David George | The a16z Show Summary

1. Key Themes

No Dark GPUs: Why This Cycle Is Fundamentally Different From 2000

The central argument against the AI bubble thesis rests on a direct comparison to the telecom bubble, where infrastructure was built with no utilization. Gavin Baker draws a sharp contrast using the "dark fiber" analogy as his primary evidence.

"At the peak of the bubble, 97% of the fiber that had been laid in America was dark. Contrast that with today. There are no dark GPUs. All you have to do is read any technical paper and one of the biggest problems in a training run is that GPUs are melting." [00:04:45]

Baker further anchors this with financial performance data from the biggest spenders:

"It is return on invested capital of the biggest spenders on GPUs who are all public. And those companies, since they ramped up CapEx, have seen, call it a 10-point increase in their ROICs. So, thus far, the ROI on all the spending has been really positive." [00:05:14]

Valuations Are Not Stretched Like 2000

Beyond utilization, Baker points to valuation multiples as a second pillar of the non-bubble argument.

"I think Cisco peaked at 150 or 180 times trailing earnings. NVIDIA's at more like 40 times. So, valuations are very different." [00:03:53]

David George adds the balance sheet dimension:

"They collectively generate like $300 billion of free cash flow a year... and they have $500 billion of cash on the balance sheet. So whenever people are like, oh my God, it's a bubble, is it going to pop? I'm like, I think it's kind of fine." [00:06:21]

Round-Tripping Is Real But Strategically Rational, Not Fraudulent

The podcast addresses a hot-button concern about NVIDIA investing in customers who then buy NVIDIA chips. Baker reframes the motivation entirely.

"NVIDIA's biggest competitor, it's not AMD, it's not Broadcom, it's certainly not Marvell, it's not Intel, it's Google. And more specifically, it is Google because Google owns the TPU chip. And this is by far, maybe perhaps today, the only alternative to NVIDIA for training and maybe the best inference alternative." [00:08:00]

"If Google is going to a lab like Anthropic and saying, I'm going to help you fundraise and give you chips for competitive reasons, it's very hard for NVIDIA not to respond." [00:09:20]

The Real Chip War: NVIDIA vs. Google TPU, Not NVIDIA vs. AMD

Baker argues the competitive framing in most coverage is wrong. The existential threat to NVIDIA is not AMD but Google's vertically integrated TPU ecosystem.

"I think it is really a fight between NVIDIA and the Google TPU... What Broadcom is saying to companies like Meta is, hey, we will build you a fabric that can theoretically compete with NVIDIA's fabric... And hey, we'll make you your version of TPU, which by the way, took Google three generations to get working." [00:23:20]

He also signals skepticism on custom ASICs more broadly:

"I personally believe most of those ASICs are going to fail... In the next three years. I think you'll see a bunch of high profile ASIC programs canceled, especially if Google starts selling TPUs externally." [00:24:19]

AI Is a Sustaining Innovation for Mag7, Not Just a Disruptor

Baker makes the non-obvious argument that unlike the internet, AI may reinforce existing incumbents rather than topple them.

"The raw ingredients of kind of data, the capital to buy compute and distribution, which is what you need — all of today's biggest tech companies have all of those in spades. So as long as they execute well, hire good people, and have a sound strategy, like I think you could see it be a sustaining innovation for a lot of members of the Mag7." [00:11:30]

However, he notes the stakes for underperformers:

"I do think it's existential. And if you don't execute, you know, IBM might be a good fate." [00:11:55]

SaaS Gross Margin Pressure Is a Feature, Not a Bug

Baker makes a contrarian case that declining gross margins in AI-enabled SaaS companies should be celebrated rather than feared, using the cloud transition as precedent.

