Deconstructing the Investment Philosophy of Gavin Baker, Early Nvidia Investor: Long AI Infrastructure Bottlenecks, Short Overall Market Risk

Compiled & Edited by: Deep Tide TechFlow

Hosts: Ejaaz Ahamadeen (EJ), Josh Kale (Josh)

Original Title: What The Best AI Investors Are Buying Right Now

Podcast Source: Limitless Podcast

Release Date: May 28, 2026

Editor's Introduction

This podcast episode primarily discusses the investment philosophy of Gavin Baker, founder of Atreides Management and a long-term investor in Nvidia and Cerebras. His core judgment is that AI is not a bubble, but rather a super-cycle driven by electricity, wafers, and computing power. The real alpha isn't in large models or chatbots, but in the "picks and shovels" layers like GPU connectivity, memory, inference chips, advanced processes, and power supply.

Gavin Baker hedges against overall market pullbacks through QQQ puts while concentrating bets on AI physical bottleneck assets like Astera Labs, Unity, Micron, Nvidia, Cerebras, and Positron. He reframes the "AI bubble" debate from sentiment to supply-demand constraints, arguing that as long as TSMC, ASML, high-bandwidth memory, and the power grid cannot quickly become oversupplied, AI capital expenditure may not be a repeat of the 2000 dot-com bubble.

Key Quotes

AI Bubble or Super-Cycle

• "AI is not in a bubble; on the contrary, it's in a super-cycle."
• "The biggest returns are not in SaaS, not in chatbots like OpenAI or Anthropic, but in power, compute, and silicon manufacturing."
• "This is not the internet bubble, because the buyers are the world's smartest, most cash-rich companies. They aren't buying compute with debt leverage."
• "If the entire market cannot be oversupplied, it's hard for it to suddenly crash like a traditional bubble."

The Real Bottlenecks: Power, Wafers, Tokens

• "Gavin's theory is simple: look only at the bottlenecks in the AI infrastructure layer. Whoever increases performance per watt and lowers token cost has value."
• "AI labs are increasingly focused on one thing: how many tokens can be generated per watt of electricity."
• "Power and wafers are two brick walls, and also the two key constraints limiting how fast AI can accelerate."

The Shift from Pre-training to Inference and Post-training

• "After a model is pre-trained, it doesn't mean it's a genius for life; it still needs to absorb new information during the post-training stage."
• "Inference inherently requires massive computation, which is why inference chips and inference infrastructure will be the next phase's focus."
• "The cost or revenue opportunity from inference alone could be 5 to 10 times that of pre-training compute investment."

Vertical Small Models, On-Device Models & Sovereign Infrastructure

• "In the future, you might not interact with Claude every day; what you might truly need is a personalized AI agent trained on your own data."
• "The speed of infrastructure deployment itself is a moat. The iteration speed in the digital world is far faster than the construction speed of physical infrastructure."

"Whoever can compress physical deployments that take months or years into weeks can command high prices in the AI infrastructure space."

Gavin's Investment Approach: Long Bottlenecks, Short Overall Market Risk

• "He strongly believes AI winners will emerge, but that doesn't mean he's optimistic about the entire market; QQQ puts are his hedge against overall downside risk."
• "TSMC actually limits how fast a bubble can accelerate; as long as chip capacity cannot expand instantly, capital expenditure is less likely to spiral out of control."
• "Gavin is like an older, steadier, more cycle-proven Leopold: the former's success is measured in decades, while the latter's is currently measured more in quarters."

Assets Worth Betting on in the AI Super-Cycle

EJ: Gavin Baker is an exceptionally prolific AI investor who is almost unknown to the general public. Over the past 20 years, he started investing in AI companies that later became household names long before they hit the mainstream. He made early bets on Nvidia (AI GPU and accelerated computing core supplier) and Cerebras (AI chip company), and has a very clear view: AI is not a bubble; on the contrary, it's a super-cycle.

