What Happens When the AI Boom Runs Out of Money
Treat AI infrastructure claims like any capital cycle: separate the technology’s long-term usefulness from the timing of its cash flows. Today, list the AI tools or companies you rely on and ask three questions: what does each use cost at scale, who ultimately pays, and whether capacity is being bui
1h 25mSummary published by 1% Better, updated .
Key Takeaway
Treat AI infrastructure claims like any capital cycle: separate the technology’s long-term usefulness from the timing of its cash flows. Today, list the AI tools or companies you rely on and ask three questions: what does each use cost at scale, who ultimately pays, and whether capacity is being built faster than revenue. This prevents enthusiasm about AI’s potential from becoming blind faith in every AI investment.
Episode Overview
The episode examines whether the AI boom could face a financing and overcapacity crunch before its economic promise is realized. The speakers discuss inference costs, monetization, advertising, cloud infrastructure, semiconductor supply chains, and why AI may resemble historic railway and memory-industry capital cycles.
Key Insights
A Great Technology Can Still Be a Badly Timed Investment
The speakers argue that AI progress can continue even if the sector suffers a sharp financial reset. Like railways and the dot-com era, an infrastructure buildout may create lasting value while investors endure a painful mismatch between upfront capital spending and delayed revenue.
Inference Makes AI Different From Traditional Internet Software
Traditional software and internet distribution often benefited from near-zero marginal costs. AI services require ongoing compute spending, and heavier use—especially test-time scaling—raises the cost to serve, making pricing and unit economics central rather than incidental.
Consumer Monetization Is Hard; Advertising May Be Inevitable
Consumers generally resist paying for productivity software, while businesses can justify spending when tools improve employee output. The conversation suggests that advertising, commerce, and enterprise products may be more durable ways to finance broadly used consumer AI than subscriptions alone.
Shortages Create the Seeds of Oversupply
Compute shortages encourage companies to build more fabs, data centers, and memory capacity. But long construction lead times mean that investment decisions made during scarcity can arrive all at once, producing the same boom-bust dynamics seen in shipping and memory.
Build for Yourself First, Then Sell the Capability
Amazon is presented as a model for AI infrastructure: develop systems for demanding internal use, become the first and best customer, then offer the capability externally. This approach creates feedback loops that improve products before they must compete in the open market.
Frameworks or Models
Aggregation Theory
1. Start with a market where distribution and transaction costs approach zero. 2. Recognize that when supply is abundant, discovery—not distribution—becomes the key bottleneck. 3. Platforms that solve discovery gain users and monetization power through positive feedback loops. 4. Assess whether AI changes those economics by adding meaningful ongoing inference costs.
Amazon’s Internal-Customer Model
1. Build a capability to solve a difficult internal operating problem. 2. Use it internally at scale, making the company the first and most demanding customer. 3. Improve the capability through real usage and measurement. 4. Expose it as a standardized external service once it is reliable and scalable.
Notable Quotes
"The same Time, this Decline Even if it happens, AI disappears. Not going. That Improving It won't stop."
"In the future This Continued To implement Required Profit To earn We built the bridge. Shall we set up? About You can worry."
"Insurance Expected Value Negatively While , nobody Unnecessary For insurance To pay They won't."
"They What for yourself? Even if created , they are the ones who First best As a customer There are . Any The matter To start They measure They offer , they For others They sell ."
Action Items
-
1
Run an AI unit-economics audit
For every AI tool you pay for or deploy, document its monthly price, likely usage intensity, business outcome, and whether its cost increases materially with more use. Remove low-value tools before renewals and focus spending on measurable output.
-
2
Separate technology conviction from valuation conviction
Write two short theses for any AI-related investment: one explaining why the technology matters over 10 years, and another explaining why the company can finance its current spending before revenue arrives. Do not let the first thesis substitute for the second.
-
3
Watch capacity lead indicators
Track announcements on data centers, GPU purchases, fabs, memory, and cloud capital expenditure alongside AI revenue growth. Rising capacity commitments without comparable revenue growth should be treated as a cycle-risk signal.
-
4
Use the internal-customer test
When evaluating an AI product or business, ask whether its creator uses it intensively in its own operations. Prefer tools with a demanding internal use case and a clear path from internal capability to external service.
Full Transcript
Transcript of What Happens When the AI Boom Runs Out of Money from Invest Like The Best. Auto-generated from episode audio; may contain minor errors.
United States Success Getting Very Complicated Will be That I I think. AI If controlled Others Than yours Army Strongly Will be Of Exaggerated Fantasy Environment Let's take it. . In this world China Excellent Game Theory Reaction How? Will there be? TSMC- Breaking the five. Especially Military and National Safety Technically, we A Notable Dominant Status If it reaches, it To the world Very Dangerous That I I think so. If you believe Believe me, We are the last Do this. Seven years It's over. The world is very It has changed, Then AI No.
We Often Collection Theory (aggregation theory) about We talked, that's it. Today's In conversation too That it will come I hope. The world is so much Changed In the situation, AI In the competition United States Success Getting What does it mean? That your Comment Listening To get started A Interesting As a place Will be That I thought. United States Success Getting Very Complicated Will be That I I think. AI If controlled Others Than yours Army Strongly Will be Of Exaggerated Fantasy Environment Let's take it. . Production Department Adjustment Like, real With the world Related AI does things Will do That I I don't think so.
. But this In the world, China Excellent Game Theory Reaction How? Will there be? TSMC- Breaking the five. Game Theory is Very Complicated May change That For me It appears. But this For me That's all. Complicated Unknown. Silicon From the valley Coming With comments , especially a From the laboratory Coming With arguments For me Basic Contradiction There is. Military and national Safety Technically, we A Notable Dominant Status If it reaches, it To the world Very Dangerous That I I think so. But that In context, TSMC Being demolished Beyond this How far?