"It is definitionally impossible, given what we just discussed, to succeed in AI without gross margin pressure. And I do not know why they have concerns because we have an existence proof that a software company can deal well with declining margins — Microsoft." [00:15:32]

"If you're an application SaaS company, like what I would just say is don't be scared and look at declining gross margins kind of as a mark of success rather than a badge of shame or something to be feared." [00:16:28]

David George reinforces this from the investor perspective:

"Whenever we have these discussions about companies, basically every company that comes to present to us is like, we're an AI company. And we always look at the gross margins and it's become like a badge of honor for them to actually have low gross margins. Kind of like, oh my God, people are actually using your AI stuff." [00:16:44]

Reasoning Models Have Unlocked a New User-Base Flywheel

Baker identifies a structural shift caused by reasoning models that changes the economics of frontier AI labs in a way not yet widely understood.

"Pre-reasoning, I often said, if you are a frontier model without access to unique valuable data and internet scale distribution, you're the fastest depreciating asset in history. I think reasoning really changed that because the way RL works during post-training, having a big user base now kind of unlocks that flywheel that was at the center of every great consumer internet company." [00:20:59]

Robotics: Tesla vs. China, and the Humanoid Debate Is Over

Baker offers a clean competitive framing for robotics and declares the humanoid vs. non-humanoid debate settled.

"It's going to be Tesla versus the Chinese in the same way it's Tesla versus the Chinese in cars." [00:29:48]

"I think that debate is over because humanoids can kind of learn from watching YouTube videos and then it's easier for a human being to put on a suit and show the robot how to do it." [00:30:13]

Business Models Are Shifting to Outcomes, Compressing Legacy Advertising Inefficiencies

Baker sketches a structural shift in how AI companies will monetize, with affiliate and outcome-based models replacing traditional SaaS and advertising.

"Humans were fundamentally paid based on outcomes. And a lot of AI will be augmenting humans, but probably also replacing some humans. And that will involve being paid for outcomes." [00:27:05]

He explains why Google never shifted to outcome-based advertising — and why AI will force the issue:

"Why did Google never start a marketplace? Because people overvalue systematically their ability once they've acquired a customer through Google to keep it as an organic customer. So they systematically overpay... That's why Google never went to outcomes or marketplace. Because advertising leads to the advertisers systematically overpaying. So that inefficiency will be squeezed out." [00:28:28]

2. Contrarian Perspectives

GPT-5 Has Nothing to Do With Scaling Laws — The Narrative Is Wrong

This is one of Baker's most pointed moments — a direct rebuttal of a widely circulated market narrative.

"All these people who say that GPT-5 is the end of scaling laws — GPT-5 is a smaller model. It was not designed to be better. It was designed to be more economical for OpenAI and Microsoft to run. Any reference to GPT-5 and scaling laws is crazy." [00:22:16]

AI Browsers From Non-Google Players May Be a Strategic Mistake

Counterintuitively, Baker argues that OpenAI and Anthropic launching browsers could be handing Google an opening rather than threatening it.

"I think the AI companies that have launched these AI browsers may come to regret it. Because there's something called Chrome that has, whatever it is, 5 billion users. And if you're Google... they could easily do this and probably do it even better. But they didn't want to be first. So now you have two AI native companies with their own browsers. Let them run for three to six months. Get a little head start. And then, wow, here we are. We had to do this." [00:19:44]

Google Is Already the Largest AI Company by Traffic — The Market Doesn't Know It

Baker makes a claim that directly contradicts the narrative that OpenAI dominates AI.

"I think they've taken 15 or 20 points of traffic share in the last two or three months and that's just traffic to Gemini. It does not include search overviews. I suspect on an actual traffic basis, Google is bigger than OpenAI, Anthropic, anyone today." [00:08:51]

Most Custom ASIC Programs Will Fail in the Next Three Years

Against the popular narrative that hyperscaler-commissioned ASICs will commoditize NVIDIA, Baker predicts most will be canceled.

