Teahose.
SIGN IN
NEW HERE — WHAT TEAHOSE DOES
We read the entire AI & tech firehose — so you don't have to.
PODPodcastsAll-In, No Priors, Acquired…
NEWNewslettersStratechery, Newcomer…
PAPPapersPhysical AI research
PHProduct Huntdaily launches
VCInvestor ScoutSequoia, a16z, Benchmark…
CLAUDE DISTILLS →
7 reads, 30 sec each — free, 6 AM ET.
+ a live graph of the companies, people & themes underneath.
HOME/SOURCERY/Gavin Baker on Nvidia's Crisis,…
POD
// EPISODE
SOURCERY

Gavin Baker on Nvidia's Crisis, Jensen Huang's Comeback & the AI Selloff

DATE November 19, 2026SOURCE SOURCERYPARTICIPANTS GAVIN BAKER, MOLLY O'SHEA
// KEY TAKEAWAYS6 ITEMS
  1. 01The Market's AI "Return on Capex" Math Is Flawed Because It Ignores the Open-Source Economy
  2. 02Open-Source Tokens Carry Lower Margins, Which Drags More Compute Demand
  3. 03Enterprise "AI Factories" Are a Hidden Demand Signal
  4. 04Fundamentals Strong, Stocks Discounting a Demand Cliff: "Voting Machine" vs. "Weighing Machine"
  5. 05Supplier Datapoints Confirm Acceleration
  6. 06Jensen Huang Is Irreplaceable: The Case for Patience at Nvidia's Board
In this episode

1. Key Themes

The Market's AI "Return on Capex" Math Is Flawed Because It Ignores the Open-Source Economy

Baker argues the market is dividing roughly $1–2 trillion in AI capex by a small set of visible frontier-lab revenues and concluding the return is poor. He says this misses an entire layer of demand: open-source models customized for enterprises.

"The market is saying, you know, we're spending, let's just make the math easy. You know, whatever it is, a trillion dollars in CapEx." 00:01:46

"...80 billion in Anthropic revenue, plus the OpenAI revenue. And, you know, it seems like that's been reported to be around 40. So it's 120 billion... And you still get to something under $200 billion. And that is the return on this AI investment. And I think that that calculation is flawed..." 00:02:09 (Gavin Baker)

Open-Source Tokens Carry Lower Margins, Which Drags More Compute Demand

Baker highlights Fireworks' growth (the CEO, Lin Qiao, spoke on stage) as proof of a large, under-counted demand pool. A non-obvious point is that lower-margin open-source tokens pull more compute with them.

"...her top customer, it is 13X in the first nine or 10 months of this year. And she is fundamentally serving open source models that she, that Fireworks is customizing and specializing for their customers' use cases." 00:02:35 (Gavin Baker)

"...because the margins are lower on these open source tokens, it drags more compute demand with it." 00:03:29 (Gavin Baker)

Enterprise "AI Factories" Are a Hidden Demand Signal

Michael Dell's disclosure shows enterprise adoption is accelerating and is invisible in the market's math.

"...in the first two years of offering AI factories, which is 24 and 25, they sold 3,200 of them. And they've sold, I believe, 3,300 in the last nine months. So these are over 6,000 discrete enterprises who are buying GPU servers and associated equipment from Dell and running open source AI on those servers and getting a return. And it does not show up anywhere in that calculation that the market is doing." 00:03:29 (Gavin Baker)

Fundamentals Strong, Stocks Discounting a Demand Cliff: "Voting Machine" vs. "Weighing Machine"

Baker sees a disconnect between on-the-ground demand and share prices, and frames the selloff as a fixation on second derivatives.

"...stocks are not trading like the world's ending. But stocks are trading like there is an imminent issue with demand. And I just don't think that's what we've heard here today." 00:01:14 (Gavin Baker)

"...we're in a voting machine phase of the market to quote Warren Buffett. But ultimately, the market is a weighing machine." 00:04:28 (Gavin Baker)

"...the market's been very focused on second derivatives. And the thought is once the growth rate peaks, it's all over. And that might, that may be true. But it may also be that this is like the railroad build out and it goes for 60 years." 00:04:58 (Gavin Baker)

Supplier Datapoints Confirm Acceleration

Baker cites Astera Labs (CEO Jitendra Mohan) and SpaceX/Cerebras commentary as evidence that demand remains strong.

"...the CEO of Astera. Since they reported earnings two weeks ago, business has accelerated. Things have gotten better. SpaceX, demand for compute, still very strong." 00:00:47 (Gavin Baker)

Jensen Huang Is Irreplaceable: The Case for Patience at Nvidia's Board

Baker and Brad Gerstner sent a letter to Nvidia's board urging a slow, deliberate approach and, if the facts clear Jensen, considering bringing him back as CEO.

