All-In Podcast on Tech: 11 Essential Episodes, Ranked

The All-In Podcast tech coverage that holds up: 11 episodes on chips, SaaS economics, space manufacturing, and AI policy, ranked with key takeaways and verbatim quotes from Pat Gelsinger, Jensen Huang, and Gwynne Shotwell. Free to read.

If you want the All-In Podcast at its most technical, start with Pat Gelsinger's Intel postmortem from July 2026 and Jensen Huang's March 2026 conversation about Nvidia. The first explains how a chip monopoly loses a decade. The second explains who captured it. Together they set up the eleven episodes below, which treat the technology industry as an engineering and capital problem: silicon, power, software margins, rockets, and the policy fight wrapped around all of it.

These are the All-In episodes in our summary library where technology is the main thread, mostly from the past year, condensed to eleven. Each entry explains why the episode earns its rank, the specific claims worth keeping, and a verbatim quote from the conversation. Every title links to our full breakdown of that episode, free to read, so you can preview a ninety-minute panel in about three minutes.

Three arguments run through the whole set, and we unpack them after the list: technical leadership as the real moat in hardware, the live repricing of enterprise software once agents can finish the work, and the migration of competitive advantage away from models and toward factories, fabs, and grid capacity.

1. Former Intel CEO on What Went Wrong, What's Next + Lovable CEO on the Real Promise of Vibe Coding

All-In Podcast · with Pat Gelsinger and Anton Osika · 49m · July 2026

The sharpest diagnosis of a technology failure in the All-In catalog, delivered by the person who inherited it. Gelsinger traces Intel's decline to a leadership substitution: business operators displaced technical operators, and business operators promote their own kind. From there the specifics land hard, including the iPhone chip Intel passed on and the foundry model TSMC built while Intel optimized shareholder returns. The Anton Osika segment on Lovable gives the hour its forward-looking half.

Key takeaways

  • Gelsinger's core claim: billion-dollar technology decisions resist spreadsheet analysis, so companies run by non-technical leaders systematically mis-price their own options.
  • Intel's miss on iPhone chips and its late answer to TSMC's foundry model trace back to the same shift from technical to business leadership.
  • The counterexamples he names, Microsoft, Google, and Nvidia, share one trait: deeply technical chief executives.
  • He sizes Taiwan concentration risk in energy terms, citing under three weeks of reserves on the island and brownout consequences he places above the Great Depression.
  • Gelsinger puts AI and quantum computing at the center of the next decade of technical opportunity.
When you're making these hardcore technical, you know, decisions that affect billions of dollars, you don't do that through a spreadsheet. — Pat Gelsinger

Read the full episode summary

2. Jensen Huang: Nvidia's Future, Physical AI, Rise of the Agent, Inference Explosion

All-In Podcast · with Jensen Huang · 1h 6m · March 2026

The best explanation on the show of what Nvidia now actually sells. Huang reframes the company as an AI factory business and walks through disaggregated computing, the architecture that makes reasoning and agentic workloads economic at scale. His headline figure is the one worth carrying around: computation requirements climbed roughly 10,000x in two years as the industry moved from generation to reasoning to agents. He also hands managers a usable adoption benchmark, denominated in tokens per engineer.

Key takeaways

  • Huang's framing: people pay for completed work, so AI economics change once systems finish tasks and go beyond producing answers.
  • The move from generative output to reasoning to agentic systems raised computation requirements roughly 10,000x inside two years.
  • His adoption benchmark is blunt: a $500K engineer should consume at least $250K in AI tokens, and $5K of annual usage signals a badly underused tool.
  • Open-source and proprietary models both occupy his view of the stack, since broad deployment needs each.
  • His stated national security worry is domestic adoption lag while other countries deploy aggressively.
People pay for work, not just information. Talking to a chatbot and getting an answer is super great, helping me do some research—unbelievable. But getting work done, I'll pay for that. — Jensen Huang

Read the full episode summary

3. Elon Musk & Gwynne Shotwell on AI Risks and Peer Review, Starship, Terafab

All-In Podcast · with Gwynne Shotwell and Elon Musk · 1h 4m · September 2026

The most-watched episode in this set, and the one with the least hype per minute. Shotwell describes SpaceX's expansion from launch into connectivity, compute, and chip capacity, which makes it the clearest available window into how a hardware company compounds across decades. The AI content earns its place too: Musk floats a peer-review system in which competing labs test each other's models before release, a concrete governance proposal inside a debate that usually stays abstract.

