Nov 6, 2025 · 1h 6m · mad
Intelligence Isn’t Enough: Why Energy & Compute Decide the AGI Race – Eiso Kant
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 episode of the MAD Podcast, Poolside Co-CEO Eiso Kant joins Matt Turck to discuss Poolside's multi-gigawatt Project Horizon data center, novel reinforcement learning research paradigms like RL2L, and why controlling physical energy and compute infrastructure is essential for frontier AGI labs.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Matt holds 12.5% of the talking time here. How this is scored →
Matt as informed peer 3.4 Guest teaching 4.8 Guest disagreement 1.7 Matt pushing back 1.3
speaking balance: gold is Matt, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 34:50 Rejection of expert rubric scaling narrative
Guest directly attacks the consensus industry narrative that scaling environments with human expert rubrics will lead to AGI, calling it a temporary teenage phase.
Hardest push from Matt ▶ 53:09 Host raises the Bitter Lesson argumentHost explicitly pushes back on Poolside's core focus, asking whether spending years focused on software development runs counter to Rich Sutton's Bitter Lesson.
Biggest teaching moment ▶ 11:00 Data center financial and operational breakdownGuest provides an authoritative masterclass breakdown of 250MW data center economics ($8B TCO, $2B powered shell, $5.5B compute) and explains how disintermediation reduces token costs by 30%.
Matt holds his own ▶ 37:15 Synthesis of RL2L into Option A vs Option BHost demonstrates deep domain grasp by taking guest's novel RL2L technique and instantly re-framing it into an intuitive Option A versus Option B comparison.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Matt as informed peer | Guest teaching | Guest disagreement | Matt pushing back | Why |
|---|---|---|---|---|---|---|
| Welcome Back & Foundational Vision of Poolside | 2 | 3 | 1 | 1 | Host welcomes guest back and prompts him to justify Poolside's existence in an intense AGI race. Guest explains how early contrarian RL bets laid the groundwork for their current positioning. | |
| NVIDIA Funding and Project Horizon Debut | 3 | 4 | 2 | 1 | Host cites rumors of Poolside's $2B fundraise at a $14B valuation with $1B from NVIDIA to prompt the Project Horizon announcement. Guest explains why AI labs that do not build physical infrastructure are 'cosplaying' their business. | |
| Navigating Infrastructure Scale and Independence | 4 | 4 | 1 | 2 | Host synthesizes how indie labs get boxed out without hyperscaler backing like Anthropic or OpenAI. Guest details the 500k-acre Mitchell family land partnership in Texas for their 2GB campus. | |
| Cost Structure and Financial Margins of AI Infrastructure | 3 | 7 | 1 | 1 | Host asks about unit economics and gross margins for owning physical compute. Guest delivers an exhaustive breakdown of the $8B total cost for 250MW, explaining how eliminating margin stackers enables 30% cheaper tokens. | |
| Project Financing and the Hybrid CoreWeave Model | 4 | 3 | 1 | 1 | Host demonstrates understanding of infrastructure deals by asking if it is structured as project financing. Guest confirms and explains the hybrid anchor-tenant structure with CoreWeave. | |
| Hiring Physical Infrastructure Experts and Team Culture | 2 | 4 | 1 | 1 | Host asks how a software startup hires physical infrastructure experts without falling victim to skill gaps. Guest outlines his interview framework to avoid Dunning-Kruger traps. | |
| Hybrid Modular Construction and Supply Chain Strategy | 3 | 6 | 1 | 1 | Host references xAI's rapid build speed as an industry benchmark. Guest explains Poolside's hybrid modular approach using 2.5MW skids to drastically lower on-site labor needs. | |
| Project Horizon Timelines and Environmental Considerations | 2 | 5 | 1 | 1 | Host asks about delivery timelines and environmental impact, stopping to ask guest to define an SCR. Guest educates on catalytic reduction and water dynamics in Pecos County. | |
| From Code Execution Feedback to Agentic Reinforcement Learning | 4 | 5 | 2 | 1 | Host prompts transition to RL by referencing code execution environments from their podcast two years ago. Guest details scaling from 10k to 1M environments and moving to agentic RL. | |
| Beyond Rubrics: Announcing Reinforcement Learning to Learn (RL2L) | 2 | 6 | 4 | 1 | Guest forcefully rejects the narrative that scaling expert rubrics leads to AGI. Guest publicly unveils Reinforcement Learning to Learn (RL2L) as Poolside's new research paradigm. | |
| The Four-Stage Paradigm of Foundation Model Training | 5 | 4 | 1 | 2 | Host synthesizes guest's RL2L concept into an intuitive Option A vs Option B comparison. Guest commends host's framing and outlines the four-stage training paradigm. | |
| Hot Stove Problem, Single-Sample Learning, and AGI Realities | 3 | 6 | 2 | 1 | Host prompts guest to explain continuous learning. Guest discusses gradient descent limitations using the 'hot stove problem' and deconstructs popular definitions of AGI. | |
| Andrej Karpathy's Influence and How Agents Collapse into Models | 5 | 5 | 2 | 2 | Host cites Andrej Karpathy's claim that agents are 10 years away to challenge standalone agent startups. Guest details how external tools and agents inevitably collapse into the base model. | |
| Reaching Human-Level AI Capabilities and Long-Term Business Strategy | 4 | 6 | 3 | 2 | Host presents the argument that AI scaling is hitting a wall. Guest refutes this by explaining hardware cycle bounds, RL duration scaling, and optimization efficiency. | |
| Poolside's Master Plan and Enterprise Expansion with Redpanda | 4 | 4 | 2 | 2 | Host challenges guest on whether narrow software focus clashes with the Bitter Lesson. Guest defends the master plan, demonstrating expansion into enterprise workflows with Redpanda. | |
| Product Family Roadmap, Enterprise SCIF Deployments, and FDRE Motion | 4 | 5 | 2 | 1 | Host asks about model roadmap and balancing enterprise secrecy with developer public relations. Guest outlines Laguna model lineup, SCIF deployments, and Forward Deployed Research Engineers. | |
| Global Talent Strategy and Avoiding the Silicon Valley Echo Chamber | 3 | 4 | 2 | 1 | Host asks about re-centering from Europe to the US. Guest clarifies corporate structure and emphasizes the strategic advantage of hiring outside the Silicon Valley echo chamber. |
Statements from this episode (38)
Kant: Next-token pre-training gains hit a sigmoidal curve and slowed down
“The first paradigm of kind of pre-training of predicting the next token on the web was becoming sigmoidal and was slowing down in terms of the gains that it had.”