He believes that by observing watts (electricity), wafers (silicon), and tokens (model generation and compute units)—the underlying infrastructure of AI—you can identify key bottlenecks and constraints. His conclusion is simple: the biggest returns in AI come from power, energy, and silicon manufacturing, and have little to do with SaaS or chatbots like Anthropic and OpenAI. The entire industry eventually funnels downstream to semiconductors—the "picks and shovels" assets underpinning the AI industry.

While many claim the AI industry is already a bubble, he sees it as a generational buying opportunity, especially in AI infrastructure. He expresses this view through his fund, which has about $4.1 billion in assets under management.

If you listen to him talk about these constraints, especially AI infrastructure, the theory sounds familiar. We've talked about an investor named Leopold Aschenbrenner on this show before, who has also positioned around similar themes. The difference is that Leopold has been at it for about 3 years, while Gavin has been doing it for over 20.

Leopold's AUM is roughly triple Gavin's, but as our producer Luke aptly noted: You might beat Warren Buffett for a year, but can you beat him for decades? Gavin Baker's track record suggests he might have a different perspective on this investment thesis.

For those unfamiliar with Gavin Baker, he's the founder of Atreides Management and has been investing in Nvidia for the past 20 years. If you held Nvidia for 20 years and are still working, that in itself is incredible because it should have generated phenomenal returns.

His recent wins include Cerebras and Astera Labs (AI datacenter connectivity chip company). Cerebras is an AI chip company; the show mentions its IPO valuation was shockingly high. There are also some lesser-known companies we'll explore in this episode, following his portfolio and his views on where AI investment opportunities truly lie.

So the question becomes: What exactly does he invest in, and why? Looking at Atreides Management's recent 13F filing, the fund has roughly $4 billion AUM. Breaking down its largest holdings reveals companies all pointing to the AI development bottlenecks Gavin frequently mentions.

He holds significant positions in companies that aren't glamorous and many people haven't even heard of. For example, Astera Labs accounts for nearly 9-10% of the fund. You can think of Astera Labs as the connective layer between GPUs. Imagine a datacenter as a system: GPUs are the engines, responsible for model pre-training, post-training, and inference. But for GPUs to work, they need to transfer massive amounts of data among themselves and access memory chips where data is stored.

To do this, you need a "pipeline system." I'm keeping this high-level because I don't pretend to understand all the low-level details. Astera Labs solves precisely this problem. When AI clusters scale to hundreds of thousands of chips, the bottleneck is no longer just the GPU itself, but the data transfer windows—getting the right data to the right place at the right time. Astera Labs builds this pipeline system.

I hadn't heard of Astera Labs before researching for this episode either. But I recall Cerebras was a similar case. Gavin talked about Cerebras about six months ago, and considering AI's timescale, six months is a long time. It later IPOed; the show mentions a valuation of around $60 billion, and it's up 40% since its IPO. This suggests Astera Labs could be another significant name in a similar trend.

Josh: Cerebras was a very early investment for him. He got into Cerebras very early in the company's life cycle, meaning he's been betting on this thesis for years. There are several other companies he's been invested in for a long time, with the flagship being Nvidia, of course.

Being involved with Nvidia for over 20 years and maintaining conviction throughout is remarkable. I recently listened to two podcasts featuring Gavin. When discussing his Nvidia position, he clearly expressed a belief that Nvidia can maintain its current margins and demand. This implies he thinks Nvidia has a path to approaching a $10 trillion market cap; currently it's only about halfway there.

Another notable name is Micron. We covered the AI investment stack and these companies' positions in it in the last episode; I strongly recommend watching that. Micron is one of the largest memory makers. The show mentions a staggering number: a year ago its market cap was under $100 billion, and at recording time it had surpassed $1 trillion—a 10x in a year. This shows how important the memory problem is.

There are also less obvious but interesting companies. EJ, I particularly want to mention one: Unity Software. Those familiar with gaming know Unity; it's a game engine used to build many popular games.

Why would an AI investor invest in Unity, this "video game thing"? The answer lies in 3D game engines. Unity is a world model builder; it has deep understanding of physics, how the world operates, materials, and lighting. When AI companies build AGI and humanoid robots, a crucial component is simulating virtual environments and virtual datasets for robot training. Unity happens to be one of the strongest tools for this. So, as a world model maxi, you should appreciate this example: a company known for game engines has a clear path to becoming a significant player in the AI world.

