Google’s AI Brain Drain, SpaceX's Huge Quarter, Airtable’s 90% Collapse, US Data Fuels China AI

Key Takeaway

Audit one workflow today and match the AI model to the task rather than defaulting to a single tool. Use a low-cost or open model for routine drafting, classification, and simple automation; reserve a frontier model for high-stakes, complex, or exploratory work where better output can create disproportionate value. Track quality, turnaround time, and cost for one week, then standardize the winning mix for your team.

Episode Overview

The panel examines Google’s AI talent departures and the strategic tension between owning frontier models and supplying the compute infrastructure that every model builder needs. It also discusses SpaceX’s reported growth across AI compute and Starlink, Airtable’s steep valuation reset and sale to Bending Spoons, and the policy debate over U.S.-created training data reaching Chinese AI labs.

Full Transcript

Transcript of Google’s AI Brain Drain, SpaceX's Huge Quarter, Airtable’s 90% Collapse, US Data Fuels China AI from All-In Podcast. Auto-generated from episode audio; may contain minor errors.

Now, everyone. Welcome back to your favorite podcast, favorite podcast, All-In All-In All-In . It is summer, and today is August 6th. It is a bit of a shame that not all the attendees could gather for the podcast, but David David David Friedberg is joining us. The Emperor of Science, David Friedberg, is back . You're doing well, right? nice to meet you . It's nice to meet you too. And everyone's favorite, Brad Gerst, is also with us . If you are interested in the stock market, he is like your Bruce Wayne.

Wayne. Wayne. He knows the secret to guaranteeing your success . all . all . all right. right. right. He closes deals at discounted prices and also creates special accounts like Trump accounts . great. . great. . great. Brad, welcome back to the program . It's really good. Thank you for showing us your rhymes again. Shall I give a brief introduction as well? Chamasu is currently on a business trip . You will be able to hear a field report from Cha Mas . And I called Daniel .

Sachs is planning to attend somehow, but you all know what kind of person he is. He is always late. As you know, I get calls from very important people. But eventually, it will come in . Ah, wait, I see it in the text here. ah. ah. ah. Ah, there it is. You've come. It's finally here. finally here. How about my cool summer vest? Oh, that's really cool. It fits perfectly. fits perfectly. fits perfectly. It looks warm. I really don't know how Suga can do something like this.

Now, everyone, I will give my report. As you all know, Chamasu is on a business trip. Oh, here it is . Hmm, this is a photo . Uh, Sachs went to check on the data center progress progress progress . It looks like they are building a data center in Colorado or Nevada . . . It's in Dune. Ah, it's in Dune . . . Yes, Dune 4. Ah, yes. Here, he is admiring himself. admiring himself. admiring himself. Uh, look. There is a stream here . You know how your wife starts teasing you when a meme reaches its peak ?

That's exactly it. Here it is. I think it was during a Christmas party. Ah, Brad, you Brad, you Brad, you appeared on CNBC with Andrew Ross Sorkin. okay. I'm not sure if it was only posted on my Twitter feed, but it definitely spread to everyone. This is a total viral phenomenon . It . It . It spread everywhere. Now, there is a lot to talk about. Let's stop the small talk . Google made two important personnel changes to its AI division on Wednesday . Demis .

Demis . Demis Hassabis has moved to the position of Chairman of DeepMind and Chief Scientist at Google . Some are speculating whether Hasabis has resigned or been promoted . Let's take a closer look at this part. However, Google However, Google However, Google described this as a promotion, stating that he had moved up to a higher position . Axios . Axios explained the personnel changes as follows. Google's Gemini 3.5 Pro has been delayed for months, and a company official company official told Axios that low employee morale was one of the reasons.

That 's interesting. 's interesting. 's interesting. Several top Several top Several top researchers, including the co-representative of Gemini, have moved to competing AI research labs . . . Three AI superstars, including Jeff Dean, left Google and founded a company called Discovery Loop. Dean is a legendary figure to Freeberg , and I know that you also worked with him at Google . He is one of the world's leading AI engineers , , , was the 30th employee to join in 1999, and as far as I know, has worked at Google for 27 years.

Discovery Loop plans to focus on in-depth scientific innovation in the field of AI . . Google's stock price fell 4% on the news of Dean's departure. It could be seen as a loss of $200 billion in market capitalization. Freeburgh, Google is also your alma mater; what do you think about this situation? Do you consider this creative destruction? Or perhaps they wanted new talent because the existing employees did not meet expectations ? Or did Google's original talent succumb to the temptation to start a startup in an era overflowing with infinite capital and opportunities in the AI ​​field Google's original talent succumb to the temptation to start a startup in an era overflowing with infinite capital and opportunities in the AI ​​field ?

It ? It ? It could be a third category . The board of directors and management are discussing how to allocate capital most efficiently . Google announced that it will invest $200 billion this year in building AI infrastructure and data centers . Thanks to capital expenditures and accelerated depreciation, investing in AI computing in the U.S. is currently very advantageous from a tax perspective . In addition, because the demand for computing is very high, the Return on Invested Capital (ROIC) model is very clear. Therefore, if you make such an investment, there is significant demand for the computing infrastructure, and you are highly proficient in operating the computing infrastructure, the capital can generate massive returns with very high confidence within the forecast period .

. Building a state-of-the-art laboratory-based model requires tens of billions of dollars in capital, and the key question is whether that model can generate revenue . In a world where open source technology advances, open-weight models are rapidly catching up, and all cutting-edge research labs are overtaking each other , is it really reasonable to invest tens of billions of dollars in model development ? In ? In ? In my view, my view, my view, the scientists leaving now are the ones who were at the core of developing these cutting-edge models models models .

They . They . They were certainly pioneers. If you look at an early interview with Jeff Dean a few years ago , you can see that Google had already internally developed a model similar to Chat GPT a year before releasing OpenAI's Chat GPT . Google did not disclose it out of concern that it would cannibalize the search market. Then, Sergey stepped forward, bringing new vitality to this field . However, as time passed and everyone competed in the model race , it became clear that it is very difficult to achieve a return on investment for the capital invested in model development, even if one , it became clear that it is very difficult to achieve a return on investment for the capital invested in model development, even if one , it became clear that it is very difficult to achieve a return on investment for the capital invested in model development, even if one invests in computing infrastructure and is not bound by the model .

Google probably possesses the world's most powerful enterprise computing infrastructure. Therefore, we have secured the largest number of corporate clients and consumers, and in both cases, we can operate a massive business even without having the best models . Google . Google can collaborate with Anthropik, OpenAI, and SpaceX without being tied to a specific model . Google . Google holds significant stakes in SpaceX and Anthropik and can utilize and host all open weight models. Now, if you are a brilliant computer scientist like Demi or Jeff Dean working at Google, and the company invests capital in infrastructure, infrastructure, data centers, and supporting a broad model ecosystem rather than your models or the field you are most interested in, you would you would you would think this: "If I could go find people like you, Brad Gerst, to prove I'm the best in the world and attract billions of dollars in investment with just a few PowerPoint presentations , that might be a better path ." I ." I think that very moment is a crucial point.

In my opinion, capital invested in data center infrastructure (CapEx) has a high alpha return and a low beta return. On return. On return. On the other hand, while model development can theoretically yield high alpha returns , it is a very risky method of investing capital due to its very high beta returns returns . So, if I were the board or the management, I would invest more capital in computing infrastructure and less capital in model development . In my opinion, that is the current situation. situation. situation.

Brad, what do you think ? ? It seems David hit the nail on the head. It's the same at Microsoft. CEO Satya Nadella said this week, citing a Morgan Stanley report, that the return on investment for Tokens as a Service ( Tokens as a Service) exceeds 30%. I mean infrastructure projects. David is correct . It's a really good business. When capital is invested, everyone ends up borrowing that capital , and it is precisely scientists who want to participate in superintelligence research , , , scientists who want to cure cancer, and scientists who want to participate in cutting-edge model research who utilize that capital.

Google is currently facing a channel conflict issue . Google . Google . Google Cloud wants to lease all computing resources to Anthropic to Anthropic to Anthropic , and , and , and companies developing advanced models internally want those computing resources to compete with Anthropic. Therefore, there is an inherent channel conflict among companies seeking to develop models . David explained this part really well. And I think this is a big challenge for Google . It seems that Google is addressing this issue by strengthening its role as an infrastructure company .