Going? Because that In the situation, we In America itself Chip Preparing Plants Creating Method We will have learned. , and that Single Terminal Only To depend on We will not. That I think so. Production and Plants or Actuators Like Basic In matters Some kind of Magical Thought To be I I think. We are China's On which To the extent We depend on is Exactly Not understood That I I think, This is a conflict. Without Will not be fixed. . Because, These Everything Fixing Completely Meaninglessly There will be.
Your Competitor From China Items When buying, You From America To buy If you start, Relatively You Very Underprivileged In the position You will be, So You do it. To do You won't. So, different There is no way. In the condition of Only You do it. You will. Like this Big key For things This Applicable. For example , Apple Company iPhone Production A part To India You can change. Them Can be forced . But, Apple Company From China Completely Not leaving Because, That's it for this. A great Example.
They To some extent They are diversifying. , but that The cost is too much. Too much. This is a Insurance Policy Like, When not mandatory So much Large amount Giving To do it Nobody They will not come forward. . This is a Kind Hypothesis, this By me Understand To buy Hard There is. Because, China Independently We Completely Exit, Taiwan They If attacked Even for us Which Affection No The environment is mine. Very Imaginarily It's raining. Such In conversations Reality Facing Character Missing That I I think.
Their In place You Keep Look. Their Purpose What? You Do you think? ? For everyone A Frightening Enemy Needed . AI trading In perspective , we are China To win Want Rather than Best strategy Nothing That I think. Of course We are China. To win Want That too I think. We Competitively To be I want. The last few Over the years, A Specific Politics For overview , China Just like us Trying to change It's sad. . We are for that. Opposite Direction To go Want That I I think so.
More Transparency , more Innovation, From above Under Imposed Controls Less, Freedom of speech Restrictions on These are less All To be I want. United States In the lead Because of existence, That By leading Success Gets. In AI United States Only Completely Dominance Payment To the world Correct position Not that You You said. This is worldwide How to be Want is your Best Expectation What? Current AI status, Taiwan Affair Like; Current The environment is so-so. Worse Unknown. Question What is, How much is this? Time Lasts That is.
But, we More than you think This is long. To last a long time Chance . Currently, OpenAI and Anthropic Companies Border In the area In the lead To be I I think. What's on Google? Happening That To whom? I know, More Grok and Meta Companies Them Following . Meanwhile, The Chinese too Very Talented people , Smart people , more They are this Samples Get 6 to 9 Months Underprivileged In the position To be Attempt They do . This To America A favorable one Good Equally For me It appears.
Now Question What is, How much is this? Time Lasts That is, Isn't it? By China Actually To advance Can Like Many questions There are. Various For reasons About this For me Still A little Doubt There is. 6 to 9 Months Lasts Advanced Reaching the status Very Difficult. Meta too Crooked Actually In progress How? Are operating is Interestingly There will be. AI Improve AI- Yes Used To the world We When going, This is for sure. Happens That I think. OpenAI and Anthropic Both Recently Fast Growing up We see, This Expected Only one.
To that extent That's true. If so, By you Actually Overtake them Is it possible? This Service Fee Or Marginal Cost For both Which one is this? To the extent Applicable That is Another Question. Through AI Your Structure Improve, Data analysis If you do, Your service Fee Others Less than Will there be? Open source Models The thing about This is it. They are That it's free What is said For me Strangely There is, Because Marginal Cost Isn't there? ? You are GLM. Or Gemini Like Inference To operate I have to.
. Gemini's service Very expensive High. A For the answer The cost is significant. Too much There is. So, people Everyone do this. That it's free Saying, Free If It's free. They They think It seems . Now I can use AI. Free of charge To use Can. No, You can use AI Free of charge To use No way. AI To create Required For research You are money. To pay No need. , but it Operating The necessary cost Of course To pay I want. Current status For me Liked , but for this Resistance Let's go, Yes, it is.
Just now So There is. That's right. Won't last is Fair It's resistance. , but For me Unknown. So, we More than you think This is long. Time It can last. How is this? Going is About, Its balance Where will it be located? is About Definitely To say Is it possible? This The S-curve ( S-curve) Length Like? This On the S-curve How much are we? Distance We have arrived. ? A In the grid These are balanced. Can be achieved, Maybe Without reaching Let's go. Future How Will be That To know What are you?
To know Do you want ? Mythos and Hugging Face ( Hugging Face) By events People Being scared , dangers To reduce Instead, we New ones Publishing Stop Lead That I I'm afraid. Thus, Current What is the limit? That Diagnostics We have the ability We will lose. . This is a mistake. Safety Feeling Creates . For example, Everyone Mythos About Fable ( Fable) They decide. , but With mythos Compared Fable Actually How much? Better? That Gap Over time Still Increase That's it. I I think. So, this is a Important Question, for this Answer me.
Unknown. AI His ability Thane Improve This is what I will do. Recycling Recursiveness Question about This. This Any one Big For change (takeoff) Will you go? Of the day In the end, many In ways Term exam ( timing) Questions Arise. For investment Revenue, that Investment Retain To the extent Enough Will there be? The time of Timing mismatch About me I'm worried. . We are capital. Capital curve Moving down We have . We first Free With cash flow (free cash flow) We started. Technical Companies Credit markets Very fast How it was handled Amazing .
This has a Year Only became. Now Google Shares Publishing . Nvidia For this Arrangements Does, For you I know... This is 500. Billion Dollar Valuable Thing. Pension Finances and Insurance Amounts Impressive 500 Billion Dollar plan This. For that What next? Will it happen? For that Then the money From where? Coming? Best In the situation, we Free again Cash flow By this To finance I want. But By then, a Gap Happened, we Enough Revenue Not earned If, a Big decline To meet It will happen, Isn't it?
The same Time, this Decline Even if it happens, AI disappears. Not going. That Improving It won't stop. It continues. Progress. Dot-com Time, Railway As time goes by In history Many things happened Bubbles Like, human Clan Overall In development These Over time Becomes unimportant ; Intermediate These in time For many Great Affect Even if it causes Even. Railway What do we have? Learn Gives? This is the last one. Big Construction Not an attempt, Isn't it? GDP Percentage Basically If you look. About it Forget it. . This In time We still On a large scale Maybe, Or This is it.