Gavin's Investment Thesis and Strategy

EJ: The theory behind world models is simple: current AI models or LLMs primarily understand the world through text and books, like a student sitting in a library, but lack real-world experience. World models aim to unlock this: put a game character into a simulated environment to understand how physical reality works. What happens if I drop my phone or kick a ball? What are the next steps? What should you do? World models address this.

Currently, there aren't many players capable of doing this at scale. The leader might be Google with projects like Genie 3. The show also mentions Google's recent release of Gemini Omni, but these models haven't truly had their "ChatGPT moment" yet.

What I like about Gavin is that his portfolio resembles a barbell strategy. On one end, it's very traditional—people need GPUs, need memory, so he invests in the biggest players like Micron and Nvidia. On the other end, it's very forward-looking—he thinks the puck is going there, so he invests in Cerebras because he believes inference will be crucial; he invests in Unity because he believes world models will be the future way to train robots and next-gen LLMs.

His portfolio also includes Positron, which makes inference chips. If that sounds similar to Cerebras, yes, both focus on inference. Gavin recently emphasized a trend in interviews: the AI model infrastructure stack, especially the training stack, is shifting from pre-training to placing greater emphasis on post-training.

If you're in the AI space, you know this shift is already happening. Gavin is very focused on this. A model still needs to understand new information, new data, and update itself. Just because it completed pre-training on a dataset doesn't mean it's a genius for life. It needs to learn new information, which happens in the post-training layer, requiring significant compute.

Secondly, if you need an AI model to truly reason about a problem—like how we, upon receiving new information, might think: Does this angle hold up? Is there another theory that explains it?—that's reasoning. Reasoning also requires massive computation. Current estimates suggest the cost or revenue opportunity from inference alone could be 5 to 10 times that of pre-training compute investment.

So, a major pivot is underway within AI labs and chip makers. You've already seen Nvidia launch many inference-focused GPUs to support agentic applications. Gavin is also expressing his bet on inference through a series of investments.

One final point I find interesting is Gavin's take on China. The narrative in the AI race has always been China vs. the US. China has a unique configuration with relatively abundant energy and the ability to expand chip manufacturing. The US currently struggles in this area, which is why many steps are outsourced to Taiwan's TSMC.

His explanation is that China has a unique opportunity to create an AI infrastructure or chips very different from the US, as they will be heavily focused on inference. You could say Gavin, through his US investments, is leading the bet on the buildout of US inference infrastructure. I think this could be a massive opportunity in the future.

Josh: It's worth noting that this bet isn't all upside. He also holds a significant position in QQQ puts. QQQ is an ETF tracking the Nasdaq 100—a basket of stocks and the second-largest ETF by trading volume in the US. It has performed exceptionally well: up 55% in 2023, 25% in 2024, 20% in 2025, and 17% so far in 2026.

In other words, QQQ as an index fund has performed very well; it's easy to buy, being a basket of the top 100 stocks. Gavin is hedging against it. He's not saying AI won't win; he's saying he wants to invest in the critical manufacturers solving bottlenecks, but he doesn't appear overly optimistic about overall market sentiment. QQQ puts provide downside protection: if the overall market crashes adversely, even if AI wins long-term, he has this hedge.

Four Categories Worth Investing In

Josh: We can break down the investment bottlenecks he deems most important into a few categories. The first is verticalized small language models. Ordinary LLMs, like Claude and ChatGPT chatbots, are generalized LLMs. They have broad understanding of the world and can answer specific questions. But training a model around a specific vertical domain or problem is another matter.

These specific problems often exist within enterprises, especially those deeply entrenched in a particular problem or companies that have carved out a niche in a specific segment. Verticalized SLMs address precisely this: they are frontier models but highly optimized to run efficiently on specific enterprise data or locally on devices.