Let's take a look elsewhere for a moment . SpaceX also announced related news this week. SpaceX is also experiencing channel conflicts . I am trying to build my own model using Grok and Cursor while renting computing resources on Anthropic . There are channel conflicts at both Google and Microsoft. Of course, it seems that Microsoft is no longer actively pursuing the development of advanced models. Meta is considering entering the Infrastructure as a Service (IaaS) market market market . And there are no channel collision issues at all in Anthropic and Open AI .

They . They say they only do model projects, not infrastructure projects. So, I think this perspective clarifies that if all these people leave in the future, the companies at the forefront of model development could disappear could disappear . However, assuming the Capital Expenditure (CapEx) depreciation bill is passed and the corporate tax rate is 26% , you can receive a 26% tax credit on all amounts invested in CapEx this year . In other words, it amounts to getting your invested money back. yes. yes. Mr. Sachs, Mr.

Sachs, Mr. Sachs, would you please share your opinion on this matter as well ? According to Polymarket, which company will have the best AI model by the end of this year as of December 31 ? Of course, Anthropic is Anthropic is Anthropic is not on this list because they won last time. They are the winners. But who will lead this market in the future ? OpenAI accounts for 32%, Google 20%, and Alibaba 14% , while Moonshot, xAI, Meta, ByteDance, etc. account for about 10% . Mr.

Sachs, which business do you think is more promising ? Is developing language models or advanced models a better business? Or would it be better to focus on the token sales business as these fields are quickly commercialized? Or, since the token sale business will eventually be commercialized as well, would it be better to focus solely on the application layer? In terms of market structure, what I think is that when I heard the news of Google's acquisition, I thought, "Now only two companies remain ." As Brad said, just a year ago, five major companies were competing to lead in high-tech labs and cutting-edge models models models .

However, now only Anthropic and OpenAI remain. The high-tech intelligence market has now become an oligopoly. Of course, Elon Musk is still participating in the competition . Google will say that it is still exploring . However, as Brad said, they may have conflicting motives conflicting motives conflicting motives . This is because they can achieve significant results with only sufficient computing resources. Therefore, I believe that the cutting-edge intelligence market has transformed into an oligopoly. It is also a very powerful oligopoly system. I do n't think it is being commercialized.

Rather, I believe we are evolving into a market structure divided into two tiers: the cutting-edge intelligence market and a market that lags behind by about six to twelve months, which can be called a kind of commercialized intelligence or latecomer intelligence. A market for these tokens, or models, clearly exists. However, the reality is that costs cannot be charged on the weights themselves . You may be charged for computing resources, inference results, and consulting services that assist in building the overall system . However, . However, . However, unless it is state-of-the-art technology, you cannot charge for the model layer itself.

If you possess cutting-edge technology, you can command a premium, and Anthropic and OpenAI are exactly such companies . You can tell this just by looking at the growth rates of these two companies . . . According to recent news, Anthropic's annual recurring revenue (ARR) has surpassed $80 billion. It was $10 billion at the beginning of the year, and although the year-end ARR was initially expected to be $100 billion, most most most experts experts experts said it would be impossible. However, it appears that the goal will be achieved with a few months remaining until the end of the year.

Anthrope's forecast has also been revised upward, and the year-end ARR is expected to be $110 billion, $120 billion, or more . Open . Open . Open Eye is also accelerating its growth. Therefore, it can be seen that the current market is clearly divided into two . Two companies with state-of-the-art models capable of setting a premium are monopolizing the market . Think about the relationship between Apple and Android . Apple is competing with open source companies . Although Android has more users worldwide , , , Apple takes all the revenue.

This is because people are willing to pay for premium experiences. Similarly, people are willing to pay a premium price for cutting-edge technology . However, . However, even if it is not cutting-edge technology, the market is large, but it is provided in a commercialized form. People are simply willing to pay for computing power . From my perspective, it seems that this phenomenon is currently occurring. Jason, what do you think? Google Cloud Google Cloud Google Cloud recorded 82% revenue growth year-over-year , an unprecedented figure in the history of cloud service providers .

Elon . Elon . Elon Musk and xAI recently announced SpaceX's earnings. I will cover this in detail later, but the service sector, which I named 'Elon Web Services,' also also also showed tremendous growth. Taking Google as an example, I still think that Google will remain the number one company in the field of AI . This is because countless people are already utilizing AI within Google products . Currently, Google has five products with over 3 billion monthly users . Android Search, Gmail, Chrome, and YouTube all have over 3 billion users.

If you have used these products recently, you have likely noticed that they are becoming AI-centric . In particular, AI tools are appearing not only on YouTube , but also on Chrome and Gmail . And as Mr. Freeberg mentioned, Google has secured over 1 billion users across a total of 13 products, including Gemini . In the second quarter, Gemini's monthly active users active users active users exceeded 950, a threefold increase year-over-year . I think Google will become the overwhelming number one AI company this year in terms of consumer usage .

That does not mean, however, that cutting-edge models are not a great business . It is clearly a great business, but so far I have only used standard models rather than high-tech ones, and I am sufficiently satisfied with 95% of the work I do . I . I . I posted about this topic a while ago, and I also discussed this with Elon Musk. I think you probably mentioned this in the group chat as well . A few days ago, I posted on Twitter that the difference between the open source model I use and the Frontier model is already minimal.

I think posted on Twitter that the difference between the open source model I use and the Frontier model is already minimal. I think that is true in the work I do , but Elon replied, " Actually, there is a huge difference ." Unless you need to duplicate video games or require tremendous speed , the , the , the Frontier model is no longer necessary. I just do n't need it. People who use the Frontier Model do so because their company set it up that way , and it is because implementing open source is too difficult right now.

However, it will get easier and easier in the future . So I still support open source , and I think Gemini is a leader in this field . I am wondering if my idea that affordable general-purpose intelligence—that is, a mid-range model in the market—is sufficient fits your use case. To be honest, an wondering if my idea that affordable general-purpose intelligence—that is, a mid-range model in the market—is sufficient fits your use case. To be honest, an Android phone is enough for me. You can use it sufficiently even with an affordable Android phone .

Even so, since I use this product so much, I end up paying a high price for it. For example, if a company is in a highly competitive industry like a hedge fund , it would be impossible to run its models without utilizing the best intelligence, right ? In such a highly competitive industry, you have no choice but to invest in the best intelligence . As . As . As mentioned in the blog post by Decagon , it is important to utilize cutting-edge technology when looking for use cases .

Because . Because . Because in use cases that are not yet mature, it is impossible to know where value will be created. When looking for opportunities to utilize AI, it is much more advantageous to use the best technology. This is because the profit gained from discovering such use cases is much greater than the small premium currently being paid . Therefore, considering that we are in a highly competitive industry and are just beginning to adopt AI, and given the use cases that are not yet mature , I , I think there are many cases where people want to use all the features conveniently.

In that case, how about using a state-of-the-art model ? ? ? Think about it again. As long as employees do not engage in foolish behavior such as creating leaderboards and abusing tokens, the cost is not that high, and the benefits are enormous. Many people want the best , and are willing to pay a premium for the best . . . I would like to present a slightly different perspective . I don't think it is a matter of choosing whether to use the best model or an open source model source model .

. The trend is to appropriately mix the two. At least, that is how it seems to me . For example, most simple workflow applications use open source weights. weights. weights. However, for specialized applications requiring a high level of model expertise, expertise, open source models are used . For example, if you are doing genome modeling in the field of life sciences, you will choose a premium model. If you are doing AI video rendering work at a media company, you will use Gemini's video processing models processing models processing models .

This is because it is the most suitable model for the application . Therefore, I believe that the idea that there is a single model that fits everything is a false assumption false assumption false assumption . . . From a consumer perspective, since services like ChatGPT, Gemini, and Claude are available for around $20 to $40 per month, the From a consumer perspective, since services like ChatGPT, Gemini, and Claude are available for around $20 to $40 per month, the likelihood of using open models that do not have specific weights applied is low.

This is because, from a consumer's perspective, the service can be operated for a cost of about $40 per month . However, in the case of companies, it is expected that they will actively select the most efficient combination of models. A wide variety is available, ranging from very affordable open models that expected that they will actively select the most efficient combination of models. A wide variety is available, ranging from very affordable open models that individual individual individual employees can use to develop simple workflow applications, to more complex models required for core workflow tasks , and specialized models models models .