"We sent a letter to the NVIDIA board asking that they take a very slow, measured, deliberate approach... There is no need to rush to hire another CEO." 00:06:07 (Gavin Baker)

"I've been doing this 27 years and I met one Jensen, and I met one Elon." 00:06:33 (Gavin Baker)

"You do not lightly let someone like this walk away. And Jensen, there is nothing stopping him... he could go be the CEO of a competitor." 00:07:03 (Gavin Baker)

Jensen's China Policy Was Pro-American Security, and Its Reversal Carries Risk

Baker argues Jensen's "ship American chips" approach kept China's Huawei ecosystem starved of use, and the new policy forces Chinese labs onto Huawei hardware, which will improve it fast.

"Now that Huawei hardware is going to be used by every Chinese AI lab, I think it might get better really fast." 00:09:17 (Gavin Baker)

"...to make the hardware better, it needs to be used." 00:09:17 (Gavin Baker)

"...the only thing I've ever heard him say about Chinese hardware is the absolute importance of making sure that American AI hardware dominates. And we do not let an alternative Chinese AI hardware ecosystem develop because that AI ecosystem could be given to our real adversaries, Iran, North Korea, Russia." 00:08:00 (Gavin Baker)

Jensen's Asian Supply Chain Relationships Are a Moat That Can't Be Replicated

"...the relationships that Jensen has with the Asian supply chain are unparalleled... He's never had a contract with Taiwan Semi. It's been reported that Morris Chang actually long ago asked him to be the CEO of Taiwan Semi." 00:13:38 (Gavin Baker)

"...it's going to be hard to replicate somebody who can go to Taiwan, speak the language, the native dialect, have a handshake and walk out of there with more volume than anyone else at lower prices." 00:14:04 (Gavin Baker)

American Open Source as a Strategic and Cultural Priority

Baker argues a vibrant American open-source ecosystem preserves value diversity, and Jensen was its biggest proponent.

"I do not want to live in a world dominated by one, two or three frontier AIs. Because these AI systems are imbued with value systems." 00:11:24 (Gavin Baker)

"Jensen was the biggest proponent of American open source. And I think America needs that." 00:12:47 (Gavin Baker)

SpaceX: Reusability Unlocks Orbital Compute, Direct-to-Cell, and a Growing AI Business

"...the rapid reusability of Starship, it means we're going to see orbital compute. Brad earlier today said they're going to put up the first orbital compute satellites in 2027." 00:17:18 (Gavin Baker)

"...I think the size of SpaceX's first-party AI business will probably... it's getting pretty close to $10 billion. That's a very significant amount of very high margin revenue that is growing really, really, really fast." 00:20:03 (Gavin Baker)

2. Contrarian Perspectives

Wall Street's AI ROI Calculation Is Wrong Because It Counts Only Frontier-Lab Revenue

Consensus holds that trillions in capex versus under $200B in visible revenue signals a bubble. Baker's counter is that open-source and enterprise deployments generate real returns that never appear in lab revenue figures.

"...over 6,000 discrete enterprises who are buying GPU servers and associated equipment from Dell and running open source AI on those servers and getting a return. And it does not show up anywhere in that calculation that the market is doing." 00:03:56 (Gavin Baker)

A Peaking Growth Rate May Not Mean the End: Think Railroads, Not a Bubble

Investors treat deceleration of the growth rate as the top. Baker suggests a multi-decade build-out and that even single-digit growth leaves Nvidia cheap.

"...it may also be that this is like the railroad build out and it goes for 60 years." 00:04:58 (Gavin Baker)

Restricting Nvidia Sales to China May Backfire by Accelerating Huawei

The conventional hawkish view is to deny China the best chips. Baker, echoing Jensen, says using inferior chips in China was actually slowing Huawei, and forced adoption will improve it.

"Before the Chinese government was trying to encourage Huawei, ironically restricting the imports of GPUs... That was slowing down Huawei because to make the hardware better, it needs to be used." 00:09:17 (Gavin Baker)

Jensen Deserves the Benefit of the Doubt, and the Board Should Not Rush a Successor

Against the instinct to move quickly on a CEO under national-security scrutiny, Baker argues for due process and possible reinstatement, noting widespread support from stakeholders.

"...a truly vast number of people have reached out to me. NVIDIA employees, shareholders, partners saying they want to back his CEO." 00:15:12 (Gavin Baker)

Elon Has Done More for the Climate Than Activists Combined

"...if you add up all the climate activists, they've effectively, I think, been negative for the climate because they annoy people, at least here in America. Elon's done more for the climate than all the activists combined." 00:17:18 (Gavin Baker)

3. Companies Identified

Nvidia

World's most valuable company; AI GPU leader. Why mentioned: Subject of the leadership crisis and Baker's call for patience; considered a "cheap stock" even at modest growth.