We've discussed on-device or locally run models before. The reason is that your phone or other devices hold a lot of highly personal data you might not want to hand over, and companies might not be able to access it—like medical records, financial details. I saw OpenAI release a financial AI agent that could access your bank account but couldn't actually act on your behalf because it contains personally identifiable information like social security numbers, bank details, etc.

Local models or SLMs can solve these problems. Gavin largely bets they will become important in the future. There's one company he is very bullish on: Apple. While he might not have expressed explicit investment interest, he believes Apple will be one of the primary device makers enabling local models to run on devices.

If the future unfolds this way, we might no longer think of Claude as the model you must interact with daily. You might need a personalized AI agent trained on your own data, which is what an SLM could eventually become. A general version could run on your phone, while many enterprises would run highly optimized, specialized models trained on their proprietary data to better sell or market products.

EJ: Apple is in such a great position for this. I'm really looking forward to WWDC; it's coming up soon.

Josh: Yes.

EJ: Apple's developer conference is only weeks away; they'll announce new AI software and how it integrates with hardware. This will be very important, and we'll continue to cover it. I'm excited to discuss it.

Josh: The second pillar is sovereign infrastructure. We often say bits move much faster than atoms. This is evident in AI infrastructure: model quality is improving almost exponentially; the intelligence generated per watt, per token, is only going up.

But physical deployment speed isn't increasing at nearly the same rate, and that in itself is a moat. Hardware is incredibly complex, with transistor precision nearing atomic scales; deploying at scale in a world where existing infrastructure is already under strain isn't easy. With EV acceleration putting more pressure on grids, many places are near capacity. Now AI brings an energy problem and a chip problem.

Gavin strongly bets on the fact that infrastructure is hard; building takes many days, months, even years. He's betting on those who can compress this cycle into weeks. So, the speed of physical deployment itself is a moat. He's narrowing his focus to find companies that can deploy quickly.

The first example that comes to mind is SpaceX and the speed at which they built Colossus and leased it to Anthropic, potentially to other companies in the future. This infrastructure pillar is one of Gavin's key focuses.

Looking at Leopold's portfolio, this is also a core part. The reality is: building stuff is very hard, and those who can build it can charge a premium. The show mentions that SpaceX's largest revenue stream now is renting out datacenters, not rockets. This shows how important this pillar is.

EJ: He cares about speed, but also about cost. He repeatedly mentions one metric: performance per watt. What he's really saying is that AI labs are increasingly focused on how many tokens can be generated per watt.

If you consider that just five companies this year are spending tens or even hundreds of billions of dollars on GPUs, compute, and the power to run them, you definitely want a high bang for your buck. Especially as hyperscalers expand to this scale, cost is a core issue.

Consider a hypothetical: It costs me 2 cents to ask Claude a question and get an answer; it costs $1 to ask ChatGPT. Even if Claude is only 95% as smart as ChatGPT, I'd probably use Claude. Because I can ask more questions and eventually get the answer at a lower cost.

So the cost of accessing this intelligence is crucial. Just this week, Microsoft and Uber announced they are actually reducing their use of Claude Code because their annual budgets were roughly exhausted in about four months.

You can see this in Gavin's portfolio: Cerebras, Positron, Astera Labs. He identifies very niche infrastructure bottlenecks and makes a simple bet: if this company solves this bottleneck, achieves a certain performance per watt, lowers token cost to a certain level, then AI labs will buy more GPUs, more products, more of these things.

So his theory is actually simple, even if the specific technology is complex: I only look at bottlenecks in the AI infrastructure layer. If I can find a company that increases performance per watt and makes tokens cheaper, I bet it will be valuable in the future, either through IPO or a high-priced acquisition.

Josh: For this part, if someone wants to replicate Gavin's trades, a few names to know: Astera Labs, Cerebras, SiFive, and Positron. These four companies are key in this segment.

The fourth and final direction is the combination of energy and space. As we said earlier, the terrestrial grid largely limits energy supply, and building new energy is very hard. The show mentions a statistic: about 40% of new datacenters face strong opposition—people lobby, protest, don't want these datacenters built.