We cannot yet rule out the possibility that Gemini will build an amazing specialized model . . . Gemini possesses the best video and life science data and has been conducting research in the life sciences field for much longer than Antropic or OpenAI. Gemini is far ahead in this field . Demis is still running the Eye Somo Pick Lab . Therefore, I believe Gemini will prove its true value in the field of verticalized models, that is, models specialized in specific fields such as video, . Therefore, I believe Gemini will prove its true value in the field of verticalized models, that is, models specialized in specific fields such as video, .

Therefore, I believe Gemini will prove its true value in the field of verticalized models, that is, models specialized in specific fields such as video, life sciences, and protein folding . Of course, all companies will use a mix of various models . However, if there is a cloud service provider that offers various models like Google GCP , I would choose GCP rather than using only Antropic . . . These changes These changes These changes led to downward pressure on prices. We saw OpenAI and Claude drastically lower the token price .

In other words, they are reacting. They aren't sitting still. And the integration between these models is being established in many places within the company. In that case, what are your thoughts on the downward pressure on the token price ? Or is it good for both consumers and businesses ? ? ? Look, we are facing tremendous competition . That's exactly it . The United States is winning . This is exactly what we want. we want. we want. We have a very competitive market . China's . China's .

China's open source, the United States' open source, and frontier national laboratories are active in their respective fields . . . There is downward pressure on prices. David David David mentioned an oligopoly, but I think it is difficult to call a market that is only a few years old an oligopoly given the existence of mentioned an oligopoly, but I think it is difficult to call a market that is only a few years old an oligopoly given the existence of mentioned an oligopoly, but I think it is difficult to call a market that is only a few years old an oligopoly given the existence of giant corporations like Amazon, Microsoft, and Google .

. However, I agree with what he said . I believe they have emerged as pure enterprises . . . Looking at their earnings, as you know , it seems they are increasing their wallet share . However, I would like to mention two of the opinions that came out this week . One is Elon Elon Elon Musk's answer to Mr. Jason. Right? For the past two weeks, everyone has been saying that China has caught up, open source tokens have caught up in terms of intelligence, and have become much cheaper.

saying that China has caught up, open source tokens have caught up in terms of intelligence, and have become much cheaper. However, Elon said, " Not yet. We are entering the singularity, and cutting-edge models are models are models are far ahead of what people think." He said. I said. I said. I think that is correct. Your use case is very similar, but I don't think it is the most sophisticated use case that people are trying to learn and experience and experience . And Jensen said this week that the closed-type model is actually cheaper.

As you know, if you do not need to build it yourself, you can save on training costs and the expertise required for fine-tuning, maintenance, and setting up safety measures. Therefore, he argues that not only is the Frontier model fundamentally more advanced, but the cost difference between the two platforms is not as significant as people think , and he believes this is the reason they continue to lead in terms of revenue . However, I think it is healthy competition. Jay Carl, you are right. In most use cases, I believe that while token consumption by open source developers is increasing, increasing, increasing, Frontier Lab's economic share is increasing increasing increasing .

This is what we want to see. Yes, it is just like a showdown between Android and iPhone. One platform generates revenue, while the other platform secures the majority of users, at least globally . Ah, or to take a closer look, they released their first earnings report as a listed company . The stock price fell 13%, likely due to concerns over surging AI investment spending. It has fallen 30% since its listing in June, but it is currently trading and appears to be stable at a market capitalization of $1.4 trillion.

At the time of the IPO, it naturally exceeded $2 trillion. The second quarter results were truly phenomenal . Sales were $7.8 billion, up 92% from the same period last year . This figure represents a 67% increase compared to the previous quarter . AI-related . AI-related . AI-related revenue, particularly that of Elon Web Services, recorded $2.6 billion, an increase of more than three times compared to the previous quarter. This This This is separate from the Cursor acquisition. The cursor argument is not yet complete, but it will be one of the greatest arguments in history.

Elon Web Services leases Services leases Services leases computing resources from the Colossus server collection to Anthropic and Google. However, total investment spending in the second quarter was $18.4 billion, a sixfold increase compared to the same period last year . Naturally, if you calculate it, annual revenue of about $75 billion could be generated . Brad, I would like to hear your thoughts on the SpaceX IPO . I understand that you have been keeping a close eye on this situation and have offered many opinions . . .

First of all, it is surprising that this company has generated a massive value of $1.4 trillion . It is noteworthy that it fell 40 to 50 percent from its IPO to its lowest point . Do you remember the chart we showed you a few weeks ago ? Within six months of an IPO, almost all tech stocks fell 50% from their peak to their lowest point. The same phenomenon is occurring at SpaceX as well . I think the performance this quarter was very solid very solid very solid .

His performance forecast was also quite excellent . It . It announced that it would achieve $100 billion in annual recurring revenue (ARR) by the end of the year and brought forward its announced that it would achieve $100 billion in annual recurring revenue (ARR) by the end of the year and brought forward its target for a $1 trillion ARR from 2031 to 2030 . For reference, Morgan Stanley's projected revenue for 2030 is $325 billion, which is also a massive figure. Considering that last year's revenue was only $18 billion , it is clear that whether it is Morgan Stanley's figures or Elon Musk's, the market is not yet properly reflecting this company's potential .

. . The estimate of $2 trillion was looking a few years ahead, but I think the current value now reflects that level . The market is raising several questions . First . First is the computing resource rental business. Elon Musk leased a massive amount of computing resources to Anthro Peak . This is a question we have been discussing continuously . The question is whether to utilize these computing resources internally to internally to internally to build innovative models or to lease them out . . . And if we lease it out , would we be able to find people with the capital to purchase those computing resources ?

He mentioned a massive scale of 10 to 20 gigabytes , but people are questioning how that funding can be raised . And keep in mind that GPU rental businesses generally businesses generally businesses generally tend to trade at very low multiples . . Look at CoreWeave, etc. I think Grok is a hidden gem in the Frontier Model Project. He said in a conference call that Grok tokens tripled during the month of July . . . This is a figure excluding the cursor. Cursor was already expected to grow from $3 billion to $10 billion by the end of the year .

If . If Cursor and Grok are combined, it could reach $10 billion to $20 billion by the end of the year . This will become an incredibly valuable asset, enough to be traded at a much higher multiple than the data center business. And of course, become an incredibly valuable asset, enough to be traded at a much higher multiple than the data center business. And of course, we haven't even talked about Starlink yet . I . I think he will achieve tremendous success in the mobile market .

This is a standard adjustment process. process. process. Various funds across Silicon Valley are distributing stocks. The stock price fell slightly. It is not surprising to me at all. Now, everything depends on execution . The two most important things are, first, how Grok and Cursor's year-end sales will turn out, and second, second, how how successful Starlink will be in replacing existing mobile carriers. It seems the distribution started today or yesterday. I also received my first distribution from the fund I invested in. I am investing in several funds that have invested in SpaceX.

It seems like everyone is investing in SpaceX these days . Naturally, downward pressure will be applied to those who have held for a long time and wish to cash out . . However, I However, I However, I intend to continue holding this stock for my grandchildren . Mr. Sacks, what do you think about this amazing performance ? Yes, I don't know how to put it. The results from a field that was not SpaceX's business area until just nine months ago are truly amazing truly amazing truly amazing .

. . Yes, I think this earnings announcement was very positive very positive very positive . It was a bit unexpected that the stock price fell after the announcement, as not only were the results better than expected along with strong earnings . It was a bit unexpected that the stock price fell after the announcement, as not only were the results better than expected along with strong earnings , but Elon Musk also spoke extensively about the company's future plans. To add to what Brad mentioned, this is about the Starship.

Elon Musk said that the Starship test flight was successful, as we all saw. It was said that the Starship was floating on the sea and the heat shield was functioning properly. Thanks to this, future It was said that the Starship was floating on the sea and the heat shield was functioning properly. Thanks to this, future Starship test flights can proceed more quickly , which will serve as a stepping stone for the development of V3 satellites, significantly expand the bandwidth of the Starlink network , and ultimately enable Direct Device Connectivity (DTC) business .

The telecommunications business as a whole appeared to be proceeding smoothly, and the company is very optimistic about this sector . And . And . And there is also an AI data center business. Speaking of data centers , computing , computing , computing capacity is expected to increase from 1.4 gigawatts to about 2 gigawatts by the end of the year . Elon . Elon . Elon Musk mentioned that the spot price of computing resources is between $30 and $50 per watt . As you can see if you calculate it , since 1 gigawatt is 1 billion watts, $ 30 to $50 per watt means $30 billion to $50 billion per gigawatt .