Biggest Could have been . Yes. Yes. For you I know, On the railway Real time Duration There was a mismatch. A train The path Tie it in Profit Earning is a Decade Or many Decades Takes It is an attempt. . Short During For that Fee To pay You Money To publish Had to . In the world Money It's over. , right? That The feature is For this That is the reason I I think. Why do people Railway Department They choose If, Everyone Enough Computing Facility Will there be?
That They are talking. . For us Enough Electricity Is it available? Perhaps, Recent Question What is, With us Enough money Will there be? Is that so? About this Thinking Strangely It may seem. 1870- That's what happened. It happened. In the world Money Truly Is it over? ? But Funny Thing What is, Railway Continued It worked, They Western Parts Expanded , more GDP -to (GDP) Their Contribution Too much There was. They are Even now For GDP Contribute Coming. Berkshire Hathaway From the company Google Within the company Money that goes, Railway Came through It's money.
No, this is Very Direct True. Truly Berkshire To Hathaway This Problem There is. Me Regarding, This Nvidia Agreement Google's Share With the release Related; When it happened That's it for me. Shockingly There was. Why is that? Shockingly Was? Because it Google, By them To raise money Can. Then they Why stocks? They publish ? They do this. Completely If you believe, why? Their Profit They are reducing ? But Berkshire Hathaway Comparison Interesting , because Roughly If you say, To them Famous ' Cheese Candies'( See's Candies) is, Isn't it?
Very high Profit margin Occupation That's it. High profit Limited Many In professions The problem is What is, Percentage profit Too much Maybe, But the total Profit Controlled There will be. Opportunities Less. That's right. You Just Money Adding Are you there? . So BNSF ( BNSF) Railway Intelligence What is, They are cheese. Candies Profit Taking , "Low profit Limited Another "Here is the business" That investment They did. But its Total money Very valuable More Because, Low profit In the limit High profit Got it. For example, 2025- In BNSF The company is a During the year Earned Free Cash, Cheese candies Its lifespan Throughout More than earned Too much.
You Low profit Limited Occupation About Even if you speak, This is very High profit Limited Leave the profession Excellent. Berkshire At Hathaway A feature There is, your Capital Biggest When changing, You are the percentage Based From the world Variable, Complete Numbers Basically To act You will start. . That story Why me? Very Interestingly Appeared If so, that Google Accessible Direction Clearly To show It seems. So they On Google Investment To do Very Code Characteristic. Google search called Incredible High profit Gives Occupation Contains, This is all Also in time Excellent Business In models One.
In all Pure Coordinator This is it. All In directions Expanding, For this, which Investment To do No need. For each Additional cost Zero. This Surprisingly There is. Meanwhile, This AI Opportunity There is, for this Biggest Money in quantity Required , which is money A burning one Thing. But, Imagine Look: AI is Intelligence If so, its Market size All Office Tasks Includes, Finally Robotics By Everything Including, Profit is very high. Too much While , Google search A small As a matter of fact Does it appear? That's it.
Happening It appears. In that world, Your whole Cash flow You are using . They did it. Have done . You Hundreds Billion Borrow dollars In the market You are receiving . They did it. Have done . You Shares You are publishing , because the share What is the output? Will you? This Your Shareholders Interest Affects. So, To you Small percentage Stock only There will be. Okay, Biggest Profitability in scale When available , in the end Nobody complains. To do They won't. This is very Code Characteristic That I I thought.
Berkshire That share Publication Symbolically It is clear. They A Google search Investor Not only that, Google Going One for the path As a model Are there any? ? Ramp Your Finance Committee Slim, Fast, Better Change Created The only platform, this For businesses Per year On average 5% Savings Provides, So you In development Attention You can pay. Ramp Customers Average American More than businesses 3.2 times Fast Revenue They raise . Visa, Versailles, Cursor, Stripe, Notion, Eleven Labs, Shopify and 70,000 others Businesses Now On the ramp They are running.
My business too That's right, You too Do the same. I want. At ramp.com/invest Learn more. Open AI, Cursor, Anthropic, Perplexity and Versailles For both A common The thing is. They Everyone WorkOS I They use . At the organizational level Adoption Reach, SSO, SCIM, RBAC and Audit Records Like key Skills You provide I want. These Yourself Create For months Spending Than, first On the same day Everything Get WorkOS APIs Can be used . That's why You Heard Many great AI Groups Already on WorkOS They are running.
Institutional Ready To be , your Preparation Like Important In matters Attention To pay WorkOS is very fast It is through. To get started workos.com Visit . Felix by Rogo is your Company Templates , environment and Standards Using, An ordinary Command To the customer Providing Final As a task A change Personal Finance Assistant . "To Felix" This Comments Take For me Change it. That one Email Send. Or, "This Emails The environment With my Tracker Update "Leave it." Felix For you Ready PowerPoint Decks, Excel Models and With evidence Together Research Return Sending.
Felix Your The work of the committee In the manner Acting, Quickly Precisely 24 hours a day Tasks Finished Gives. Learn more rogo.ai/felix Visit . I have one. Doubt, You Describing Business and Competition Dependent Features Skipping If you look. This Department About Thinking With others Compared , Technology Traditionally Which one are you? AI to the extent You believe ? I like this. Some In matters More Hope and , in some Less Hope and There is. So, For everything Applicable Common In argument For me Complete Agreement No.
For example, AI In programming (coding) Excellent Works . One year People before How do you do this? Did That is for me Surprisingly There is. That is, Truly Codes Write down They had . Of course, This In mathematics Very Excellent. But, these Verifiable Domains That is Transparent True. Verifiable In the fields Better Working, To check Impossible Or long Term Verification Cycles With In the fields Better Active For What evidence? Is there? That Still To be determined Needed Together There is. This Interesting , because I This question I woke up, Then In laboratories Some people who were In a group There were.