There are two types of solutions. One is creating out-of-the-box energy, i.e., portable energy. You can bring a datacenter and power it with a small energy unit. Blue Marble, which Leopold is very bullish on, falls into this category.

The other is orbital compute, which Gavin is very focused on now. The biggest, most central company in this space is, of course, SpaceX. It's the only company capable of becoming the highway to space, launching payloads into orbit, sending racks and datacenters to low-Earth orbit, and generating enough intelligence and power to transmit back.

I think SpaceX's significance goes beyond SpaceX itself. I'm a bit surprised Gavin's portfolio doesn't have more space stocks, given he sees it as a huge industry. Maybe the reality is it's still too early, and SpaceX is the linchpin unlocking this industry.

Next, watch closely for Starship V3 launches. We just saw a Starship launch last week; it performed well. If Starship doesn't truly operate, there's no space energy, no racks to orbit. It's a necessary condition because the payloads needing launch are massive. So SpaceX is definitely a must-watch company, though many second-order companies will be affected.

Why This Isn't Another Internet Bubble

Josh: The next question everyone will ask is: Why isn't this just another dot-com bubble? Gavin has been asked this many times and gives a very strong answer, which I largely believe; his argument is convincing.

His logic roughly goes: The 2000 internet bubble was debt-fueled. Many people borrowed huge sums to invest in unproven theories and products no one really used or cared about.

Comparing that to what Gavin calls the AI super-cycle, just OpenAI and Anthropic alone are projected to reach $200 billion ARR this year. And this isn't made-up money; it's money already contracted, with a large portion—the show says 40% to 60%—prepaid by enterprise and retail customers. So money is actually flowing.

Looking at GPU compute buyers—not the model labs, but who's buying from Nvidia—Google, Microsoft, Amazon, and Meta are paying from their own cash reserves, not borrowing. Amazon just reached the end of its free cash flow; if they start borrowing, we can worry. But the key point now is they aren't leveraging up.

Moreover, these are among the world's top five and, in a sense, smartest companies, given their market caps, scale, and stature. Compared to the internet bubble, where countless unknown companies raised lots of money and burned it unreasonably, in this cycle, the world's smartest companies are spending unleveraged cash.

The quarterly reports we discussed on the show in recent weeks also show profits are being optimized around these moves, models are still improving, getting smarter. So Gavin's core argument is: This isn't the internet bubble because it's not driven by leveraged money, and the bottlenecks we're discussing are constrained by physical atoms.

Buying a bunch of memory chips and GPUs is one thing, but Nvidia can't oversell GPUs, Micron can't oversell AI memory chips because they don't have enough chip fabrication capacity. So his simple point is: If you can't oversupply the entire market, it's not a bubble. We are limited by not having enough picks and shovels to do this, and he's investing precisely in those.

Another great point: Gavin believes that if TSMC could supply them, Nvidia could have sold $2-3 trillion worth of GPUs this year and next. That is, TSMC is a key element in the bubble boundary.

The reason is, if TSMC could satisfy these companies' demand and provide that many chips, it would consume enormous capital. Looking at the charts currently, there isn't a huge disconnect between CapEx and operating cash; the cash generated by companies is still sufficient to fund the buildout.

But if TSMC told Nvidia tomorrow they could triple capacity overnight, Nvidia wouldn't refuse; it would start spending massive amounts to buy chips. Other companies would also be forced to borrow to buy these chips. Then the CapEx bubble would start to inflate, widening the gap with corporate operating cash flow.

But because there are supply constraints at every level—memory constraints, chip manufacturing constraints, energy constraints, and especially TSMC's constraints on advanced chips—we simply cannot ramp construction that fast. Therefore, TSMC is blocking the bubble's acceleration.

As long as TSMC's chip capacity remains limited, as long as Samsung and other chip makers don't surpass their market share, the growth is relatively sustainable. It looks fast, but there's still massive unmet demand because we simply can't build fast enough. As long as this dynamic holds, I think things are okay for now.

EJ: Another point: You can't assume demand remains static, because it won't. AI-related demand is growing exponentially, and the growth rate outpaces the production supply of these chips.