The current price seems to be close to the upper end of that range . Therefore, when Elon Musk said, "We will achieve an annual recurring revenue (ARR) of $100 billion by the end of the year, " he meant that to achieve that goal, 2 gigawatts of computing resources would have to be operated at $50 per watt . . . Starlink, the Launch Project, the Grockers Project, etc. were not included here. That seems to be why they are putting forward such an optimistic outlook. Jack, as you said, there are many ways to succeed.

There are many ways to succeed with this stock . I think Starlink is a company with truly tremendous cash-generating capabilities. You can see this by looking at the financial statements, especially the reports on the space, connectivity, and AI sectors sectors sectors . Starlink . Starlink recorded an adjusted EBITDA of $2.6 billion in the connectivity sector. This can be viewed as operating cash flow . The space sector recorded a deficit of $200 million, so it could be considered the break-even point break-even point break-even point . The AI ​​sector posted a surplus of $1.1 billion.

However, as Brad pointed out, in the AI ​​sector, it is unclear whether the rent for computing resources is a temporary premium due to the current shortage of computing resources, or if people who need computing resources are paying a premium to Elon Musk . Therefore, . Therefore, . Therefore, questions remain regarding future prospects. However, Starlink's connectivity segment generated $4.3 billion in revenue in the quarter , and its adjusted EBITDA reached $2.6 billion. The number of subscribers doubled from the previous year to 12 million . The monthly ARPU is $66, which is the amount people pay every month .

And he grew by 20% quarterly . Expanding on this, the number of annual subscribers is expected to approach 24 million. If we apply this multiple and assume that the business targeting corporate clients, such as airlines, scales up just like the consumer business , Starlink alone could generate about $40 million in revenue, and a significant portion of that could be converted into free cash . This means that it can generate $30 million in free cash flow within one year . This amount of cash flow alone can support most of Elon Musk's businesses .

And if we apply a 30x multiple here, I think it is quite possible because subscription businesses have very high renewal rates and very low customer acquisition costs (CAC) . With the Starlink business alone, the market capitalization could surpass $1 trillion within two years, or rather, within 18 months . I think all the remaining funds will be invested in science projects and growth potential . So . So . So I have an optimistic outlook . It is truly amazing how well the Starlink business is doing .

You can also see that effect at AT&T, Verizon, Hughes Net, and Via Sat. These companies have completely failed. I previously installed a Hughes Net satellite antenna at my ranch in Sonoma County to use the internet . I had to use that. I think it was about 200 dollars a month . Since it was a high- orbit satellite, the signal was slow and the service was terrible. However, that However, that However, that market completely collapsed because of Starlink . If he starts a mobile phone business, the growth in subscribers will accelerate even further.

Even now, 2 million subscribers are joining every quarter, and that number could increase to 4 to 5 million. There is a possibility that the number of consumer mobile subscribers will actually surge . There are as many as 400 million mobile subscribers in the U.S. alone, you know. I think we can have a sufficiently optimistic outlook based on Star Link alone. The rest is " How will Elon Musk invest the excess capital coming from Starlink? What will he do with that money?" It comes down to the question of...

Well, could we entrust that money to someone who can do things like Elon Musk, such as Starship, AI computing, or TeraFab ? Good ? Good heavens, this is like a science fiction novel set in Texas County . This is a way for the U.S. to break free from its dependence on semiconductors for Taiwan and China . . If Elon Musk takes charge of this business and realizes the vision he showed here today, this place will become the largest semiconductor manufacturing facility on Earth . . David, to add to your question , how many CEOs or founders wouldn't take other risks with just one exceptional business like Starlink— that is, a business that could grow from $1 trillion to $2 trillion ?

They probably wouldn't have invested in Terra Fab . He probably wouldn't have tried to build a data center himself , nor would he have attempted to create his own model . . . This investment currently underway carries a high risk, but it is a very important investment. Showing an unyielding passion for innovation like Elon Musk is a truly heroic and important thing . I wish more CEOs and listed companies would take on this level of risk . I just mentioned that some CEOs tend to choose safe investments.

However, Elon refuses safe investments . He is reinvesting all the profits from this special business into sectors that are very important to the United States . . Bryce, that is a really good point . Other CEOs and executives are also joining this trend . They started to realize that investing in the future is more important than buying back shares or paying dividends paying dividends . DoorDash received harsh criticism due to excessive capital expenditure . Google was also hit by capital expenditure issues. This kind of thing keeps repeating.

repeating. Speaking of the headwinds SpaceX will face , while I am likely to be wrong, a valid question worth discussing is, "Will the demand for tokens and computing continue to increase?" Or, "Will demand eventually decrease as on- premises, desktop, and open source models evolve to become increasingly smaller and more efficient efficient ?" no see . I don't think that will be the case. I am not sure if there is a limit to the demand for on-demand intelligence . Secondly, . Secondly, Starlink is for people living in rural areas .

You cannot install Starlink if the building is connected to Verizon or Spectrum fiber optics . So, what will likely be explained within the next one to two years is that Starlink will be installed in all Tesla vehicles . . Once the merger is complete, all mobile phones will be able to connect to the Wi-Fi network of all Tesla vehicles . For example, all robo-taxis on the road will also be able to connect directly through the next-generation Star Link . In other words , once visibility is clear, you will be able to connect to all Tesla vehicles on the road via your mobile phone .

Because all future Tesla vehicles are scheduled to have Starlink built-in. This is something I am really looking forward to. And finally, there was a lot of speculation regarding the company's value . Brad, just as you mentioned. I think I mentioned securing liquidity when people asked you . Unlisted . Unlisted companies or venture capital are means of exercising voting rights. And once it is listed, it will serve as an evaluation indicator . . . And sometimes there are moments when concerns about corporate valuation grow . And that concern peaked last quarter .

When Tesla first went public, its price-to-sales ratio was 160 times. 160 times, right? If you compare a market capitalization of 2 to 3 trillion dollars with much smaller revenue, you know. However, know. However, know. However, as sales increased, the price-to-sales ratio has now dropped to 45 times . So, a kind of balance is being achieved . Yes, Brad, the balance between the private and public markets , as well as revenue growth, are all having an impact . . . Honestly, I think all of this is very wholesome.

wholesome. wholesome. SpaceX's IPO was really amazing . The current adjustment in corporate value is fully predictable. And there is a company with a market capitalization of $1.4 trillion right now , and if you look ahead about three or four years—whether it is Morgan Stanley's or Elon Musk 's estimates or whatever— considering it is a very reasonable valuation, I think you can triple the money invested in this business. I have always made such predictions . I acknowledge that Elon Musk is a top innovator and a capital allocation expert allocation expert allocation expert , but it is true that entry price is important .

Buying . Buying . Buying over $2 trillion worth of this company's stock on the first day of its IPO is only possible if you are convinced that such a thing will thing will thing will happen happen happen . I appeared on CNBC on the day of the IPO and said that I wanted to own shares of this company , but that I was n't sure if now was the right time to buy them . Right? So, the entry price is important. This is a basic principle.

However, let me give you another example . We've been talking about Entropy's IPO , and many people expect it to take place at the end of this year . I have heard many people talking about around 1.5 trillion or 2 trillion dollars . David . David said earlier that sales would exceed $100 billion by the end of the year . It's about 10 to 15 times the sales . It is not that large for a company that has grown tenfold and is rumored to be profitable in the second quarter .

. Looking at the market, regarding the correction in July—that is, the Leopold bottom formed in July— many semiconductor-related stocks fell at that time. I mean the Leopold floor. Wait, Leopold is doing really well. The well. The well. The stock price has risen by 80% this year, and it recently made a large-scale private investment as well. I think it is a truly amazing feat to have grown the company so significantly in such a short period of time . However, the market was in a panic, so he had to overcome the crisis .

I see all of this as really positive really positive really positive . So, buoyed by the SpaceX IPO and ahead of other companies' IPOs in the second half of this year, I think we are in a really good situation. This is especially true if this sales growth trend continues. Elon Elon Elon Musk is truly excellent at building physical factories, that is, physical facilities physical facilities physical facilities . . In this world where everyone is competing to secure data centers and semiconductor manufacturing facilities, physical hardware physical hardware is essential for providing software services, and this is Elon Musk's core competitive advantage .

Look at the gigafactories built around the world . . . This is precisely his core competency . If you compare Brad, Elon Musk, Dario Parsons, Sam Rogers, or even Alphabet, which has been in business for 27 years , you can see the key advantage Elon Musk has if the world comes down to this kind of competition . Just like he said in a conference call, " Building a data center is nothing compared to launching a rocket." Building a data center isn't as difficult as rocket science .