Key takeaways

  • Shotwell's hiring rule is absolute: hire the best people anywhere, hand them hard engineering problems, and protect their time.
  • SpaceX's roadmap now spans reusable Starship, satellite connectivity, orbital compute, and domestic chip capacity under the Terafab effort.
  • Musk's AI-safety proposal borrows the structure of scientific review: competing labs test one another's models before release.
  • Shotwell's product-risk principle is to obsolete your own products on your own schedule.
  • Her management diagnosis is friction: high performers stay engaged when approvals, duplicate tools, and unclear ownership get cleared out of the week.
We hire the best people. Not the best people that we can. But the best people. And then we give them really hard projects. Really hard engineering problems. And let them fly. — Gwynne Shotwell

Read the full episode summary

4. The IPO Comeback: Why Tech Giants Are Finally Going Public

All-In Podcast · with Brad Gerstner, Andrew Feldman (Cerebras), and Will Marshall (Planet Labs) · 32m · June 2026

Thirty-two minutes with two CEOs who just did the thing everyone else theorizes about. Feldman and Marshall treat the IPO as an operating event with a next morning attached, and Feldman's answer on timing (get it wrong for a decade, then get it right) is the most honest line about company building anywhere in the catalog. Marshall's case for Planet Labs explains why earth observation and AI converge: text-trained models are blind to physical conditions.

Key takeaways

  • Feldman's claim, backed by the studies he cites: more money gets made after an IPO than before, in percentage and absolute terms.
  • The morning after listing, product shipped and engineering progress stand exactly where they stood, with new constituents to communicate with.
  • Feldman's architectural argument for Cerebras rests on latency, since users abandon slow systems within a few seconds.
  • Marshall frames Planet Labs as indexing the earth the way Google indexed the web, making floods, farm fields, and security conditions searchable.
  • Both CEOs treat good timing as the output of a decade of persistence.
I think the first thing is a lot of people asked us about how we got the timing right. And I think the answer is by getting it wrong for a decade. I mean that's really the right way to get timing right. — Andrew Feldman

Read the full episode summary

5. SpaceX-Cursor Deal, SaaS Debt Bomb, New Apple CEO, SPLC Indictment

All-In Podcast · Chamath, Jason, Sacks, Friedberg · 1h 30m · April 2026

The episode that puts a price on AI deflation in enterprise software. The hosts work through SpaceX's $60 billion acquisition of Cursor and Medallia's collapse under private equity ownership as two faces of one trade: coding agents make custom internal tools cheap, which strands leveraged software vendors. Sacks supplies the valuation evidence, naming public SaaS companies at a billion in ARR trading near three times revenue.

Key takeaways

  • The deflation math they walk through: functionality that delivered $10 of value for $3 in SaaS fees can now be rebuilt internally for pennies.
  • Sacks points to public SaaS companies with a billion in ARR, 20% growth, and 80% gross margins trading near three times ARR.
  • Chamath's upside case: an enterprise that cuts half its SaaS budget and redirects the capital into growth expands both the company and the economy.
  • Over-leveraged software companies carry the real risk, since debt service limits their ability to compete with internal AI builds.
  • SpaceX buying Cursor reads as a vertical bet on owning the coding layer that builds everything above it.
If I can suddenly cut, you know, call it 50% of my SaaS budget and I can reinvest that capital in other ways of growing my business instead of managing my expenses all of a sudden my enterprise will grow and the economy will grow. — Chamath Palihapitiya

Read the full episode summary

6. Why AI Will Dwarf Every Tech Revolution Before It: Robots, Manufacturing, AR Glasses From CES 2026

All-In Podcast · with Bob Sternfels (McKinsey) and Hemant Taneja (General Catalyst) · 51m · January 2026

The episode with the hardest number on how AI changes an org chart. Sternfels says McKinsey is growing client-facing roles 25% while cutting non-client roles 25%, which severs the link between revenue growth and headcount growth that consulting ran on for decades. Taneja supplies the capital-side version by setting Stripe's twelve-year climb to roughly $100B against Anthropic's jump from $60B toward several hundred billion inside a year.