"Let's say they grow 8% a year, 5% a year, 10% a year. It's a cheap stock." 00:05:22 (Gavin Baker)

SpaceX

Launch, Starlink, and AI company (Grok/xAI-related first-party AI business). Why mentioned: Starship catch, orbital compute, Starlink direct-to-cell, and a first-party AI business approaching $10B.

"SpaceX is the most important national defense contractor for America." 00:16:49 (Gavin Baker) "...America controls space because of SpaceX." 00:16:49 (Gavin Baker) "They're operating 11,000 satellites today." 00:18:35 (Gavin Baker)

Fireworks AI

Platform that serves and customizes open-source models for enterprise customers. Why mentioned: Growth proof point for the open-source demand layer; Baker calls it the best of its type.

"...they're the Atreides portfolio company. So I think they're the best one doing this." 00:03:01 (Gavin Baker) "...her top customer, it is 13X in the first nine or 10 months of this year." 00:02:35 (Gavin Baker)

Dell Technologies

Enterprise hardware provider selling "AI factories." Why mentioned: 3,200 AI factories sold in 2024–25, then about 3,300 in nine months.

"...they sell enterprises, what they call an AI factory." 00:03:29 (Gavin Baker)

Astera Labs

Connectivity semiconductor company (CEO Jitendra Mohan). Why mentioned: Business accelerated after earnings two weeks prior.

"Since they reported earnings two weeks ago, business has accelerated. Things have gotten better." 00:00:47 (Gavin Baker)

Cerebras

AI chip company (CEO Andrew Feldman). Why mentioned: Encouraging tone on AI demand.

"...the tone from Cerebris on, you know, the state of AI demand was encouraging." 00:00:47 (Gavin Baker)

Anthropic

Frontier AI lab. Why mentioned: Cited at roughly $80 billion revenue in the market's capex-return math.

"...80 billion in Anthropic revenue, plus the OpenAI revenue." 00:02:09 (Gavin Baker)

OpenAI

Frontier AI lab. Why mentioned: Reported to be around $40B in revenue in the same calculation.

"...it seems like that's been reported to be around 40." 00:02:09 (Gavin Baker)

Grok / GrokBot (xAI)

Elon Musk's AI assistant. Why mentioned: Baker's most valuable personal AI tool, now rivaling his time on X.

"I now spend more, almost more time in GrokBot than I do on X, which I thought would never happen." 00:19:36 (Gavin Baker)

Taiwan Semiconductor (TSMC)

Leading chip foundry. Why mentioned: Jensen's handshake relationship and Morris Chang's reported CEO offer.

"He's never had a contract with Taiwan Semi." 00:13:38 (Gavin Baker)

Huawei

Chinese hardware/AI chip maker. Why mentioned: Risk that forced domestic adoption improves its hardware quickly.

"...Huawei hardware is going to be used by every Chinese AI lab, I think it might get better really fast." 00:09:17 (Gavin Baker)

Starlink

SpaceX satellite internet. Why mentioned: Direct-to-cell and humanitarian impact.

"...they are tangibly saving lives and making lives of the poorest people in the world better by connecting them to the modern world." 00:18:05 (Gavin Baker)

Microsoft

Where Baker first met Nvidia CFO Colette Kress.

"I think I first met Colette when she still worked at Microsoft." 00:06:07 (Gavin Baker)

Gemini (Google)

Mentioned as part of the long tail of AI revenue.

"...Grok and Muse and Gemini and all these other things add up to something." 00:02:09 (Gavin Baker)

4. People Identified

Jensen Huang

CEO and co-founder of Nvidia. Why mentioned: Baker's central subject; a "singular" CEO whose return he advocates.

"I have known him for 27 years at this point... He is a friend." 00:07:32 (Gavin Baker) "I do not know a single CEO who is more patriotic about America than Jensen." 00:09:46 (Gavin Baker) "In Taiwan, Jensen, like, he is a living deity." 00:14:04 (Gavin Baker)

Elon Musk

Founder of SpaceX and xAI. Why mentioned: One of the two singular figures Baker has met in 27 years.

"I've been doing this 27 years and I met one Jensen, and I met one Elon." 00:06:33 (Gavin Baker)

Colette Kress

Nvidia CFO. Why mentioned: Praised as the stabilizing force during a CEO transition.

"Colette is a superb CFO, okay? She's been at NVIDIA a long time." 00:06:07 (Gavin Baker)

Brad Gerstner

Investor; co-signer of the letter to Nvidia's board. Why mentioned: Partnered on the letter; said first orbital compute satellites go up in 2027.

