Oct 7, 2024 · 1h 19m · news
Eiso Kant, CTO @Poolside: Raising $600M To Compete in the Race for AGI | E1211 · 20VC with Harry Stebbings
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this deep-dive interview, Poolside CTO and co-founder Eiso Kant discusses his company's mission to achieve Artificial General Intelligence (AGI) through a focused, deterministic approach to software development, alongside the structural, financial, and psychological realities of the global AI landscape.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Harry holds 15.1% of the talking time here. How this is scored →
Harry as informed peer 4.2 Guest teaching 5.9 Guest disagreement 2.0 Harry pushing back 3.5
speaking balance: gold is Harry, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 1:05:00 Rejection of China AI lag premise
Eiso flatly disagrees with Harry's question asking if China is two years behind, responding with an immediate 'No. No, they're not' before explaining their sophisticated open research strategy.
Hardest push from Harry ▶ 41:48 Challenging OpenAI $6B adequacyHarry directly challenges Eiso's logic regarding OpenAI's $6.6 billion round, insisting he cannot understand how that amount is remotely sufficient compared to Big Tech spending hundreds of billions.
Biggest teaching moment ▶ 19:40 Fab relationships and chip margins breakdownEiso provides an extensive breakdown of the hardware stack, explaining how Amazon deals directly with foundries while Google works through Broadcom, detailing the unit economics of custom silicon versus NVIDIA margins.
Harry holds his own ▶ 1:01:20 Chase Coleman Netscape value capture citationHarry demonstrates deep market domain knowledge by citing Chase Coleman's statistic that 99% of internet value was created long after Netscape, questioning whether foundation model builders will capture real economic surplus.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Harry as informed peer | Guest teaching | Guest disagreement | Harry pushing back | Why |
|---|---|---|---|---|---|---|
| What Is Poolside and its Focus on Coding? | 2 | 6 | 1 | 1 | Harry opens with a broad introductory request asking for context on Poolside. Eiso provides an extensive breakdown of model compression, token availability across code versus English text, and the need for intermediate execution reasoning data. | |
| Simulating Data: AlphaGo, Tesla Autopilot, and Code Execution | 3 | 7 | 2 | 2 | Harry uses a cinematic metaphor from The Social Network to ask how abstract thinking is captured. Eiso uses DeepMind's AlphaGo and Tesla's FSD to educate Harry on deterministic versus real-world non-deterministic data and RL execution feedback. | |
| The AI Triad: Algorithms, Data, and why Compute Matters | 4 | 6 | 1 | 2 | Harry frames the progression of AI around compute, data, and algorithms, asking where the main bottleneck lies. Eiso explains why compute underpins synthetic data generation and parameter scaling limits. | |
| Algorithmic Efficiency and the Synthetic Data 'Oracle of Truth' | 4 | 6 | 2 | 4 | Harry asks a sharp question about model collapse and whether synthetic data is a snake eating its tail. Eiso clarifies that an external oracle of truth like code execution feedback prevents model degradation. | |
| Scaling Laws and the Economics of Model Distillation | 3 | 6 | 1 | 2 | Harry asks about the trajectory of scaling laws. Eiso explains the economic dynamics of training ultra-large models and distilling them down into cost-effective smaller models for deployment. | |
| Model Price Wars, Custom Silicon, and the Hyperscaler Advantage | 4 | 8 | 2 | 3 | Harry inquires about model pricing trends over the next 12 to 24 months. Eiso delivers an in-depth breakdown of custom silicon across hyperscalers, fab relationships, chip margins, and price wars. | |
| A Historical Perspective on Bundling Human and Machine Intelligence | 3 | 5 | 2 | 2 | Harry asks if model distillation will remain necessary in five years. Eiso contextualizes the answer through historical shifts in technological connectivity and the bundling of human and machine intelligence. | |
| The Capability Gap, Economic Value, and the Data Moat | 5 | 6 | 3 | 5 | Harry challenges Poolside's position by bringing up GitHub's massive codebase advantage. Eiso clarifies the distinction between public code access and private code boundaries, outlining the four capabilities of frontier AI labs. | |
| The Compute Economy: Funding, GPU Clusters, and Physical Constraints | 4 | 7 | 3 | 6 | Harry directly asks if Poolside's $600M funding is enough. Eiso candidly answers no, detailing physical cluster interconnect limits and GPU network scaling physics. | |