Key takeaways

  • Sternfels's structural claim: growth and total headcount have decoupled, so a firm can expand its client-facing side while shrinking support functions.
  • Taneja's compression evidence: Stripe took about 12 to 13 years to reach roughly $100B, while Anthropic moved from $60B toward a couple hundred billion in about a year.
  • Sternfels reframes the core professional skill as question-asking, with AI agents absorbing more of the problem-solving.
  • The CEO tension he describes pits CFO caution about unproven ROI against CIO urgency about disruption.
  • Calacanis's advice to young technologists: skip the resume queue, email the CEO directly, and show up with their landing page redesigned.
We invested in Stripe in 2010. It became a you know a hundred billion dollar company let's say 12 13 years later. You look at Antropic, which we're also investors in, that goes from $60 billion last year to, you know, a couple hundred billion. — Hemant Taneja

Read the full episode summary

7. Software Stocks Implode, Claude's Hit List, State of the Union Reactions

All-In Podcast · Chamath, Jason, Sacks, Friedberg · 1h 21m · February 2026

The clearest record of the moment the AI-versus-SaaS debate changed shape. Chamath marks the shift from asking when software cash flows decline to asking whether they survive at all, and the panel traces a viral doomer scenario that moved stocks in real time. The counterweight arrives through Aaron Levie's supply-and-demand argument, which Sacks reads at length: lowering the cost of a constrained capability raises demand for it.

Key takeaways

  • Chamath's framing of the repricing: the question moved from the timing of software cash flow decline to its existence.
  • The Levie argument Sacks relays: cheaper supply of a previously constrained capability expands total demand for it.
  • Calacanis's operating view is that a structured business process handed to an agent will simply run on schedule, daily or weekly.
  • The hosts cite Derek Thompson on the information problem: real-time data on AI's macroeconomic effects is thin enough that serious discussion turns literary.
  • The productivity cases they discuss involve engineering output multiplying while headcount holds steady.
We used to debate when. When will these cash flows disappear? Now it's like will they even exist? — Chamath Palihapitiya

Read the full episode summary

8. Epstein Files, Is SaaS Dead?, Moltbook Panic, SpaceX xAI Merger

All-In Podcast · with Brad Gerstner · 1h 19m · February 2026

The best hour on what AI does to the software stack, largely because Gerstner adds the multiple-compression argument the hosts tend to skip. Sacks's layer thesis is the useful mental model: incumbents endure as legacy infrastructure while value accrues to a new layer built on top of them. Gerstner then prices that outcome, arguing a CRM can survive the transition and still trade at 17 times free cash flow permanently.

Key takeaways

  • Sacks's layer thesis: AI becomes a new layer above existing software, pushing incumbents toward legacy infrastructure status.
  • Gerstner's valuation version: survival and permanent multiple compression can both hold, from 30x free cash flow down to 17x.
  • Roughly $300B of software market value came off in two days following Anthropic's Claude Co-work announcement.
  • The hosts argue software companies now have to prove AI acceleration in revenue growth, since merely hitting prior targets reads as stagnation.
  • Calacanis's agent setup spans Slack, Notion, and Gmail, and he treats open API access as a condition for staying on a platform.
The risk for the SAS companies, it's not that they get replaced, although that'll happen to some degree, but it's that they become an old layer of the stack that now there's a new layer that gets built on top of it becomes more legacy infrastructure. — David Sacks

Read the full episode summary

9. GPT-6 Hits AGI? Tech Euphoria 2.0, SF Mansion Shortage, NYC Bans AI in Schools

All-In Podcast · Chamath, Jason, Sacks, Friedberg · 1h 31m · September 2026

The most useful episode on cost curves. Chamath's argument is that model competition keeps driving the marginal cost of intelligence down even while hype and late-stage valuations climb, which cleanly separates the capability story from the asset-price story. The Hugging Face agent-security incident gives the panel a concrete case, and Sacks's response (meet AI-powered attacks with AI-powered defense) is the practical position on a problem guardrails handle poorly.