Their Answer me. To be irritated Did, Because They me For AI Opposite They considered. They said, "Oh, People By us Or chess? Or Go. Game or To win Can't That They thought , but we Make it easy. "We did it" They said. To that I, By us Chess To win Can That For me Know I said. They are Knowable Domains Because We can go. The game To win Can That For me I know. For both ' Scale As a solution Was, But they Both are one Within the limits There were, Isn't it?
Chess Or Go. Non, Truly A new one and Unknown In place, Before Impossible Things Will do What is an example? So, this is it. I still Completely Unbelievable part. However, On the Internet All in Data With Training AI provided . On the Internet All in Data is also a Filtered (distillation) form. This is human. Thought Everything Filtered . No, it is. Like that Didn't. This is human. Of thought Final status Only Filtered . On Reddit Typing What was done Only. In it For thought Traces No, okay?
That's really it. That Opinion, Feeling Or that The article For writing Or that Comment For posting Behind Something that was Not having . We Neuralink ( Neuralink) that Something If you say What will happen? Neuralink's Real benefit Human Of thought Traces Hold on, this Samples Skills Dramatic In the manner To expand If there is What to do? In this world, Verifiable Characteristic My concerns What is, More Data Getting Through it Let's fix it. . Plenty Jobs and Economic Activities, Artificial Intelligence Better To act Impossible In the departments To be I I don't think so.
. Actually, One in the world Specific To the extent Conscious Artificial Insights Many others like There are. They Verifiable In the departments Very Better They are working. . For them Work Given , they did it They do . This Humanity About Some kind of Desperate Visually It may seem, But that The market is very Big, So Models Now Rather than being Even if it doesn't progress , Economic Opportunities Actually Huge ones . I have some For the time being Previously a Article I wrote, For you I know, Acceleration ( accelerationist) Movement There is.
I Me a Reluctantly Together Accelerator That Call me. I got it. We Forward To go Want That I think, Because By us Backwards To go No way. We Doable The worst The thing, Will be On the spot Just getting stuck. . So, In economics Its impact and Monetization Basically Its Benefits Regarding, I am artificial. Intelligence I support. Deadline About For me Unknown. Towards it Progress How Will there be? Law or Medicine Departments like Let's take it. , them You Verifiable That Do you think That For me Unknown.
The law is A kind of Code Like. In medicine Things About Specific Understanding With us There is. Medical field Biggest In opportunities That one I I think. Yes. This Biggest In opportunities One, the same Religion is very Challenging Also, because Included in this All Terms and conditions Restrictions Such. You are a Artificial Intelligence Or Mechanical Learning All Medical Also in the records To use If possible, Very short During We Detectable Discoveries Improved Treatments Unbelievable There will be. So, that Very Confident Vision. On the other hand, it When Will it happen?
Humans My opinion on Confident Vision What is, Requirements To create Our ability Infinite. So, Over time New Opportunities And jobs In creating We are very Better Let's act. That I I think. This is still Desperate In the manner To say If you want , unnecessary Problems Obstacles In creating Our ability To some extent Infinite. For you You know, our In economics How much? Jobs Just Time For subtraction , sluggishly To be Created is For me Unknown. 2010s For starters Let's go, Or Yours Collection Theory (aggregation theory) your In mind Developed Could have had , you can do it in 2015- In You published .
That Theory, that Idea, Technology That During the period Who wins? Those who have received, Who Losers That Decided To say that Exactly It will be, it will be You In short To remind Is it possible? Finance And the market Value Basically, This Of the period Winners Which Theory Or Policies Decisive That you Do you think? To know whether I am interested. I am. That's a good one. Question. Collection Theory is About myself Many times I have thought. . Which one is this? Same as size To the point Applicable?
Yes, yes. Collection Theory Main In the components One is 'zero' Edge Zero marginal cost That is, People for this Resistance They report . Zero Edge Cost is high In ways Revealing . At the beginning I Of distribution On it Attention I paid. People said, "Oh, To me Distribution No. I For advertisements To Google To pay I have to. "That They will say. To that I said, " No, You have a Website There is. Your Problem Distribution No. Your Problem What is To you There is no demand.
Accumulators Need Their Under control Because I have , you For advertisements Money You are paying. "That I will tell you. They Need Their Under control To keep Reason, Abundant Materials In the world Difficult Problem Distribution No, that's it. Discovery It is. You Searching The matter How Actually Detect? Therefore, their In the field Detection Problem Solving Companies That In the market Dominance Pay . They are a Positive Feedback Cycle They receive. In short If you say, This is it. Collection Theory. Another Thing Transaction Expenses. In which Transaction Expenses Nothing.
By Google World For the whole To expand Can. By them User On the side Not only that, Monetization On the side too World For the whole To expand Can. Google or In Meta Advertisement Among those who do Most people Google or Meta's Any To the person No talking. They Directly Go Advertisements They buy . All By computers Are made , commercially That For computers Cost Zero. Artificial Artificial Intelligence (AI) This is when it comes Significantly Changing. Their Speculative expenses ( inference costs) Real ones. But Again, they How much?
Are they real? Currently they are Only the real ones. . How much? Real ones is For me Unknown. Is that so? That The company Depends, But the previous Examples These are better Very Real ones. Okay, let's For example Gross profit Limit or Like that Something You If you look. Of course, But To you This Incredible Gap There is. So, AI today Used In people Most people, That Basically To Google Alternatively Or cooking Notes Creating As a tool Or is it Like For something They use That I think.
For them Service To do The cost is very high. Less is less. My doubt. And a Internet Page For them To provide Equal This is the cost. I Thinking What is, This is a bit Too much Maybe, But that's all. Too much There will be no. On the other end, Test Time Scaling Actually Used People There are , right? Previously Models Zooming in Only through We are scale. Done We had . Now You Time Depending Scale Let's do it, That is About the answer How long?
Are you thinking? is Important. You are that. About the answer The days, Weeks Or Months Let's think. , it is directly Additional cost Cause. Every Second You Thinking More money To spend , which is AI and Inference About us Thinking The way About Talking. I love you. Against I heard that. About, But The truth If you say, Casual Chat GPT Used For the user, Mathematics Theorem Solve Will try For the user too Between Cost Completely Something different. This Challenge Companies In size You Very Interesting In the manner Let's see.