| The $100 Billion Entry Price and the CapEx vs. OpEx Reality of AGI | 5 | 6 | 2 | 4 | Harry quotes Larry Ellison's $100 billion entry price claim. Eiso corrects the assumption by delineating between model creation CapEx and global inference OpEx buildouts. | |
| Why Data Centers Become Obsolete and the Physics of Training vs. Inference | 6 | 7 | 1 | 3 | Harry cites Sequoia investor David Cahn regarding data center obsolescence. Eiso explains why synchronous parameter updates during training require localized data centers while inference can be distributed. | |
| NVIDIA’s Historic Dominance and the Custom Silicon Competitors | 4 | 6 | 1 | 2 | Harry asks if NVIDIA maintains a permanent monopoly. Eiso recalls his 2016 experience building Sourced and contrasts NVIDIA against Google TPUs, Amazon Neuron, and AMD. | |
| The Blackwell Delay and the Economics of GPU Upgrades | 5 | 6 | 3 | 5 | Harry pushes back on OpenAI's $6 billion round relative to hyperscaler CapEx commitments. Eiso explains how non-capital vectors like data and research limit pure cash dominance. | |
| Proprietary Knowledge, Startups Consolidation, and Poolside's Standalone Strategy | 5 | 5 | 2 | 4 | Harry probes into startup acqui-hires and corporate investors. Eiso explains why Poolside intentionally avoided Big Tech hyperscaler equity while retaining NVIDIA as a technical partner. | |
| AI Equity Draft: Evaluating OpenAI, Anthropic, and xAI | 6 | 5 | 3 | 6 | Harry poses an equity draft scenario, asking Eiso to choose between buying OpenAI, Anthropic, or xAI. Eiso breaks down xAI's infrastructure speed and OpenAI's revenue, but gently declines to pick a single winner. | |
| AI vs. Crypto: The Centralization of Scarce Resources | 5 | 5 | 2 | 3 | Harry references Peter Thiel's thesis comparing crypto decentralization to AI centralization. Eiso draws on his 2008 digital currency background to discuss scarce AI inputs like talent and research. | |
| AI Tourists, Enterprise Budgets, and Commoditized Use Cases | 4 | 7 | 2 | 4 | Harry asks about corporate AI budgets and why Poolside maintains European bases. Eiso reveals an internal mapping of 3,300 global candidates that led them to tap under-serviced European talent pools. | |
| AGI is a High-Stakes Race | 4 | 5 | 3 | 4 | Harry brings up European work-life balance stereotypes. Eiso forcefully reframes AGI as an intense global race that requires total sacrifice, dismissing geographic work ethic tropes. | |
| Value Capture and Vertical Integration in AI | 7 | 5 | 3 | 6 | Harry quotes Chase Coleman on Netscape's initial value capture versus long-term app accumulation. Eiso counters using BYD's vertical integration model to justify model layer capture. | |
| China's Role in the AGI Race | 4 | 7 | 4 | 4 | Harry asks if China is two years behind in AI capabilities. Eiso flatly rejects the premise, explaining China's game-theoretic open research publication strategy. | |
| Quick Fire: Future AI Roadblocks | 4 | 5 | 2 | 3 | Harry runs a quick fire segment touching on past decisions, AI roadblocks, and dream board members. Eiso chooses Mark Zuckerberg for his long-term conviction against consensus. | |
| Quick Fire: Yuri Milner and Global Investing | 3 | 6 | 1 | 2 | Harry asks about investor Yuri Milner. Eiso explains Milner's scientific manifesto and his global approach to placing technology bets across international markets. | |
| Quick Fire: Boat Living and Shifting Priorities | 3 | 4 | 1 | 3 | Harry asks about Eiso's lifestyle living on a sailboat. Eiso describes shedding material possessions to maintain focus and freedom while building Poolside. | |
| Quick Fire: Poolside Premortem and Eiso's "Why" | 3 | 5 | 1 | 3 | Harry asks for Poolside's premortem and Eiso's core personal motivation. Eiso explains that falling behind in capabilities or GTM causes failure, while his internal drive comes from working on the hardest problems. |
Statements from this episode (48)
Eiso Kant: Current AI era sets foundational table like mobile and internet
“We're going to look back on this moment 10 years from now, just like we would look back to the moment of mobile, internet, and realize that that was the moment where the table got set.”
Kant: Poolside's $500M round is merely an entry ticket to AGI race
“You do not want to look back on that moment and not have given it everything you've got, because it's a race, and the latest five hundred million dollar round translates to us being able to be an entrant into the race.”