I can think of only two ways to disprove this thesis. First, someone miraculously replicates ASML, and suddenly there are many ASML competitors. For those unfamiliar, ASML makes machines worth about $400 million that TSMC and all major chip fabs need. The show says ASML has only one team in Norway making these, with very long cycles and order backlogs stretching about five years.

Second, we create a completely different type of LLM that doesn't require so many GPUs or so much memory. But we see no signs of that at all.

I saw news about SK Hynix today. It's the number one memory manufacturer and supplier for Nvidia GPUs, almost the top dog in AI memory. It's reportedly receiving $50-100 billion offers from Google and Microsoft wanting to lock in supply for the next three years, to pay for the equipment needed for its expansion.

This shows how hungry these big companies are for memory, and this is just one sub-sector within AI components. SK Hynix is essentially saying: I don't want to give you supply guarantees; I'll just raise prices. Its operating margin is about 70%, almost unbelievable in the semiconductor industry.

So Gavin going all-in makes sense. It doesn't look like a bubble; maybe the market reacts that way short-term. Opening our stock portfolios before recording today, almost everything was down, but that's more of a reactionary sentiment move. The directional goal here is: We will only need more GPUs, more semiconductor chips, and supply is insufficient, manufacturers are insufficient.

Gavin's Investment Portfolio

Josh: The conclusion is: Power and wafers. That's it. They are two brick walls, two limiting factors preventing us from accelerating too fast. As long as power and wafers remain valuable, in strong demand, and supply-constrained, the good times are still ahead.

If you want the TL;DR of Gavin's portfolio, I can read out his largest holdings. Again, this is not investment advice. This is what Gavin holds, not what we hold. I don't know if these stocks will go up, down, or sideways.

His largest position is somewhat counterintuitive: a QQQ put position. Overall, he's bearish on the market, which is very noteworthy. Second largest is Astera Labs, about a 7.4% position, ticker ALAB. Third is Unity, the 3D software company.

There are many more: Ciena, Micron, Nvidia, Amazon, Lumentum, Alphabet, Coherent, Roblox, EchoStar, Twilio, Wayfair. This guy invests in everything.

If you're interested, you can look up his 13F; we'll put a link in the description. But that's Gavin's view: the bottlenecks are power and wafers. As long as these constraints persist, it's basically one-way up. EJ, how are you absorbing this info? What would you do with it?

EJ: Since Leopold's 13F came out, the market has been volatile. Recording this episode, I increasingly realize Gavin is like an older, wiser Leopold. He's been in this industry a long time. Maybe he doesn't have $13 billion AUM, but I feel he'll still be here in 10 years.

If you're listening and thinking, I don't want to track every minute, hour, day of AI progress; I just want to put money somewhere and watch it grow over the coming months or years. Then Gavin's portfolio might be very informative. Of course, this is not investment advice.

He's adopting a more cautious, longer-term, and future-oriented approach. If his trend calls materialize, like his early bets on Nvidia and Cerebras did, there could be exponential returns in the coming years. But it all hinges on his core view: We are not in a bubble.

I'm curious if listeners agree. Obviously, most people won't be as technical or deep into the weeds as Gavin. But after listening to this episode, do you think we're in a bubble or not? What are the reasons for and against? Is there something we missed? Josh, before we end, do you think it's a bubble now?

Josh: I think we are certainly in a bubble. The question is, what stage of the bubble are we in? That's debatable. Right now, it looks more like an early stage, so hopefully it remains that way. According to Gavin, as long as TSMC continues to limit chip capacity, we're okay.

That's the overall outlook. We've talked about Leopold, whose success is currently measured in quarters; now we talk about Gavin, whose success is measured in decades. Many people's own answers might fall somewhere in between.

If you liked this episode, don't forget to share it with a friend. Also, tell us which type of asset you're most bullish on. Maybe not a specific theory, but a ticker symbol worth our attention. I find this exciting because everything is moving fast, whether up or down, there's a lot of volatility, and it's very engaging. See you tomorrow, good morning.