We are building data centers by leveraging SpaceX's hardware expertise hardware expertise . That is why we were able to build more, larger data centers faster than our competitors . I have a few questions . Do you understand why Starship is so important to Starlink ? Yes, I will explain it briefly. SpaceX has developed a new V3 satellite with a bandwidth 10 times greater than the existing V2 satellite. The Starlink network is currently operating with V2 satellites , but by launching about 27 satellites at once aboard a Falcon 9 rocket, the total total total network capacity will be increased by about 2.6 terabits/second.

However, the Starship can launch 60 V3 satellites at once . . . Then the total network capacity per launch increases by 60 terabit/second . In other words, the capacity increases by more than 20 times per launch . This is the power of the Starship . So, when the starship starts operating properly... And And And in the last test, not only did the heat shield work properly, work properly, work properly, but as far as I know, they also test-launched—or rather, deployed—20 V3 satellites . We also successfully connected to those satellites and confirmed that they are working properly.

It even has a camera built in . The . The . The reason we were able to see the Starship was probably because of the idea, "Life is only once, so let's put an HD camera on it ." ." Yes, that is correct. Those satellites were basically for testing purposes, and eventually they burned up. So, I think the next important milestone will be to load 60 V3 satellites onto the Starship, place them into precise orbits, and connect with the satellites to increase network bandwidth . If that happens, the bandwidth of the Starlink network could increase tenfold, or perhaps even up to a hundredfold .

. And when that time comes, we will be able to do interesting things like direct cellular connection . Gwyn . Gwyn . Gwyn Shotwell mentioned the ground base station and gave an interesting hint about what it could potentially do. And Sachs could also acquire a company like T-Mobile . It . It is within a sufficiently manageable range. And I think Elon Musk mentioned that the Starlink network would ultimately be able to handle about half of the internet traffic , so that means this service could service could grow much larger than the 12 million subscribers Mr.

Freeberg mentioned . By the way, Brad, I’d like to talk briefly about data centers . I have a few questions . Elon Musk said that he would achieve 2 gigawatts (GW) by the end of this year, and 5 to 10 GW next year, closer to 10 GW than 5 GW . . . So, shall we say 8GW? It is just an arbitrarily chosen number, but it is a sufficiently possible range. Now, let's assume that 6 gigawatts are added . . . Then it increases from 2 gigawatts to 8 gigawatts.

Right? I have two questions. One question is how we can be certain that spot prices will maintain their current levels. Can we maintain $50 per watt ? How can I know ? How can we track it ? To ? To ? To what extent is the associated risk? During the conference call, Elon Musk seemed to think that spot prices would rise because the market is currently constrained by a memory supply shortage . They said that memory production would increase by about 20% next year, but they also mentioned that demand would increase by more than 200% .

Ultimately, the market becomes constrained by the bottleneck at that point. The current bottleneck current bottleneck current bottleneck is memory. In that case, what do what do what do you think will happen to the spot price? To what extent is the associated risk? And another thing I am curious about is that if you expand from 2 gigawatts to 8 gigawatts , the net increase is 6 gigawatts. You gigawatts. You know that building a gigawatt-class data center typically requires a capital expenditure of $50 billion. So, that means $300 billion in capital expenditure is required to add 6 gigawatts of computing capacity next year.

Of course, that is assuming the construction is completed . Of course, there will be an option regarding whether or not to build it . Then, . Then, . Then, how will this cost be funded? What is the best way to minimize share dilution? You said the payback period is within one year, but it will probably vary depending on the market price. In other words, you only need to provide funding for one year . Do you think Nvidia will fund them fund them fund them ? ?

? Or I wonder how this situation will unfold. will unfold. will unfold. David, that is a good question. First of all, the construction cost is at least $50 per gigawatt. Yes, so let's assume the investment is $300 . . To raise this capital, they must borrow money from the market or or or issue shares through equity dilution , neither of which is the way they want . Alternatively, you can obtain a warranty from NVIDIA , and NVIDIA has stated that it intends to provide more warranties in the future .

However, the problem is that Nvidia shareholders do not want unlimited warranties . This is because there is concern that the spot price of Sprite may eventually fall . If the price falls, the payback period payback period payback period changes. Looking at current spot prices, it seems like the payback period will be about one year, but no one thinks so thinks so . Just a few years ago, it was uncertain whether the investment could even be recouped, and until a few months ago, many people thought it could be recouped within four years .

. So basically, it means investing $50 and earning $10 to $15 a year . The investment is recovered within 4 to 5 years, and if you are lucky, the return on investment will well exceed 20% in the 6th year. There is currently a severe shortage of computing resources, and because cutting-edge research labs are convinced that tremendous innovation is on the horizon, they are willing to pay three, four , or even five times the market price . The . The contract between SpaceX and Antropic is exactly such an example.

Antropic will invest much more if the opportunity arises, even now . The same goes for open eye . . . Brad, you said they are willing to overpay up to five times , which means $50 per watt that David mentioned. They will be willing to pay $30 to $50 per watt to secure large-scale computing resources that can secure a competitive advantage in the market They will be willing to pay $30 to $50 per watt to secure large-scale computing resources that can secure a competitive advantage in the market .

. And remember, not many companies can afford to purchase computing resources of this scale . It was not Chinese open source companies that purchased the surplus computing resources generated as SpaceX built a 10-gigawatt power plant in Ohio . OpenAI and Anthropic Games purchased it. Therefore, most purchase agreements are coming from Anthropic Games, OpenAI, and Nvidia. When you hear that hyperscaler companies are building all these computing resources, they are producing these resources to sell to the companies just mentioned . . David, if he builds a 6- gigawatt facility next year , perhaps only Elon Musk would be able to actually build something of that scale within that timeframe .

As Jensen told me on the podcast, no one comes close to him . Even Microsoft and Google are far from being able to build that scale within that period . Hmm, I think he's going to have trouble getting the parts. Assembly itself might be possible, but will I be able to secure the memory, chips, long power cases, etc., in time ? It ? It ? It seems like there is demand, though. However, to explain the situation more clearly, when Anthropic and OpenIal were combined this year, the initial computing capacity was about 5 gigawatts .

He . He says he will gradually add more capacity than when the two companies merged, but considering that Anthropic is growing tenfold every year and OpenIre is OpenIre is OpenIre is currently growing about fourfold, that isn't that big of an increase. There is clearly demand in the world , and I think that will be the case for the next 12 to 24 months or so, but it seems there is anxiety lurking in the market . The reason the stock price fell in July is that Kimi created anxiety among people, making them think, "Oh my god, Nvidia is going to cut Frontier Labs' revenue ." If ." If Frontier Labs' revenue decreases , who , who , who will cover all these computing resources?

This is precisely why the stock prices of companies like CoreWeave plummeted by 40%, and why both semiconductor- related and AI-related stocks declined declined declined . . . Furthermore, it is predicted that developing 10 gigawatts (GW) of computing resources in the future will cost $500 billion, but it is questionable where this funding will come from. Should we raise funds through a stock issuance? Will Nvidia raise the funds on its own ? Or should we establish a Special Purpose Vehicle (SPV) like other companies ? As you know, the fact that we are having this conversation, that everyone is aware of this fact, and that the market is fully aware of it means that if the plan is not realized or the pace slows down, people can respond to the changes in real time.

However, I expect that it will be difficult for this trend to continue for more than two years . But as our good friend Bill Gurley always reminds us, he says, "I can't believe we are accepting this level of seller financing so casually ." They often say that. Right? He would call this 'circular profit '. However, '. However, '. However, the market has already become accustomed to this situation . And . And as we saw in July, when concerns about demand arise, the entire industry declines. yes declines.

yes declines. yes . . . Everything falls together. This is because that is precisely due to the leverage injected into the system . In effect, it supports people's ability to build ahead of profit . Therefore, if there are signs of a decrease in demand, it drops by a much larger margin. Well, as you know , I , I don't see any such signs right now for the next 12 to 18 months, but there is unpredictable uncertainty unpredictable uncertainty unpredictable uncertainty , and this certainly makes people anxious .

Credit spreads continue to widen, and widen, and spreads in these transactions remain wide . Therefore, there is fear of such transactions in the market . Now, everyone. In September, you are...the 5th annual event. All In Summit is back! The fifth annual event will be held. Yes, that is correct. The best experts are joining David Friedberg for this summit . NVIDIA . NVIDIA Founder and CEO Jensen Huang. If you are interested in the future of AI, do not miss this . . . Microsoft CEO Microsoft CEO Microsoft CEO Satya Nadella, a fan of our podcast, is appearing for the second time .