"I personally believe most of those ASICs are going to fail, particularly... In the next three years. I think you'll see a bunch of high profile ASIC programs canceled, especially if Google starts selling TPUs externally." [00:24:19]

He points to the difficulty of replicating what Google has achieved:

"Which by the way, took Google three generations to get working." [00:24:19]

Karpathy's "AGI in 10 Years" Is Being Treated as Bearish — It's Actually Wildly Bullish

Baker flags how distorted the Overton window on AI timelines has become.

"Karpathy, you know, whatever, two days ago, is being painted as like a skeptic for saying AGI is 10 years away. Are you kidding? It's insane. 10 years? That's wild. Yeah, sign me up." [00:29:26]

3. Companies Identified

NVIDIA

The dominant GPU and AI systems company. Discussed as the central infrastructure winner and the standard against which all other chip competitors are measured.

"Jensen's one of the two best CEOs along with Elon I have ever known. And I think he's playing a strong hand really well." [00:10:11] — Gavin Baker

Google / DeepMind / Gemini

Described as arguably the leading AI company today by traffic, controlling the TPU — the only credible alternative to NVIDIA for training — and positioned to leverage Chrome's 5 billion users in the browser wars.

"I think you could argue that they're the leading AI company today. I think they've taken 15 or 20 points of traffic share in the last two or three months and that's just traffic to Gemini. It does not include search overviews." [00:08:24] — Gavin Baker

Cursor

Highlighted as the benchmark AI coding tool that legacy public coding companies have failed to compete with, and now building an insurmountable data moat.

"Cursor now, they have a trillion tokens. And there will be a point where they have enough coding tokens that it's tough to catch them." [00:17:49] — Gavin Baker

Anthropic

Discussed as a key lab that is effectively a Google/Amazon captive and may be acquiring tens of billions worth of TPUs, signaling deep infrastructure alignment.

"Anthropic is really going to run on TPUs and Traniums... it was just rumored Anthropic wants to buy tens of billions of TPUs." [00:08:51] and [00:24:47] — Gavin Baker

OpenAI

Referenced as the ChatGPT / Netscape Navigator analogy — the first mover in AI but not necessarily the ultimate winner, given how early the cycle is.

"If we're going to make an analogy and say that ChatGPT is to AI what Netscape Navigator was to the internet — at this point in the internet boom, Google had not been founded. Mark Zuckerberg was in middle school." [00:10:37] — Gavin Baker

Tesla / Optimus

Cited as the leading Western humanoid robotics competitor, with videos impressing roboticists and a clear competitive frame mirroring Tesla's EV success.

"It's going to be Tesla versus the Chinese in the same way it's Tesla versus the Chinese in cars... you can all watch the Optimus videos. Every roboticist I know is extremely impressed." [00:29:48] — Gavin Baker

Broadcom

Identified as NVIDIA's key enabler of a competitive alternative — building Ethernet fabric and custom ASICs for hyperscalers — but ultimately dependent on Google's TPU strategy.

"What Broadcom is saying to companies like Meta is, hey, we will build you a fabric that can theoretically compete with NVIDIA's fabric... it's really Google and the TPU versus NVIDIA, enabled by Broadcom for now. And Google can take the TPU away from Broadcom whenever they want." [00:23:49] and [00:25:16] — Gavin Baker

Amazon / Annapurna / Trainium

Praised for having perhaps the most talented silicon team at any hyperscaler, with Trainium 3 expected to be a meaningful improvement.

"Amazon, like that's a very talented team, arguably the most talented silicon team at any hyperscaler, the Annapurna team. Like I think the Trainium 3 will probably be a much better chip than the Trainium 2." [00:25:16] — Gavin Baker

Microsoft

Used as the canonical existence proof that a software company can successfully navigate a margin-compressive platform transition while rewarding shareholders.

"Microsoft, they transitioned from on-premise perpetual licenses with maintenance to a cloud model. And it was a pretty good stock for 10 years." [00:15:59] — Gavin Baker

Adobe

Briefly cited alongside Microsoft as examples of companies that navigated the cloud margin transition successfully.