"...my friend Brad Gershner and I, we... sent a letter to the NVIDIA board." 00:06:07 (Gavin Baker)

Michael Dell

Founder and CEO of Dell Technologies. Why mentioned: Provided the AI factory sales data.

"...Michael Dell said something really important today." 00:03:29 (Gavin Baker)

Lin Qiao

Founder and CEO of Fireworks AI. Why mentioned: Growth of 13x at top customer; open-source specialization.

"...we heard from Lynn today. And she has a business that is growing really fast." 00:02:35 (Gavin Baker)

Jitendra Mohan

CEO of Astera Labs. Why mentioned: Reported accelerating business since earnings.

"...Jitendra, the CEO of Astera." 00:00:47 (Gavin Baker)

Andrew Feldman

CEO of Cerebras. Why mentioned: Encouraging demand tone.

"...I had a great conversation with Michael Dell, with Jitendra, with Andrew Feldman." 00:00:34 (Gavin Baker)

Morris Chang

Founder of TSMC. Why mentioned: Reportedly asked Jensen to become CEO of TSMC.

"It's been reported that Morris Chang actually long ago asked him to be the CEO of Taiwan Semi." 00:13:38 (Gavin Baker)

Walter Isaacson

Elon Musk's biographer. Why mentioned: Recounted SpaceX engineers thinking the Starship catch was impossible.

"...they thought it was impossible. And we've done it." 00:16:21 (Gavin Baker)

Warren Buffett

Investor; source of the voting/weighing machine quote.

"...to quote Warren Buffett. But ultimately, the market is a weighing machine." 00:04:28 (Gavin Baker)

Rick Pitino

Basketball coach; used as an analogy for succession after a legend leaves.

"...Larry Bird's not walking through that door anytime soon." 00:07:03 (Gavin Baker)

Brian Armstrong

Coinbase CEO; prior guest at the event.

"...we just had Brian Armstrong." 00:00:14 (Molly O'Shea)

Becky (Baker's wife)

"...me dressed up as Jedi for Halloween with Becky wearing Yoda ears." 00:21:53 (Gavin Baker)

5. Operating Insights

Triangulate Narrative Against Ground-Level Data Points

Baker's method is to compare the market's implied story against firsthand signals from operators in the same week, then act on the discontinuity.

"...it does feel like there's a little bit of a discontinuity between really strong fundamentals in the present and on the ground. And what stocks are discounting." 00:04:28 (Gavin Baker)

Look for Revenue That Never Hits the Headline Line Item

Dell's AI factory count and Fireworks' customer growth show how real value creation can sit outside the data points everyone cites. Build your own bottom-up model from customer counts, not top-line lab revenue.

"So these are over 6,000 discrete enterprises... And it does not show up anywhere in that calculation." 00:03:56 (Gavin Baker)

Trust Built on Handshakes Is an Operating Advantage

Jensen's refusal to rely on formal contracts with TSMC, backed by decades of relationship capital, yields volume and price advantages.

"...a handshake and walk out of there with more volume than anyone else at lower prices." 00:14:04 (Gavin Baker)

Use Long Tenure and Personal Relationships to Act Decisively in a Crisis

Baker leveraged a 27-year relationship to write a public letter with conviction, which attracted unsolicited backing from employees, shareholders, and partners.

"...a truly vast number of people have reached out to me. NVIDIA employees, shareholders, partners." 00:15:12 (Gavin Baker)

Use Your Own Product Intensively to Gauge Real Traction

Baker's heavy personal use of GrokBot informs his view that its adoption is underappreciated.

"I think GrokBot has a lot more traction than people realize." 00:20:03 (Gavin Baker)

6. Overlooked Insights

Usage Improves Hardware: Export Controls May Have Been Handing Huawei a Training Handicap

Mentioned in passing, this reframes China chip policy: the Chinese government restricted GPU imports to help Huawei, yet the better American chips were what labs preferred, thereby slowing domestic hardware iteration. Removing the "choice" flips the feedback loop.

"...to make the hardware better, it needs to be used. Now that Huawei hardware is going to be used by every Chinese AI lab, I think it might get better really fast." 00:09:17 (Gavin Baker)

Orbital Compute Is Gated by Starship Reusability, Not by Cooling or Design

Skeptics focus on thermal engineering, but SpaceX engineers consider that solved; the real constraint is launch economics. Catching the upper stage is the unlock, with first satellites targeted for 2027.

"...what they're actually stressed about is Starship and reusability. And so catching the upper stage of Starship, that's a huge step towards making Starlink direct-to-cell, orbital compute, all of that a reality." 00:19:06 (Gavin Baker)

Also notable is Baker's offhand detail that a Starlink V3 satellite already consumes "10 to 15 percent of the power of, like, an orbital compute satellite" 00:19:06, implying the power and thermal envelope is already within proven range.