NASA's Jared Isaacman, Brad Gerstner, and Bill Gurley (BG2) also return. Many great people are participating, including SpaceX's Gwen Shotwell, Jason Paul, and Nick Sully . Martin Schrelli might also attend. I'm really looking forward to it! Apply right now at all-in summit.com . You can access either allin.com or all-in summit.com summit.com summit.com . We plan to take over Universal Studios once again and create our own special space. Dave Friedberg, you did a great job running the summit . Casino . Casino . Casino Night Toyo. yes that's right.

It's going to be a really grand casino night . It will be the largest ever, and a concert is also scheduled to be held. The lineup has n't been announced yet, but it will be fantastic. fantastic. Speaking of the summit, people from over 60 countries attended. Meeting so many people like this is a truly amazing experience. It is about meeting entrepreneurs, investors, and people who are genuinely interested in the topics we are discussing . We . We try to have the most important conversations in the world , but what matters most is the amazing community experience .

That is exactly what makes people come back. So, we are increasing our investment every year to make the summit an even better experience better experience better experience . It is different from other events that simply showcase cool content on stage. How can I have a truly special experience for a few days? I'm really looking forward to it . We are focusing on three things . First, you will learn something. Right? You will learn something from the great people on stage . You will meet new people, network , and have a wonderful experience .

It is an event that has all three elements . Brad, you're looking forward to it, aren't you ? Are you happy to be back? What is the date? Could you please say that again? Look at the calendar. You're sleeping . It . It . It will be held in LA from September 13 to 15. There couldn't be a better date for this summit . Because the midterm elections will be held within 60 days, and Antropic's IPO will be held within 30 days . It will be a really hot time .

It seems that the SaaS crisis— not the Socks crisis, but this very SaaS crisis—is now slowly coming to an end . The . The . The indigestion must be subsiding a bit too. The Air Table Air Table Air Table was acquired for a price lower than the fundraising amount, you know. It is a profitable SaaS company, you know. Air Table, a company that possessed excellent products , recorded $480 million in sales and an annual growth rate of 20%, and would have been worthy of respect if it were a publicly traded company, was sold for $1.28 billion .

This is a price that is only about 10% of the highest enterprise value of $11.7 billion in 2021 . . . Including cash holdings, the sale price would have been $2.25 billion . Air Table was acquired by a company based in Milan, Italy called Bending Spoons. Bending Spoons is a company that acquires struggling but interesting businesses, having acquired AOL's existing businesses, Evernote, Eventbrite, Evernote, Eventbrite, Evernote, Eventbrite, Vimeo, Meetup.com, and others . Bending Spoons went public last month, and its stock price rose 15% on news of the acquisition of Air Table .

Air Table had been successful in many aspects of management and possessed a massive amount of cash, but there were rumors that the founders appeared somewhat unstable. You might be tired. Some of the investors who invested at high prices may be exhausted. What lesson can we learn from Bending Spoons' recent trade ? Are Bending Spoons the last buyers now? Well, I think Bending Spoons is building a great business on its own . This is because this acquisition will be a highly profitable one . One more thing to add: prior to this acquisition, Airtable spun off its AI agent business unit, HyperAgent, HyperAgent, HyperAgent, into a separate, independent company .

In my opinion, the company's founders and talent believed, "We do not need to continue operating the existing product that has become the target of private equity investment. (I will explain what this means not need to continue operating the existing product that has become the target of private equity investment. (I will explain what this means shortly shortly shortly .) We .) We want to focus on a new business, namely the AI ​​business. That is the field where we can create significant value in the future ." ." It seems they decided that.

So, basically, the talent focuses on venture investment, while the private equity investment division is sold to Bending Spoons. Why do I think it would be a good acquisition target for Bending Spoons ? There is an interesting piece of data I saw in a related commentary: only 30% of Airtable's sales team achieved their sales targets . The sales achievement rate was merely 30% . This figure tells us a lot about this business. Reading between the lines, I see that the company was implementing a successful PLG (Product- Driven Growth) strategy .

In other words, they were recording an annual growth rate of about 20% through organic, product-driven growth . However, the board was not satisfied with that . As you know, the board consists of investors who invested in a company valued at up to $11 billion . They . They expected results similar to those of a venture investment. So, what happened? The board pressured the founders to do things that, frankly speaking, were unfamiliar to them . It was exactly what they said: "Hey, to grow faster, we need to add a traditional sales-driven approach here." Say that.

Will that be effective? No, you will probably get some growth, but the achievement rate will only be 30% . It amounts to hundreds of salespeople working in vain , and the growth rate is slowing down. So, what are the opportunities for the acquiring company? Bending Spoons can Spoons can eliminate 85 to 90 percent of its cost structure, just as Elon Musk did on Twitter, and move away from a sales-driven strategy to return to its original approach of product-centric growth . Then, it will become a very profitable company while maintaining a 20% growth rate for most of it .

I think we can probably achieve a return of about 80% . Some . Some people say that net profit will recoup the acquisition costs within a few years , while others say the EBITDA margin will only be 30%, but I think 80 to 90% is possible. I do not think it is necessary to maintain most or most parts of the business . . . I mean the cost structure related to this project. Air Table is a company with a fan base . I think the fans will probably continue to support the air table .

And it will be able to grow by about 10 to 20 percent while generating an annual EBITDA of around $300 million to $400 million . Bending Spoons' strategy is precisely this part. However, However, However, investors, especially investors, especially investors, especially companies like Sachs, are satisfied with recouping their investment and moving on to the next investment . It’s . It’s not like a situation at a blackjack table where you have to raise the price tenfold to catch up, but rather a situation where you have to raise it not like a situation at a blackjack table where you have to raise the price tenfold to catch up, but rather a situation where you have to raise it tenfold again to satisfy the investors .

That wo n't happen. Ultimately, the key question is whether Bending Spoons can acquire this currently unprofitable business, generate $400 million in annual EBITDA, and cover the costs . . . In just 3 years. That's surprising. That's surprising. That's surprising. Why couldn't the company do that on its own ? In my opinion, the structural problem is that both the venture capitalists (VCs) on the board and the founders find it very difficult to switch to private equity mode . This is because they have to tear down what they have built up so far.

I have strong loyalty to the team, so I don't want to think about how to reduce the cost structure by 80% or 90% strong loyalty to the team, so I don't want to think about how to reduce the cost structure by 80% or 90% . Because that is n't their way. What the founders want, and the result desired by the board members, is the outcome achieved through attracting venture investment. I think they could have easily done that . You could have done it like a bending spoon.

But they But they But they don't fit into that structure, Sachs. It's just as you said. Not only do they not fit into such a structure, but the capital structure itself is flawed . It is because they have massive liquidation priority priority priority . All these investors must get their investment money back . It reached a corporate value of $11 billion , and as you know, that value continued to rise . . . Brad, the incentive system has collapsed. And your company, Altimeter, also made quite bold investments during this period, didn't it?

I am not sure if Air Table was one of them, but you have invested in SaaS companies several times, and some of them were valued at high valuations . How do you look back on that period ? ? ? Are there any lessons to be learned as you move forward? Sales multiples can drop in an instant . Right? It is effective when a company grows by more than 50% , but you must keep in mind that this is merely an empirical indicator indicator indicator . It is just a very rough estimate used almost entirely in Silicon Valley .

Just like people say, "Wow , this company sold for double its revenue ." But if you look at it internally, it was probably sold for about 30 times its free cash flow. I don't think raising EBITDA to $400 million is an easy task easy task easy task . If it were, the board would have already done so . I . I . I invest in companies like this, and when a company loses its growth momentum, employee morale hits rock bottom, and customer churn begins to skyrocket.

A vicious cycle starts . So, the reason I go to work every day is ultimately, ' How am I going to survive?' That is what is happening. I am not entirely sure what the core product is or what the turnover rate is like for it, but I think that as the sure what the core product is or what the turnover rate is like for it, but I think that as the pace of its development slows, the core product has actually started to wither . The turnover rate in the product development department is skyrocketing skyrocketing skyrocketing .

Now, people say, "It is almost impossible for software companies these days to retain good salespeople or product developers." This is because they say, "It is almost impossible for software companies these days to retain good salespeople or product developers." This is because they all want to work in the field of artificial intelligence . "I . "I agree, but it is not necessary for this product . I do agree that the market is efficient. I think Vending Spoons has an advantage that the board and founders lack: they already have the infrastructure in place .