"We have an existence proof that a software company can deal well with declining margins — Microsoft and Adobe." [00:15:59] — Gavin Baker

Figma

Highlighted as a positive example of a company that proactively communicated AI-driven gross margin compression to public markets and was rewarded for it.

"Figma, for example, when they went out, they are extremely high gross margin. And they're like, hey, we're going to pretty aggressively distribute our AI tools and our gross margins are going to go down. And investors asked a few clarifying questions and then they were like, oh, that actually would be a good thing." [00:18:21] — David George

Decagon

Mentioned as a portfolio company exemplifying the outcome-based business model shift in customer support AI.

"We're investors in Decagon, customer support. Like you can pretty easily see a business model that is priced on the resolution of a task because it's so measurable." [00:26:21] — David George

XAI / Grok

Identified as one of the four leading frontier labs and referenced as a personal AI of choice by Baker due to its personalization potential.

"My own AI will be a version of Grok because we're both XAI shareholders. It will be a version of Grok that knows me and it likes me." [00:27:34] — Gavin Baker

Level 3, Global Crossing, WorldCom

Referenced as cautionary tales from the 2000 bubble — the dark fiber builders who laid infrastructure that was never lit.

"I vividly remember companies like Level 3 or Global Crossing or WorldCom would come in and they say, we laid 200,000 miles of dark fiber this quarter." [00:04:15] — Gavin Baker

4. People Identified

Jensen Huang (NVIDIA CEO)

Praised as one of the two best CEOs Baker has ever known, credited with strategically evolving NVIDIA from a chip company to a systems and data center architecture company.

"Jensen's one of the two best CEOs along with Elon I have ever known. And I think he's playing a strong hand really well." [00:10:11] — Gavin Baker

Elon Musk

Named alongside Jensen Huang as one of the two best CEOs Baker has encountered. Also cited for a recent tweet about work becoming "optional" in an AI-abundant world.

"Elon tweeted today that work would become optional. You know, like instead of buying your vegetables at a supermarket, you can grow your own garden if you want. Now, who knows how long it takes us to get there. But that doesn't sound wildly implausible to me." [00:28:56] — Gavin Baker

Andrej Karpathy

Referenced for his framing of AGI timelines — his "10 years" estimate is being misread as bearish when Baker considers it extremely bullish. Also mentioned in the episode opener regarding agents potentially being "ghosts."

"Karpathy, you know, whatever, two days ago, is being painted as like a skeptic for saying AGI is 10 years away. Are you kidding? It's insane. 10 years? That's wild." [00:29:26] — Gavin Baker

Larry Page (Google Co-Founder)

Quoted for an internal statement that encapsulates the existential mentality driving hyperscaler AI spending.

"Larry Page apparently internally said, I'm happy to go bankrupt rather than lose this race. And I think that is the mentality for sure at Google and perhaps Meta." [00:07:09] — Gavin Baker

Richard Sutton

Author of "The Bitter Lesson," referenced as the intellectual foundation for why AI models will remain compute-intensive and why frontier lab gross margins will structurally lag SaaS.

"The nature of AI because of scaling laws, Richard Sutton's The Bitter Lesson — they're just more compute intensive. So their gross margins are structurally going to be lower." [00:13:10] — Gavin Baker

Mark Zuckerberg (Meta CEO)

Cited as a serious competitor in the AI race, with Meta described as taking an existential view of the technology.

"Mark Zuckerberg's trying hard. And we'll see. Yeah. A lot of smart people in there now." [00:21:26] — Gavin Baker

Dwarkesh Patel

Referenced in the context of an interview with Andrej Karpathy that sparked renewed debate about SaaS viability.

"With Andre's Dwarkesh interview he just did, it's like the market's reacting positively to it and it's like a whipsaw reaction." [00:14:11] — David George

5. Operating Insights

Embrace Margin Compression Publicly and Frame It as the Cloud Transition Analogy

Baker offers a specific operational and communications playbook for public SaaS companies trying to compete in AI. The hesitation to compress margins is holding companies back from taking the aggressive moves needed to stay relevant.