Key takeaways

  • Chamath's cost thesis: broad availability of strong models keeps pushing the price of an incremental unit of intelligence lower.
  • Sacks's security position: AI-powered defense is the realistic answer to AI-powered attacks, given how weakly guardrails perform.
  • The education segment lands on adaptive learning platforms as the default mode of schooling for the next cohort of kids.
  • Calacanis describes using an assistant to audit and consolidate his own bots' skills and instructions, a template for cleaning up accumulated automation.
  • Open versus closed model markets, data center politics, and the reported Venezuela oil arrangement fill out the rest of the hour.
So I think the right way to think about this for most people is we are seeing incredible intelligence capabilities be broadly available. And we are seeing the cost of that incremental unit of intelligence being driven further and further down. — Chamath Palihapitiya

Read the full episode summary

10. Inside America's AI Strategy: Infrastructure, Regulation, and Global Competition

All-In Podcast · with David Sacks, Michael Kratsios, and Maria Bartiromo · 47m · January 2026

The policy episode worth your time, because the people describing the strategy are the ones executing it. Sacks (White House AI and Crypto Czar) and Kratsios lay out a three-pillar plan of out-innovating, building infrastructure, and exporting American AI, then attach a number to the softest variable in it: 83% AI optimism in China against 39% in the United States. Kratsios on the 50-state regulatory patchwork is the specific cost startups absorb.

Key takeaways

  • The three pillars as described: lead on innovation, build the necessary infrastructure, and export American AI technology.
  • Sacks's utilization claim: every GPU going into a data center is in use, with no idle capacity sitting around.
  • The sentiment gap they cite: 83% of people in China see AI as beneficial, against 39% in the United States.
  • Kratsios argues a 50-state regulatory patchwork falls hardest on early-stage companies building on frontier models.
  • Residential electricity rates are the stated political constraint on continued data center buildout.
There's no such thing as a dark GPU right now. Every GPU that's being put in a data center is getting used. — David Sacks

Read the full episode summary

11. Jensen Huang: The Doomer Hoax, Superintelligence is Here, and The Future of AI

All-In Podcast · with Jensen Huang and President Trump · 46m · September 2026

Huang's second appearance in this set, aimed squarely at the policy conversation. His argument is that safety and leadership operate together, which collapses a trade-off the debate usually treats as fixed. The listening value sits in two places: his standard for regulation, which is to solve observed problems, and his claim that the AI race turns on application quality across a whole economy.

Key takeaways

  • Huang's central claim: safety and fast execution are compatible, so the trade-off framing collapses on its own terms.
  • His regulatory test is narrow and practical, holding that regulation should address actual problems.
  • He locates competitive advantage in application, arguing the race goes to whoever exploits the technology best.
  • Nvidia's stated strategy is to move as far up the stack as needed while staying as low as possible.
  • Open and closed models both have roles in his view, alongside distributed data center capacity.
Safety and leadership are not false. They're false choices. You're able to innovate quickly. You're able to execute quickly. And America's able to lead and to do it safely. — Jensen Huang

Read the full episode summary

What these episodes have in common

Technical leadership is the moat in hardware

The four best episodes in this set are all, underneath, about the same variable. Pat Gelsinger supplies the negative case: Intel's decline began when business operators replaced technical operators at the top, and the compounding damage showed up as a passed-over iPhone chip and a decade-late answer to TSMC's foundry model. Jensen Huang, Gwynne Shotwell, and Andrew Feldman supply the positive case from three different industries. Huang can describe disaggregated computing architecture himself. Shotwell talks about Starship, chip capacity, and orbital compute as one engineering portfolio. Feldman argues Cerebras's design choices from latency physics rather than market sizing.

The practical read for anyone evaluating a hardware company: ask whether the people making billion-dollar architecture calls can do the underlying math themselves. Gelsinger's line about decisions that affect billions of dollars failing when routed through a spreadsheet is the whole thesis in one sentence, and it is worth more than any market forecast on this list.