Kant: Full AGI is far off, but specific high-value capabilities will arrive first
“At some point we are going to be in a world where across all sets of capabilities that we have as human beings, Machine intelligence is going to be as capable and if not more capable than us and surpass us. Now, our point of view is, is that that world is stil…”
Kant: Internet has 3T usable code tokens vs 15T English text tokens
“Usable code for training is what we refer to as about three trillion tokens. And if you look at kind of usable language in English on the internet for training, we're talking about anywhere between 10 and 15 trillion tokens. There's a massive amount of code th…”
Kant: Poolside is generating intermediate developer reasoning data to unlock software AI
“The missing data set in the world to go from where models are today to being as capable as humans at building software is the data set that represents being given the task, all of your intermediate reasoning and thinking, the steps that you do, the code that y…”
Kant: Tesla will inevitably build the most capable full self-driving AI
“And so to me, Elon has one full self-driving. I think it's inevitable that, that the most capable AI for full self-driving is coming out of Tesla because they've been gathering and building up this data set.”
Kant: Poolside runs world's largest code AI training environment with 130,000 codebases
“We put it in an environment, say it's an environment with a 130,000 real world code bases, several orders of magnitude, the largest environment in the world.”
Kant: Data is the real differentiator between competing AI models
“The real differentiation between two models is the data.”
Kant: AI hardware and algorithmic optimizations have decades of gains left
“And we've got probably decades, if not, you know, hundreds of years of improvements still left there and different forms of it over time.”
Kant: Synthetic data only improves AI when evaluated by an objective oracle
“If you have something that can determine an oracle of truth that can help say, this is better and this is worse, or this is correct and this is wrong, that's when you can actually use synthetic data.”
Kant: Software development will be first domain where AI closes human gap
“Our view of Poolside is that the first majorly economically viable capability that is going to close the gap between human intelligence and machine intelligence is software development.”
Kant: Google relies on Broadcom for TPUs while Amazon works with fabs
“Because while Google has to work with Broadcom to be able to bring TPUs into the world, Amazon is working with the fabs directly.”
Kant: LLM providers are in a drunken bar fight over model pricing
“I often refer to it as a drunken bar fight that's happening in our industry, is that we are, you know, all these companies are massively incentivized to drop the cost of their models as quickly as possible.”
Kant: Speech recognition AI has effectively closed the human capability gap
“Models today, in my opinion, are pretty much there. Maybe there's a tiny bit left to say, but we've closed that gap, you know, into an incredible amount.”
Kant: Neither OpenAI nor Poolside is allowed to train on private code
“But private code, no one's allowed to train on. Not us, not OpenAI. So all of us have access to the same public data.”
Kant: Poolside brought 10,000 GPUs online in summer 2024
“The 10,000 GPUs that we've now brought online this summer, you know, that came from this capital, allow us to make incredible advancements in model capabilities”
Kant: Interconnecting over 32,000 GPUs is currently extremely challenging
“Today, interconnecting more than 32,000 GPUs is extremely challenging.”
Kant: $100B is merely the starting price to become an AI hyperscaler
“If you want to become a hyperscaler that is able to put data centers all over the world with GPUs in it that is, that are going to allow you to serve these models to everyone, an infrastructure player, that's probably it. And that's probably just a starting po…”
Kant: Model training is CapEx, while inference is OpEx
“The creation of models is CapEx. The operating of them, the inference to running of them is OPEX. But the OPEX to run them requires extremely large scale physical footprint in the world.”
Kant: AI infrastructure build-out is largest since cloud boom
“I think this is the, one of the largest build outs that we've seen in physical infrastructure since, you know, the last couple of decades in, in the cloud.”
Kant: Data Centers Built in Two Years Will Look Radically Different
“The data centers from, you know, two years ago versus the data centers in terms of size and power requirement that we're going to see in the next two years look radically different, not just because the scale of number of servers and nodes that we're interconn…”
Kant: AI training requires co-located servers, while inference can be distributed
“For inference, we don't need all of the machines to be connected to each other in the same place. For training, we need them all to be connected to each other in the same room, in the same place.”
Eiso Kant: NVIDIA, Google, and Amazon Lead AI Hardware Volume
“The reason I mentioned those three specifically is that they are all building extremely large volume of chips. And constantly iterating on faster and better and better generations of chips for training and for inference. They're, for me, the three primary play…”
Nvidia GPUs yield 2x AI training performance gain every two years
“Pretty much what we've seen consistently with every two-year generation of training from NVIDIA is about a two X, one and a half to two X, you know, performance increase. But training is about two X every two years.”