"If you're an application SaaS company, like what I would just say is don't be scared and look at declining gross margins kind of as a mark of success rather than a badge of shame or something to be feared... you can run your new AI products at breakeven and catch up to the leaders." [00:16:28] and [00:17:19] — Gavin Baker

The actionable implication: if you have a profitable legacy business, use it as a subsidy to run AI products at breakeven and compete aggressively, rather than protecting margins and ceding ground to pure-play AI natives.

Outcome-Based Pricing Is the Inevitable End State — Build Toward It Now

Baker and David George both point to customer support (Decagon) as the proof case that outcome-based pricing works and is the natural business model for AI. The broader implication is that any company still pricing on seats or time-based SaaS is leaving money on the table AND is more vulnerable to disruption.

"You know, going back to the consumer business model, everybody's talking about affiliate fees... there will probably be some sort of affiliate fee. And again, that's just being paid for an outcome and kind of closing that loop." [00:27:34] — Gavin Baker

The Coding Token Moat Is Already Being Built — Move Now or Lose

Baker signals that the window to compete with Cursor in coding is closing fast due to token accumulation creating a data flywheel that will be very hard to replicate.

"Cursor now, they have a trillion tokens. And there will be a point where they have enough coding tokens that it's tough to catch them. But I think today, if you're a public coding company and you said, I'm going to lean in, I'm going to run it breakeven, I have an existing business, I'm going to attach it to everything — hey, you have a chance." [00:17:49] — Gavin Baker

The operating insight: in AI, token accumulation from real usage is becoming a durable moat. Any company that touches a specific domain (legal, medical, coding, finance) should be aggressively accumulating domain-specific inference data now, before the window closes.

6. Overlooked Insights

The Chinese Open-Source Model Ecosystem Is a Strategic Lifeline for Any Lab Without a Top-4 Checkpoint

This was mentioned in a single sentence but carries enormous strategic weight. Baker points out that any lab trying to train the next frontier model is at a severe disadvantage if they don't have access to a leading checkpoint — UNLESS they can use Chinese open-source models as a bootstrap.

"In a strange way, the Chinese open source model ecosystem is a godsend to any American company that's trying to catch those four leading labs. Because the problem is, if you don't have Gemini 2.5 Pro, or a later checkpoint of Grok that we don't see, or a later GPT checkpoint, training the next model, you're at a big disadvantage." [00:21:56] — Gavin Baker

The non-obvious implication: the geopolitical framing around Chinese AI models (DeepSeek, etc.) as threats may be obscuring the fact that they are actually a competitive equalizer that benefits the entire American startup ecosystem trying to challenge the Mag7 labs. Any second-tier lab or well-funded startup should be thinking carefully about how to leverage open-source Chinese model releases as training bootstraps.

Google's Litigation Posture, Not Capability, Is What Kept the Browser Open — This Window Is Closing

Baker makes a throwaway observation that Google deliberately held back on launching an AI browser because of active antitrust litigation with the government — not because it couldn't execute. This means the moment that legal constraint lifts or shifts, Google has 5 billion Chrome users it can instantly redirect into an AI-native browsing experience, potentially making OpenAI's and Anthropic's browser launches moot overnight.

"They are very cautious. You know, they're currently in litigation with the government. And they could easily do this and probably do it even better. But they didn't want to be first. So now you have two AI native companies with their own browsers. Let them run for three to six months. Get a little head start. And then, wow, here we are. We had to do this." [00:19:44] — Gavin Baker

The investment implication: the AI browser opportunity is not a durable wedge against Google — it's a window that Google's legal situation has temporarily left open. Any company betting on browser as a long-term moat should stress-test that thesis against the scenario where Google's antitrust posture shifts and Chrome becomes an AI-native interface overnight.