That I think. Microsoft Recently Their Institutional Plans Modified Coming. They are E7 Plan Introduced , this is a For the user $100 per month , in which Specific Size Usage Includes. But even that Beyond For use They Fee They charge . For Microsoft This is a Difficult That the situation I I think, Because Microsoft Type About Positively Thinkers , as a business For you Required Everything They They do That They believe . Every Personal Part Best Even if not , you Everything At the same price You are receiving , they Often Together Are operating .
You are a Small or Medium As a business Or big As a company However, In it Real Value There is. That's true. This Life Makes it easier . You How much money? You are paying. That To think Begin At the moment, It is a new one. Finally Changing, And that Employees With numbers Not bound , right? Microsoft Benefit received What is, You are a New Employee For work When taking , that Employee's About the cost You will think. , in which Their For license $100 per month Or $50 Cost Includes.
This Microsoft For the company A Thoughtless Revenue As evidence There was. Now, You The app About If you think about it, Every Monthly How much does it cost? To do Want To think that Will have to . This is two. Problems Creates . First, Most Companies This To do Not ready. They During the year Just once Budget They will put it. Monthly Basically Budget About Thinking This idea For them I don't understand. They Capital Expenses (CAPEX) Or Once Will do Expenses Only about To think They are used to it.
I Referring For the employee Cost, capital Not the cost, But that Almost It's like that. You At the very beginning The result Take it away , then it To think about You won't. Conclusion Already Taken Considered . But The app About When thinking , you Again Again To do Will have to . Finally, Every Monthly Microsoft Bill Looking, How much? We used That When you see , what is money for? We pay. That To think You will start. . Every Preparation How much? Best That You will think.
. Just like this Separately To give Shall we start? To think that I want. More Tokens Using AI- Handling Serious For the user, 100 per month Than a dollar Microsoft For the company More It costs, That's why They do this. To do Had to That I think. By them That Users Bear No way. But, Most For employees Same standard To continue the price They want , because Serious Users About Customers A little To think Want That they They expect . However, They More Thinking They Don't like , because it This model Strange In ways Destroy .
Low At cost Service Receive For users One more Best Business Model Not formed In that For you Isn't it amazing? ? Google and Facebook Like Companies Previous In time Business Completed . They do this. Still Found Unknown. For me Disappointingly There is, But Surprisingly No. This Of course By advertisements To be supported The market. That's why Advertisement Always Consumer Business As a model There is. Consumer To pay Don't like . Silicone fence Every 10 Per year Once Again To learn Needed Consumer Two about Things There are.
First, Consumer For software To pay Don't like . Second, For consumers Productivity About No worries. Early SaaS During We do this. Pass We have arrived. . In my mind For this Best Example Dropbox ( Dropbox) company. Dropbox A Incredible Preparation. Especially that First of all When introduced . Business At school While studying , Dropbox Used First In people I am one too. That Biggest Welcome Received. My Free In Dropbox Even more Storage facility There is, Because my Code For many I gave it. Drew Houston This Wonderful Preparation Created , this To use Very Easy and Seamless.
About this He is very Clearly Was That I think. He is a Consumer The company Create Liked. He is Steve. Jobs Met a Famous story Yes, apple. Company Dropbox Interested in buying Was, But they They themselves are a The company Create I wish. They said, Steve for that You are a Not the company, A feature Only He said. Apple Company iCloud Drive Feature Created, Dropbox Very fast Grown, But then For them Two Years Stagnation Occurred. Those two Year In stagnation, They The app Absolutely To modify Had to , because Enough Consumer Money for that.
To pay They won't. Companies Its value Realize and pay Paying, But For companies Permissions, Control and Many Managing Need convenience, Their The processor is for that. Suitable There was none. So they The whole Again To build Had to , For companies Just selling. To earn income That the only way They felt. Companies Why money? Pay ? Because Companies For employees Salary They give, So Employees High production Ability As having If changed, For them Investment More on Profit Available. This For consumers Completely The opposite. Consumer, Day Throughout Work After doing it To home After arriving, why?
Again Hard To work Want That They think . I am on the sofa. Sitting Reels Just look. I would like . You can do this. AI In the matter You see. . More Advertisements On a Common Disbelief There is. Advertisements Respectful As one Because, My deck Read to me Highly recommended Got it. My old Articles Back When you see , they are correct Direction However, That's all. Better No, that's not true. It appears. However, Advertisements Written about I am the only person. Because I have more Support Got it.
Everyone Blog and On Twitter To be Wanting In the world, no one Advertisements To talk about Don't like . Even now Silicon On the fence, Advertisements Money through In earning Some kind of Shyness prevails, Especially Last 8 Over the years Facebook's Bad By name, Best Engineers This On the issue To work Don't like . That is, open AI (OpenAI) Dropbox ( Dropbox) story Again Once Performed, But it's a hundred times bigger In size There was. That is, no, We For consumers Subscriptions To sell We are going.
That they They said. As is They did, A lot They sold, But that Enough No. You Consumer In the market To be Are you going? If, Advertisements Just do it. I want. Now They Advertisements They do . Strange Thing What is, They are the same. In time For companies Services Towards They are returning. , because Anthropic ( Anthropic) them Backwards Pushing, So What are they? They do is Exactly I don't understand. They Advertising Most of the features Fast Published They are coming. For example, Limitations Determining and Retail With sellers Link To cause , thereby A purchase Did it happen?
Whether Knowing, All Monitoring And that Like Things To do Can. So that How? Going That I see. Very Curious I am. ChatGPT ) Successful Immediately They In advertisements Attention If paid , now To them A great Advertising Preparation Would have been That I I think. If only it were so. Google Big In trouble Would have been That I think. Meta (Meta) Big In trouble Would have been , because To you This cycle If so, Consumer By Advertisement Doing Special What is, Consumers Income generating Ability. Prices will rise.