Vending Spoons currently has a core team managing dozens of assets . . . Therefore, they can apply AI immediately . In some ways, AI makes their work easier. In the past, the reason we couldn't completely eliminate talent and infrastructure was that we needed someone who knew the organization 's experience and knowledge—that is, the codebase . But . But now, AI can learn the codebase instantly. That is an interesting perspective. So, So, So, using AI using AI using AI makes maintenance much easier because we no longer need past knowledge .

AI . AI can reconstruct past knowledge . . Mr. Freeberg, may I speak for a moment ?" Hmm, when comparing the lessons of the Zerp era, the rapid growth of the SaaS market, and today's AI market , , , what similarities or lessons can be found between the two periods ? ? Between the Zerp era and the AI ​​era, you know. you know. you know. In the Zerp SaaS era, corporate value was very high, and amidst the enthusiastic atmosphere, there were many people who turned a blind eye to reality.

In the current AI era, computing costs are incredibly high, and many companies are recording price-to-sales ratios (PER) of up to 100 . . . What similarities exist between these two eras ? ? ? It is an easy question to answer, but it is a completely different paradigm. The focus of capital on AI capital expenditure (CAPEX) implementation and model training—that is, on capital investment—is capital investment—is not on obsessing over high revenue multiples like in the past SaaS market. It means that if you get a PER of 20, you can turn $1 into $20 .

. . Shall we talk all day? This has a completely different structure, strategy, and capital allocation process. So I don't think the two are related in any way . To . To be honest, that's an easy question . I intentionally tried to elicit a response. Look, there are two reasons why SaaS companies are overvalued in the era of zero interest rates . One . One reason is that the interest rates were artificially low. A speculative asset bubble had formed. Another reason is that people regarded these things as if they were guaranteed pensions, and even growing pensions.

They said the net retention rate was 120% , so they thought this business would grow by 20% every year forever according to the basic scenario . That is why the price was set that way. However, disruptive changes are clearly taking place in the field of AI . As Brad said , by now the customer churn rate would have been high, and SaaS is not a pension. The situation can change at any time. So, naturally, these products are now being traded at much more heavily discounted prices.

However, even taking all of this into account , I do not think it is right to generalize the entire SaaS market based on a single company called Airtable . Airtable has several features that are distinctly different from other companies like Salesforce or Workday . Air tables have always been somewhat unique products . When this company was at the height of its popularity, people popularity, people said that the Air Table was just like a new Excel or Google Sheets. Basically, it is an office product , and I thought it was a spreadsheet for Word .

People . People perceived the air table as a new spreadsheet that deals with words rather than numbers . However, the air table failed to live up to those expectations . Everyone . Everyone knows how to use spreadsheets, and everyone uses spreadsheets, right? The air table has not reached that level . Most . Most . Most people still people still don't even know what an air table is. Of course, it had an enthusiastic fan base, but it was a product with fierce market competition . It was a product that needed to be explained to people .

When should I use it? Just like a cult. However, the air table did not achieve that level of public acceptance. It wasn't clearly explained why it should be used or what purpose it serves serves serves . That is why we could n't do marketing properly . . . To be honest, products like Claude Code, Purple Rex City, and Agent are doing the role that Air Table used to play. Air Air tables have failed to create a clearly established niche market regarding when to use them .

It was like a jumble of various no-code tools mixed together . The no-code tool market is currently the sector undergoing the biggest changes in the SaaS market . What is Claude Cod really good at? Things like Claude Code and Purple Rexity are the best no-code tools, you know . . Taking tools like AirTable or Retool as examples, in the past you could use them without knowing coding, but now you had to learn how to use AirTable or Retool. They were things like alternative programming languages. However, now there is absolutely no need to learn such things.

If you use Claude, you just have to give the command for what you want. For example, if you want to create a new dashboard or a voice spreadsheet, you just need to tell Claude what you want . There is no longer a learning curve . While the entire SaaS industry is currently being affected , it seems this sector is taking the biggest hit . . . Based on the situation with AirTable, while it may not be applicable to other fields in the same way , I believe it is not advisable to replace CRM, ERP, and HR systems with other systems built using Vibe coding .

Because any system requires certainty in everything related to coding. Compliance with regulations? To be honest, honest, regarding the SaaS tool our team developed, do you use off-the-shelf SaaS tools like portfolio management? We We made this by coding it ourselves. Yes , it's the same for us. The tool our team created is truly amazing; if we had used off-the-shelf software, the software cost alone would have been $250,000, and the integration work alone would have cost $1 million and taken two to three years. However, we made it in a month .

You can now perfectly grasp the entire portfolio, competitors, founders, and all other situations . . Remember that one of the reasons Leopold suffered a huge loss was precisely because he bet on the collapse of SaaS. You did n't invest for the long term simply because of the semiconductor stock correction, did you ? I short- sold stocks of SaaS companies, especially Adobe and several other SaaS companies. All of those investments went in the wrong direction. I would like to reiterate that I believe these points should be taken into consideration .

It is too generalized to conclude that all SaaS will completely disappear disappear disappear . yes . yes . yes . . There was a really good article regarding this, and it contained the following: The reason people buy Microsoft is not because Microsoft makes the best code, but because Microsoft is the foundation that drives everything . Employee . Employee . Employee IDs are stored in Active Directory, meeting materials and financial data appear in Excel, and compliance-related conversations compliance-related conversations compliance-related conversations are recorded in Teams. Because Azure has received the highest level of FedRAMP certification and Department of Defense Impact Level 5 security approval, security approval, security approval, defense contractors defense contractors defense contractors cannot arbitrarily replace it with cheaper products .

As such, large corporations have sufficient reason not to spend tens of millions of dollars to dismantle a system that they use at a cost of millions of dollars annually. That doesn't make sense. have sufficient reason not to spend tens of millions of dollars to dismantle a system that they use at a cost of millions of dollars annually. That doesn't make sense. And I just saw that Benny Off posted on Twitter 5 minutes ago that all 15 government departments are using Salesforce , so there is no way the government would tear down Salesforce and replace it with another system.

Something live-coded . So, the point is that not all SaaS is the same in this regard . . . I bought some Figma. I think that SaaS companies with great founders who run their businesses with a long-term perspective and a passionate passionate passionate user base will leap forward with AI-first products . Figma is one of those companies, too. To wrap up, IGV has risen 20% over the past six months and 20% over the past five years . . Could you explain IGV to me? It's the high-growth software stock index, right?

Snowflake has risen 88% over the past six months. IGV is an ETF of growth software companies . As David said, there was panic regarding software companies, and a massive sell-off. To be honest , software , software , software companies performed quite well, and as he mentioned, they rose in July when many semiconductor and AI stocks fell. And some of these companies, such as Databricks, Snowflake, and ClickHouse, are achieving tremendous success . As I just mentioned, Snowflake has Snowflake has Snowflake has risen 90% over the past six months, which falls into the same category as semiconductor AI-related stocks.

Therefore, as David said, not all companies can be put into the same category , but it seems these no-code application software application software application software companies have realized that the game is over and are trying to sell the company to maximize profits . The . The . The important thing in the Airtable story is the late investors. We investors. We turned down investment in the last three funding rounds . They were worth $2 billion, $5 billion, and $11 billion, respectively. However, the late investors, who were the most likely growth investors, investors, investors, all withdrew their investments .

And the early investors made huge profits . Therefore, the failure of Airtable can be considered a fairly successful failure in Silicon Valley . . . This This This is one of the points raised right there. At the time, I used to think, "This company has a solid team and revenue base , so if we can secure the investment funds along with a sale option, it might be a good investment." The same was true for AI-related investments. However, in this case, the liquidation priority actually played an important role .

It is usually not important, but I think the investors eventually got their money back . As far as I know, it was n't something like participating preferred stock that offers a 7% interest rate or double the investment. Does anyone happen to know ? I ? I ? I took a close look at this part, and the return on investment was quite good. The net profit after cashing out may have been slightly less than the total investment, but it seems that ultimately, no one suffered a loss.

But what if there had been a time when liquidation priority of 1x had to be guaranteed, like Sachs ? During the Sachs era, a 1x liquidation priority was the standard. That means you get your investment back before others see a profit, which is only natural. natural. And the interest rate was excluded, wasn't it? It seems these transactions took place when the common stock reached its peak before making a profit . . . Standard conditions, that is, the so-called ' clean conditions,' clean conditions,' are, simply put, 1x liquidation priority.