Kant: New Nvidia Blackwell GPUs do not force immediate training cluster upgrades
“The Blackwell generation for us from a training perspective doesn't unlock anything new. It just means that we have to, we can do more with a certain set of chips. My H 200 become less valuable in the world, but it does not necessarily mean I have to go upgrad…”
Eiso Kant: Capital scales AI compute directly, but not talent or data
“What we are going to find is that for compute, dollars have a direct one-on-one effect. But when we look at data, when we look at proprietary applied research, and we look at talent, It is not as straightforward as dollars in magic, you know, success out on th…”
Kant: Poolside deliberately excluded Google, Microsoft, and Amazon from $500M round
“If look at our capital raise, our last five hundred million dollar round, you'll see that there's none of the big hyperscalers, Google, Microsoft, you know, Amazon were part of the round.”
NVIDIA participated in Poolside's funding round without taking an outsized stake
“There is one corporate that became part of our round, and that was very deliberate, was NVIDIA. Not in an outsized stake manner at all. And it's because we collaborate really closely with them On, you know, the next generation of their chips, on the software a…”
Big tech investing in frontier AI startups is game-theory optimal
“The nature of large technology companies choosing to invest, you know, in frontier AI companies is, is frankly the game theory optimal thing for them to do.”
Kant: Very few capable frontier AI startups remain available for acquisition
“I think there's very few left to be acquired, to be very honest.”
Kant: xAI built a 100,000 GPU cluster in Tennessee in mere months
“They built a 100,000 GPU, three, 32 K interconnected clusters in Tennessee in the span of months, and XAI showed up with Elon's strength, the ability to build physical infrastructure incredibly fast in the world.”
Kant: Elon Musk repeatedly built companies that shouldn't have existed
“I think Elon is one of the most impressive examples of someone who has done this For such a prolonged amount of time, at moments in time when the entire world, you know, refused to align around his view, I think there's companies in the world that get built be…”
Kant: Poolside, OpenAI, and Anthropic will rival Big Tech in AI
“What I would like to see and what I think is going to happen is that it's not just Google, Amazon, and Microsoft. It's going to be, you know, an open eye, an anthropic, a poolside, and a set of companies who are able to get that massive escape velocity needed …”
Kant: Software development will stay developer-led while becoming increasingly AI-assisted
“AI for software developers. I don't think anyone in the world anymore questions that software development moving forward is going to be a, for the foreseeable future, a developer-led, AI-assisted world, an increasingly AI-assisted world.”
Kant: AI image generation is seeing steady commoditization
“While there's still a bit of a gap, I think image generation is one already where we're seeing more and more commoditization over time.”
Kant: DeepMind built the primary European AI talent pool in London
“The number one company we have to give credit to is DeepMind. DeepMind built an incredible talent base, and they built it out of London.”
Kant: Unlike normal startups, building AGI is a direct competitive race
“Most startups are not racist. Most startups are against your, yourself. But AGI is a race, and so our view always has been, is that the team that we build is a team that is deeply passionate to be in that race”
Kant: Elite tech talent willing to make extreme sacrifices exists worldwide
“I have found no shortage of people in Europe that, or in America, there's a stereotype about Europe But the fact of the matter is that people who want to join races and do truly their life's work, they're built differently, and you can find them all over the w…”
Kant: AI value will accumulate across the stack, not just model layer
“Because I agree with you that the value is not going to only accumulate at the model layer. It's going to accumulate all the way to the end user.”
Kant: Third parties will build more value on Poolside than Poolside alone
“But I still actually look at this thinking that there will be more value built on top of us in the future than what we can possibly unlock only ourselves.”
Kant: The vast majority of interesting open AI research comes from China
“The research that still gets published openly, right, that doesn't get held back, that is most interesting, is all coming out of China in vast spades and majorities.”
Kant: China is not years behind the West in AI progress
“No, I think China is at an incredible level of capabilities and in no way should be discarded or thought of as years behind on AI or AGI progress.”
Kant: Rejecting GitHub's all-stock acquisition offer was my worst financial decision
“It was probably the dumbest financial decision of my life, considering it was an all stock offer and GitHub sold to Microsoft, I think, less than a year later.”
Kant: Believing AI progress will halt in 10 years is a misconception
“That it's going to, that progress is going to halt.”
Kant: Only global conflict disrupting chip supply could halt AI progress
“Global conflict that disrupts the supply chain of chips.”
Eiso Kant: AI regulation disproportionately harms young startups over funded incumbents
“The reality of regulation in many cases is that it becomes an expensive bureaucratic overhead. And that harms the most young startups. It doesn't harm companies that have raised massive amounts of capital.”