When, that too Rising, yes. Because Advertisers Price increase Facing . So, there Elasticity (elasticity) related Problems Nothing. To the consumer Fee When charging , Netflix ( Netflix) Raise If desired, This is it. Their Subscription In the plan The problem is, Consumer Away To go Or Service To give up Before How much? Price Raise Can Whether They To think Want, Isn't it? To the people Money Charging Difficult. Yes. For the people For free Items Providing Easy. Open AI This is before Not doing Very It's sad. .
You are some Big Investment Companies and capital With evidence Time Spending I know. Currently Their How is it interesting? There is and The future About them What They think ? Because of this Annual capital Expenditure (CapEx) About 800 In billions Will be That I think. Next year 1.3 Trillion It will be That Current Rating Says. That From there Continued Increasingly There will be. Will be installed All Computing Ability We Immediately Using We leave. How much is this? Strange If the environment, Ability Online As soon as it arrives Immediately To use It ends.
Yes, that's it. Thing. Currently some Term Contraindications Happened Have . On Twitter Everyone in Always, With us Enough Computing No ability That They are saying. . With us Enough Computing No ability. Well, 2023 and Enough in 2024 Investment Because it wasn't done , we have Enough Computer skills No. Of course, it is True. Also, Enough Computer skills No, that's not true. You If you think about it, TSMC Company 2023, 2024 And 2025. Over the years Their Development Rate Reduced . So, next Some Over the years Ours Computer skills Shortage Still To worsen Going.
Because, a Production The center (fab) To create Time limit Data center More than. So, today Enough Computer skills No, that's not true. We When I say, Companies Invest today Money making Immediately tomorrow Computer Skillfully It won't change. Yes. No, it is. All 2028 And 2029. Over the years It's the computer. Skillfully Revealed. Year Report At meetings, Andy Jazzy too Satya Nadella, " We are the data. Centers Only "We are building" That They say . That is, these Just building Settings Only. We Now Those Without using Maybe, In the future Can be used , more needed While Only We are GPU- Friends We buy.
" This To say A great story. How much of it? Truth There is That For me Unknown, That Often Just talk That's it. I think. Because, Reality What is, You are that. The building If you tie it , that money Right there Will be paralyzed . You do it. To lie idle You won't let go. . That is, You are a Standard Investment If you did , that's it Commodity Markets ( commodity markets) full Logic too. Technical To the department Overall Commodity Markets About I don't understand.
That I think. Technical Department Often This is where Attention Paying : I am a Unique Preparation If created -That As software Maybe, Developers With As a network Maybe, Or social As a network Maybe- I Uniquely While , high price Determining My profit through Limit Can be obtained . For this Best Example Apple Company, Isn't it? To them Own Ecology, Software and Third Party Applications As everything Because, By them 50% off on iPhone To make a profit It ends. Everyone An apple Best Business As a model They are watching.
. Yes, that's it. A profession Conduct. But a Commodity In the market, Price Marginal By distributor Determined . For service Cost only Important. That's right. In Taiwan Ship Transportation In the field My existing Good friend One Was. Hmm, that's a Interesting Department. This is the plane. Companies Like, That Department I Observing For me Like. You are a The ship You are buying. If, That Ship's Cost is Its Depreciation Only, Your Marginal Cost Actually Very Less. This The ship To operate Fuel and For the crew Cost, and Your Port Fees Like.
That's all. Not much. This means What is, You are that. The ship Completely You will operate , that is, which Anyway You do it. You will operate . Also, Your Marginal Expenses To compensate, a Container Price How much? To keep it low If you want That's all. Less You will reduce . This In the environment, Your paper The losses are too great. Too much Maybe, Because Your Accounting At a loss Depreciation Includes. But, that Depreciation is a Accounting Fantasy Only. You Already That Money You have paid.
. Therefore, the market What price? Hold on, that At the price You The ship You will operate . Container Beauty What is, It is a pure It is a commodity. So, Market The price is Marginal At cost There will be. Now, that Enough If it decreases, A In the grid People They will leave. , because Their Marginal Expenses Actually Them To lose Will do, Isn't it? Just Paper money Not only that, Real money Loss Will happen. They They will leave. , but Distribution When decreasing, Price again Rising, Like this In the market Entering Exiting A contact There will be.
But Covid Like time The market is very Too much If there is," Wow, now Enough Distribution Because of absence We are a lot. Money We earn. "That Appears. Then Of ships Distribution Enough No. So, Containers Price usually There will be $3,000, $ $4,000 to $17,000 , $18,000 High. This ship Companies Very short During Earned money Believe it or not Impossible. So then What's next? Will it happen? Okay, More Ships. For us Still More Ships If only How Will be Imagine that Done Look, Okay? Problem What is, A ship 2 years to build It is.
So, Everyone is the same. In time This decision If you take it, Suddenly To you A lot Ships Will be, The price will fall. Let's do this. In the elements, Especially In memory We see. Memory market Its load For the dead Very Popular, People To the market Late They will enter. But the data Which centers To the extent Memory As producers Change Are they going? Now With us Enough Compute No ability is For everyone I know. So, High profit Can earn Therefore, we Definitely To invest Want That Everyone They think .
And our Repayment Payment Time Look. Problem What is, Shortage Prevailing During Your Repayment Payment Time You You are calculating. . That Repayment Payment Time Excessive Distribution In Also in time Will it last? Moreover, the market Supporters Never Distribution Too much There will be no That They will say. AI, We Always In short supply Let's stay. We Always With a shortage Let's stay. Maybe we As is Maybe. My concern What is, That's true. However, Now in this More money Investment Because it is done A gap may occur, Investment Exhausted The state of being To cope Enough Revenue Still Not available .
Again I tell you, I am artificial. Intelligence I hope. This is a Real That's the thing. I I think. This Economic Impact Biggest Will be That I I think. Social impact Regarding All Concerns Very Real ones And they A key On the way Revealed That I I think. You Everything Even if you believe, In the future This Continued To implement Required Profit To earn We built the bridge. Shall we set up? About You can worry. You can TSMC and Fabs Related Some Spare parts Producers Regarding To explain Is it possible?