Enterprise software got repriced in real time, and the hosts split on what survives

Three episodes form a sequence across early 2026. In February, Software Stocks Implode catches Chamath moving the question from when software cash flows decline to whether they survive, with roughly $300B of value coming off in two days after Anthropic's Claude Co-work announcement. The Is SaaS Dead? episode adds Sacks's layer thesis, where incumbents persist as legacy infrastructure while value accrues to a layer built above them. By April, the SpaceX-Cursor episode prices the outcome in deal terms: $60 billion for a coding platform, alongside Medallia collapsing under private equity debt.

The genuine conflict sits between two arguments the panel presents in the same breath. Aaron Levie's view, which Sacks relays approvingly, holds that lowering the cost of a constrained capability expands total demand for it. Brad Gerstner's view holds that a CRM can survive and still re-rate permanently from 30 times free cash flow to 17 times. Both can be right at once: the category grows while the vendors who taxed it lose pricing power. Chamath's version is the bull case for everyone else, where an enterprise redirects half a SaaS budget into growth and the economy absorbs the savings.

Advantage is migrating from models toward factories, fabs, and grid capacity

Across the policy and infrastructure episodes, the panel keeps arriving at physical constraints. Huang's reframing of Nvidia as an AI factory business is the cleanest statement of it, backed by a roughly 10,000x rise in computation requirements over two years. Sacks's claim that no dark GPU exists anywhere, with every chip installed in a data center already running, says the same thing from the demand side. Shotwell's Terafab push and Feldman's silicon backlog are two companies acting on that belief with capital. Will Marshall extends it off-planet, arguing that models trained on text are blind to physical conditions and that indexing the earth is the missing input.

The policy consequence is that electricity prices and regulatory geography become engineering constraints. Kratsios makes the sharpest version of that argument: a startup building on a frontier model now has to navigate 50 sets of state rules, and that friction lands hardest on the companies with the fewest lawyers. Sacks pairs it with the political variable, an AI optimism gap of 83% in China against 39% in the United States, which is the number most likely to decide how much of the buildout actually gets permitted.

Every episode referenced

Frequently Asked Questions

What are the best All-In Podcast tech episodes?

Start with three. Pat Gelsinger's July 2026 appearance is the strongest technology episode the show has produced, a full postmortem on how Intel lost its lead to TSMC and Nvidia. Jensen Huang's March 2026 conversation explains the architecture and economics that replaced it. The September 2026 Gwynne Shotwell and Elon Musk episode covers SpaceX across launch, connectivity, compute, and chip manufacturing. Each one is summarized in full above.

Does the All-In Podcast cover Silicon Valley?

Yes, and from inside it. The four hosts are active Silicon Valley investors and operators, so the coverage runs to deal terms, valuation multiples, and hiring economics. Concrete examples from this list: SpaceX's $60 billion Cursor acquisition, public SaaS companies trading near three times ARR, McKinsey growing client-facing headcount 25% while cutting support roles 25%, and San Francisco housing prices as a read on the current tech cycle.

What does the All-In Podcast say about software and SaaS?

The panel's 2026 position is that AI repriced enterprise software permanently. David Sacks frames incumbents as an aging layer of the stack with new value accruing above them. Brad Gerstner argues a surviving CRM can still fall from 30 times free cash flow to 17 times. Chamath Palihapitiya makes the buyer's case, where cutting half a SaaS budget and redirecting it into growth lifts the whole enterprise. Three episodes above cover the sequence.

Is the All-In Podcast good for tech news?

It works best as analysis layered on top of news you already saw. The weekly panel format means a given episode mixes technology with politics and markets, so the tech segments run 20 to 40 minutes inside a 90-minute show. The guest interviews are the exception and the better entry point: Gelsinger, Huang, Shotwell, Andrew Feldman of Cerebras, and Will Marshall of Planet Labs each get a focused conversation.

Which tech guests have appeared on the All-In Podcast?

The episodes ranked here feature Pat Gelsinger (former Intel CEO), Jensen Huang of Nvidia twice, Gwynne Shotwell and Elon Musk of SpaceX, Andrew Feldman of Cerebras, Will Marshall of Planet Labs, Anton Osika of Lovable, Bob Sternfels of McKinsey, Hemant Taneja of General Catalyst, and White House technology officials David Sacks and Michael Kratsios. Our library carries free written summaries of every one of these conversations.

Keep exploring