Right? Preferred stock holders receive their investment back before common stock holders participate in the profit distribution resulting from the sale of the company. Isn't that obvious? Yes, but participating preferred stock offers a double benefit, doesn't it? Yes, but we have never stuck to that method. We aim for clean conditions . I don't mean to disadvantage the founders . However, it does n't make sense that in the shareholder structure, some people make money while others suffer losses . . . Right? Right? Right? The distribution of value among shareholders is not being done properly properly properly .

Because in the shareholder structure, the value of some people is being transferred to others. So, generally speaking, it is important to ensure that investors get their investment back and that everyone can share in the profits . . . Yes, this is the fourth story. China is training AI using U.S. data provided by U.S. companies . Forbes published a research report titled "U.S. Startups Are Making China's AI Smarter." I Smarter." I believe this is also related to many of the activities Mr. Sachs carried out during the early days of his administration.

According to the report, U.S. data labeling startups are selling valuable training data to Chinese laboratories , and , and , and as a result, Chinese laboratories are catching up with leading U.S. laboratories . Two startups, Surge AI and Mercor, each have a corporate value of over $20 billion . They are selling training datasets to companies such as OpenAI, Anthropic, and federal agencies. According to this report, these companies are also selling the same datasets to major Chinese AI companies such as Tencent, ByteDance, Alibaba, and Moonshot .

. According to this report, China's top six AI research institutes spend $500 million annually to purchase content, reinforcement learning, and knowledge pipelines written by PhD-level researchers, which Forbes calls the 'secret sauce'. I invested in several companies, including Micro 1. The founder of Micro 1 was not directly involved in selling data to China . He made such a decision . Sachs, what do you think about this new variable—that is, the fact that the key secret of many models lies in the data ? ? Public data on the web is already depleted.

Last week, we talked about taking apart the binding of a book and scanning it . People are looking for data. Several companies, such as Merico and Micro 1, are providing data providing data . In that case, should these companies sell the same data to Chinese open source companies , , , or should they not ? ? ? Well, I think we need to decide what our goal is here . Are you trying to wage an all-out economic war with China? Or are you trying to prevent our companies from doing business in China ?

If ? If ? If that is our goal, we could take that position . Historically, there was a rule that caution should be exercised in the transfer of dual-use technologies, such as military applications . I think data is a kind of product. That is certainly the case with data labeling. If we prevent them from using data labeling , I , I guarantee that China will not face a shortage of labor capable of performing data labeling tasks . Rather, it is highly likely to be insufficient. What I am trying to say is that there are various ways to obtain data.

So, if we ban these companies from selling products to China , we , we should expect that China will take reciprocal measures to ban us from selling things like rare earths. These two countries are not completely independent. Of course, I hope we are as independent and sovereign as possible . We do not want any dependent relationships, but honestly speaking, honestly speaking, honestly speaking, we are currently dependent to some extent . Therefore, I believe we need to ask whether this data is truly proprietary, if dual-use is possible, and if it has military applications.

It is believe we need to ask whether this data is truly proprietary, if dual-use is possible, and if it has military applications. It is certainly not for military use. This is a scientific task, such as hiring PhD holders or top experts to create original datasets . . . China can do that, too, and I guarantee they are already doing so. so. so. I do not think this will give us a decisive advantage in the AI ​​competition . On the contrary, it will cause problems and friction, and friction, and how much do we want to worsen our relationship with them ?

Do you want to risk starting another trade war ? I am not opposed to restrictive measures when they are judged to be effective. For example, I think the Trump administration's administration's restriction on the export of EUV lithography equipment to China was a really good decision . It was probably 2019 . It was a really important decision . Therefore, I believe that goal- oriented strategic control is valid. valid. valid. However, we will have to verify whether this measure meets those standards . . Brad, what do you think about open source data being sold to China and our adversaries ?

Are you concerned about these open source models and us providing them with data ? ? ? First of all, I fully agree with David's opinion . . . We want maximum competition . As we sit here today, the United States is winning. As I mentioned at the beginning , our , our , our advanced research labs are winning , and our open source policy is also winning . And regulations have also been significantly relaxed. She is scheduled to come here in September for a bilateral meeting with the President .

We are developing relationships in various fields . Therefore, everything is proceeding smoothly, and I believe we must continue in this direction. However, I would like to say that some people in Washington will be uncomfortable with this issue . This is because we believe that these measures, along with distillation technology or other technologies, could lead to chip exports. While there is some because we believe that these measures, along with distillation technology or other technologies, could lead to chip exports. While there is some truth to it, one might question whether these measures are making it too easy for Chinese research institutes to catch up with U.S.

research institutes in the competition for advanced information . Jason, this kind of story makes the situation even more confusing, so we will have to keep an eye on it . . . However, I do not think this issue will change our position toward China . This is because we are winning. winning. winning. But what if the President asks his staff someday , that is, about six months from now, "Are we "Are we "Are we winning against China?" If asked, I would suddenly say, "No, we are no longer winning.

They have caught up and overtaken us overtaken us overtaken us ." If you receive that answer , you will be subjected to much more investigation than you are now. I think the only reason all of this is acceptable right now is that we are still ahead in the competition. Looking at Kim think the only reason all of this is acceptable right now is that we are still ahead in the competition. Looking at Kim Kardashian, Gwen Stefani, and the fact that they have used GLM 52 for the past 60 days , it , it , it is truly excellent material, so I do not think giving them favorable conditions is a patriotic act patriotic act patriotic act .

I won't do that . I . I . I ... I'm sorry, but what exactly are you referring to as the advantages of that company ? What is that proprietary dataset you are concerned about ? ? All of these datasets were created by U.S. experts using queries with errors . So, consider the case where you answer 'No' to a question that requires a high level of technical skill, for example, a question related to code, biology, or science case where you answer 'No' to a question that requires a high level of technical skill, for example, a question related to code, biology, or science .

Ph.D. . Ph.D. . Ph.D. holders are holders are holders are entering the latest content and going through the process of verifying it over and over again . This is precisely why the results of LLM (Master of Laws) programs are getting better and better . Essentially, you are helping them catch up . If we hadn't sent this technology to China, it would have been a significant advantage for the United States . I . I think one of the major reasons these models are getting better and better is due to data leakage.

However, what is the reason you think China cannot do this ? There are a huge number of PhD holders in China . For China to conduct research on this scale, it would have to hire the best scientists and experts from the West . So, . So, . So, all Western knowledge is helping the development of our LLM program in a package form . They are reselling the same packages to Chinese companies, which means that China is catching up to us us us that fast that fast that fast .

I think this is one of the big reasons they are catching up . It is similar to the distillation process; it is a really similar process. If we truly have proprietary technology, we do not want to sell it to China. So, we need to look into that part in more detail and verify if there really is proprietary technology proprietary technology proprietary technology . However, . However, the view that this will cause serious disadvantages to China is somewhat different. China produces more graduates in mathematics and science every year than all other countries in the world combined .

It is not that smart people are being sent abroad after graduation because there is a shortage; that is precisely the problem . We need to solve this problem . It . It seems that various complex problems are intertwined. I don't know how many you want to combine , but I agree with the idea that they cannot reconstruct the dataset . . . If there is really exclusive information here, if there is a dual use , if it is military-related information, but I am not sure if this is the case.

Anyway, they were all designed to be proprietary from the start, but I cannot know if they are dual-use because I do not have the dataset. Well, everyone, that was another great episode of the All-In Podcast . Brad, . Brad, thank you so much for being with us. Cha Mas, I hope you have a successful world tour. I hope you are getting plenty of rest, and it will be really hard to find a white turtleneck this season . It's sold out everywhere . So, go to the allin.com store (allin.com/store) .

. 1,000 white sweaters with Cha Mas's autograph will be in stock . You can also apply in advance . All proceeds will be donated to charity. When I say charity, I mean the Chamas Yacht Fund. Well then, I'll see you next week , everyone. goodbye . . Let the winners run to their heart's content. Rain Man David Sax. And as I mentioned, we released it as open source to the fans, and they showed a really enthusiastic response. I love you, Queen S-I Quinoa . . Let the winners run to their heart's content.

Besties is back. Sacks, that's Sacks, that's Sacks, that's my dog ​​peeing on the road in front of your house . I . I think we all need to get a room and throw an orgy party. Everyone is so useless . It feels like I have to relieve sexual tension somehow somehow somehow . . . Wet your feet. Wet your pure feet. Soak it. I need to make merchandise. make merchandise. make merchandise. Besties is back.