So far, my According to understanding, They Their Production Ability In expanding , new Fabs In creating , of the market As needed In development Very Carefully Have been , they Didn't. Maybe this This whole The matter Controller , such Biggest Excessive From the structure Us If you wait, it Good. We are some Different Things Let's talk about. Memory In the field From Let's get started. A lot before Memory Producers There were. Every time A development Even when it happens , Memory Producers Again Inside the market They will enter.
Or new Countries inside Coming, For example A in Taiwan During Memory market There was. But, you This Specific Just the mechanics. You will see. For memory Shortage If it happens, More money To earn Can, Because By you Immediately Production Ability To increase No way. This Ship Transportation It is the same. Now we Watching Like This too There is. So, this People Inside the market It will encourage you to come. Excessive Production Ability When it occurs, Prices Collapsed, People In the market From Will be expelled .
Because, The beginning of this Investment is very Too much To exist It's a problem. A ship Like buying, A fab Building More than that Difficult. Memory now Of the product Leading In the position Such EUV Machines are used . So, this Production For series Expenses Billion Dollars Reaching. Every time This development and Fall Cycles Enough, some people Inside They will enter, Many others Will be expelled . This memory Cycles Regarding Many in history Important Moments There are. More Companies As is They collapse. . Memory In the market Samsung How Dominance Paid is Very Interesting Story.
Market Slowly While Investment To do The market Capturing The way They Through history They felt . Too big for that Courage, Control , money too Needed. They did it. Done Japanese Companies From the market They sidelined . Only then South Koreans The market Completely They captured , but Finally Three Companies Only The rest. In which The problem is What is, This is a monopoly. No right However, Some kind of Companies Dominantly It has changed. Past tense Mistakes Again Don't do it. In that They Definitely There are.
They Together Not working However, Those mistakes To avoid Want The only thing Comment They have. This environment Current Status Affected . Memory For need Long-term Change one Has occurred Whether They To understand For a while Took. So, Memory Shortage Problem Over time That it will be over I I think. They Facing Another Danger Apple Company Try. Apple Chinese To get memory Lobby Does, Isn't it? Algorithm Changes Primary What is the purpose? Less Memory How? Using That is That's it. Memory Producers For themselves Themselves Great danger Resulting Long with Term Basically Themselves Destroying I I think.
Memory Producers I am with Iran. I compare. . Hormuz In the strait The problem is What is, That's very Powerful As a tool There is. That You Without using Only when That is more useful. Gives, Because it Always a As a threat There will be. Now They did it. They used , it works Did. But United Arab Emirates Emirates and Saudi Arabia Pipes New Ports Phase They are going. From now on To walk like this They They won't let go. . Now this Very Painfully Maybe, But in 2035 Iran Hormuz To close the strait If you think, Which one is it?
The result Does not cause, Because Alternative to that Ways Will be created . Memory Producers Similar The situation is the same. They have caused That is My concern. Memory In the matter Nobody Again Such In trouble To get stuck They won't. TSMC In the case Status Even worse, Because there Only one Company Only There is. Leading In technology Only one Company Only There is. Of course, Intel and Samsung that Place To catch Attempt They do. This is the same. Like That's the thing. All Markets Dangers Have.
This danger Finally Who They are going to carry That is Great question. TSMC Technology Companies Risk on Which To the extent Has imposed That Most Technical Companies Exactly I don't understand. That is my Comment. TSMC- To Worrying Very Important Danger What is, That Production Overcapacity It is just being. We Excessive Production Ability If created , it costs Only Not adding. , on the contrary Exactly Unused Standard Expenses Creates . But, We are a Factory ( fab) If created , that's thirty Years Running That We expect .
Many For years More than necessary Excessive Production We have the ability In the structure We have. So, they Very Carefully In action Interest They show . One of these Cultural Background There is. Samsung Story Similarly, TSMC's Very Interesting In stories One, Morris Song 2000s Finally Retired. New leadership Took charge , at the same time Great Economic Slowdown Occurred. So, They Their Planned Investment They reduced . He (Morris) Chang) again Come, Everyone Dismissal Did. He, " iPhone Just now Introduced , "he said." This We are so far Upon seeing The biggest Opportunity.
"" We invest To do Want, Cost To reduce No. "" They That Economic Even in a recession , collapse Also in time Investment They continued. " That's it. In time State-of-the-art Semiconductor In the field They First place To catch As a foundation Set. Morris Chang A Unique , of history The best and Impact Caused Technical In administrators He is also one. That I I think so. Fabulous Model Technical To the world Very Important, Such Brave The result is that In the environment Picking up, Especially Generally Risk To take A hesitant one In culture Being Not easy.
The truth By the way , TSMC is very Conservative Trend As if having There was. Okay, What now? What's happening? Where is the danger? Gone? TSMC," We are this Risk To take Don't want "That It has been said. Danger Never Will not disappear . It is a From the place Another To the place Moving. Now Where is the danger? There is If, Every big Technical The company also," With us Extra Computing Ability If so, High profit Might have earned "That Feels. Therefore, TSMC To them Handed over Of danger As an expression, Slipped Income, Profit Plenty.
Danger It doesn't disappear. . That To others Handed over . Some At times This danger At a loss Not appearing , rather profit Unearned As an environment Revealing . Money now Unearned To exist Reason, They are using 5G. About very much Excitedly It was. 2020, 2021, 2022 During periods They Their Factories Expansion In doing Great Investment They did. Then They said, "Okay, For now "This is enough." That They thought . What I said Like, 2024, ChatGPD 2022- Came in, in 2023 Technical The biggest in the world As a matter of fact Was, But in 2024, Their Development Rate At least.
2025- In their Development Rate Collapsed. In 2026, Now it is Again Has risen. . This is very Funny , because I About this A little For the time being Before I had written . Then, one Or two Revenue For calls Previously, CC Wei, CEO and Chairman, Suddenly AI- Yin Applications Regarding Before Always Absent Full in kind Revenue Also on call Spoke. This is the market. Developers Fearful Problem. Usually a Bullwhip Result Will be, They are its In conclusion Being Count They are worried. , where needed Evolving Chain Series Down the road When it comes, They are those At the end There are , they are more Investment When doing, It already is.
It's getting late. , thus They Their money They waste . In the case of AI This is a Bullwhip As a result If so, That's it. History Very long Bullwhip As a result Will be That I think. Still Create What you need There are many, This Companies Will be Asia this Message To understand For a while Took . But, They Often This They understood. That I think. But, Message Understood Then, it Truly Many to be revealed Years It is. Computing Deficit In consideration With this How long?
Takes That you Do you think? ? Intel and Samsung's Logic For business What is this? Indicates That is Interesting Thing. I am many. For years TSMC This on Dependency Problem Write about Bye. In fact, 2013- I wrote in First In articles One, For Intel You are a Fab Business Create Want That Advised . You are no longer Just As a designer Only To be No way. Chips In preparation A Biggest There is business. That As written I It's late. I thought. 2010- Cloud in the cloud In the waves They travel.
When done Their Stock market Value Sky Touched, in 2020 That's them. We We are lagging behind. Whether They felt . Importantly, Here is a Big Opportunity There is. We For that Completely Not ready. DSMC- To Will be Customer Service Attitude, Culture, IP Construction Blocks etc. We have No, one Real Boundary Business Create For them Customers Need. So I like this. As a problem I wrote, Chinese The problem About I wrote, That is, our Biggest Geography Politics The enemy's From the beach 60 miles Far away A The company You You are dependent.
. So, these All Big Problems. From there. This insurance Regarding I am the matter. I understand. . A big one Technical Company Intel Go, Intel, Our Chips You Create That To say. And a The thing is, because Biggest Use For you. Coming soon , because a With a partner How it works To do Whether You You will learn. , more For us. Biggest Difficulty, Because With you How it works To do That We To find out I want. We Directly from TSMC- To Could have gone .
They are very The best. With them Working Very simple, So they Good job. They will do That For us I know. Nobody Intel With the company To work Thought Logically Exactly Did not appear. That's it. Their Basic Problem. In an unchanging world, TSMC always Success Receiving There will be. But, memory Producers The same mistake made TSMC some In matters Did, I This Analogy If continued, Iran did The same Mistakes They They did. The last few Over the years They Investment Because I didn't do it, Shortage Very Severely It is going to be .
Enough Computing Facility Because of absence More Income Profit Lost big Technical Companies, Intel and Samsung Companies Logic Improving Difficulty To undertake Ready There will be. This It's a shortage. Finally IntelliJ Saved . In the future A In the grid They are a Main Partner First of all They will announce That I expect , that's great As a matter of fact Will be, But this Finally TSMC To oneself Made . At a high price Very expensive The solution is This is it. We do it. Avoid Alternative The way Let's search.
Yes. You Like, a As a researcher On the side Sitting All this When writing I Learned Thing This is it: Insurance Expected Value Negatively While , nobody Unnecessary For insurance To pay They won't. TSMC On Geography Politics Dependence Problem The only solution Way, Biggest Computing The app Just bring it. ; Only then Others To improve For everyone Economic Incentive Available, Through it We have this Geography Politics Insurance Free of charge Available. Top 10 or 15 Technical Companies About If you think about it, Whose today? Business organization Very Interestingly To be You Do you think?
? The answer to that Always It's Amazon. Why Amazon? So much Impressive If, They What for yourself? Even if created , they are the ones who First best As a customer There are . Any The matter To start They measure They offer , they For others They sell . AWS for this Very clear Example. AWS, Generally People To think On the contrary, Amazon's Surplus capacity Not. Actually, amazon.com to AWS To bring Long time Took. But that For us Taught What is, More Meetings Without Expandable A system Need That is.
API Only through Operable Computer skills We need it. To anyone To speak No need, It's right there. There will be. Just like that. Our Domestic Retail Business For groups If you do, do it. To whom Whatever Let's do it. Retail Trade The biggest Because From others. Let's start. I want. AWS Domestic For customers Beforehand Foreign For customers To serve Started, But now For everyone Service Does. To you Logistics Other like Products There are, aren't there? There for it Opposite The situation is there. Now We Our For logistics UPS, FedEx and USPS Like Outdoor Companies We use .
We do this ourselves. Create I want. Now They did it. Create, Third To the parties They offer , right? Now Others too Their Delivery Services Can be used . Every Also in the market You can do this. Let's see. They Their AI Products Or chip Products About They are talking. , right? Graviton Or Trainium's Special feature What, Especially Early In versions? Early Versions Very bad There were. But You On Amazon Their Redshift Database service Like Managed Services If used , what is in it There is an app.
That they To you To say They won't. You are a Managed Service Only You are buying. . So they Selling Services Under their Bad Processors Let's put this. For them Improve Best Change Provides opportunity . They Them Outside Selling Status They reach. They are Graviton- Best for First As a customer Because there was, That Advanced. For Trainium Best first As a customer Because there was, Trainium Advanced . Trainium now By Anthropic Running For you I know. They are AI For products This is it. They do .
Which of them Success Receives That Let's see. To them Call center Software There is. Their Call center Or Customer Experience AI By Taking place. Of course, it is Very good There is. You do it. Did you use it? That For me Unknown. Because I To America When I returned , a lot Items I bought it. Every Even in summer I am very Short During A lot Items I will buy it. Last one During the year, You are a To the chatbot You are talking. is Knowingly Let's talk, but That chatbot Very Better Works .
That's really it. That Problem Solves . So, that's it. In category To act Starting You Let's see. But, they Their own For business These AI services They create , they Widely Available They will. Some of them Will work, Some work No. But, this is a Very Elegant Approach. More real In the world They are many. Investments Because of having done , AI model From the form Their Main Business Not affected As