Kant: AI labs build general reasoning faster by focusing on single proxies
“To get to highly capable intelligence that can reason, that can do planning, that can understand the world. You're almost better off putting a set of blinders and focusing on, on one proxy for that.”
Eiso Kant: AI intelligence is going to become a commodity
“I would actually go as far as saying that intelligence is going to become a commodity.”
Kant: Frontier AI won't be winner-take-all, but limited to few labs
“I don't think it's a winner takes all market, but it does seem to be that it's a small number of companies who are able to get there.”
Kant: The three critical AI stack layers are energy, compute, and intelligence
“There's three layers of the stack that fundamentally are going to matter. It's energy, it's compute, and it's the intelligence built on top.”
Kant: If AI intelligence commoditizes, only scale and delivery cost matter
“And within this world, if you think that intelligence is going to become less distinguishable between the companies building it, And becomes a commodity probably more like oil or cloud compute than like bread at the bakery, is there's two things that matter, y…”
Kant: Foundation model labs lacking physical infrastructure are 'cosplaying' their business
“If you're a foundation model company, and you're not building physical infrastructure, you're cosplaying your business.”
Kant: AI is on track to rewrite $29T of global knowledge work
“I don't think anyone has any doubts anymore that we're now on track to reach human level capabilities and intelligence. And in that world, 29 trillion dollars of knowledge work rewrites itself, right?”
Kant: Poolside's Project Horizon is an expandable two-gigawatt data center campus
“So Project Horizon, in a nutshell, is today announced a two gigawatt campus. We can actually go far beyond two gigawatts on the site that we're in.”
Kant: Poolside starting construction on 250MW natural gas data center in 2025
“And so what you're seeing from us is that before the end of this year, we're starting construction on a 250 megawatt data center. It's natural gas powered with grid connection as a backup, and it will house an incredible amount of compute, but it will be the f…”
Kant: AI foundation model gross margins will settle near 40%
“I'm not yet convinced that gross margins in our industry will look like a SaaS company, like 80% plus. I think when we're talking about a commodity as intelligence, that we build value added services on top. Right. It's a foundation model company. On one hand,…”
Kant: A 250MW AI data center costs roughly $8 billion to build
“To energize, 250 megawatts is about half a billion dollars worth of physical infrastructure. So in the case of gas turbines, it's gas turbines, but there's different paths to it. Building a data center, like the powered shell that brings together all of the eq…”
Kant: Full-stack AI infrastructure ownership cuts token costs 20-40%
“So when you take all those margins out, all of a sudden you can start seeing that you can serve your tokens, 2030, 40% cheaper than someone else.”
Kant: CoreWeave is the anchor tenant for Poolside's new data center
“They are the anchor tenant on our data center.”
Kant: Founders entering new domains should interview 100+ candidates per role
“And what I found to be one of the best advice I got a very long time ago, whenever you find yourself in that situation, well, this is when you want to start finding the experts to join you, but how do you find the experts? Because you yourself aren't. And here…”
Kant: Large-scale data center construction is concentrated among very few players
“What was particularly interesting to learn about the infrastructure space over time is that it's an incredibly small world. Everybody knows each other. It makes tech feel like a big world, right? The construction of data centers at scale has really only been d…”
Kant: Poolside is building a 250MW data center with under 450 workers
“When we're building 250 megawatts, We're doing it with a less than 450 person on site workforce with very skilled tradesmen and construction workers that we're bringing over there and we're housing over there versus a world where you might need traditionally 2…”
Kant: Poolside first compute online late 2025, 250MW by Q1 2027
“So at the beginning of Q four, end of Q three, so next year, we've got the first compute coming online. And then we're finalizing in Q one, 20, 27, the first you know, 250 megawatts, but already next summer, the next 250 megawatts starts construction. It's a s…”
Eiso Kant: Renewable-only data centers with battery storage are not economically viable yet
“Renewables only, in combination with the battery capacity that you would need to power a data center if you want renewable only, it's just not economically viable yet.”
Poolside plans eight years of construction for West Texas data center project
“When we talk about this big multi-gigawatt project, we've got about eight years worth of construction that we are planning out there”
Poolside scales its RL environment repository to over 1M codebases
“Right now we have over a million of those environments and tens of millions of revisions, putting us in like the thirty million range.”
Kant: Expert RL will lift enterprise agent quality to 90% in years
“And I think what we'll see in the next couple of years is a lot of model companies ourselves as well. We go into these enterprises. We find high value use cases. We use our agents to get to 60 or 70% You know, of quality, and then use additional reinforcement …”
Eiso Kant: Expert-created rubrics and environments will not scale to AGI
“But the singularly collecting of skills This way by creating environments and experts we don't think will scale to artificial general intelligence.”
Kant: Internet data lacks the thought processes needed for AGI
“But the reason that never got us to AGI is because, well, the internet never actually included the data set of the thoughts and actions that created it. It's the final piece of code, it's the final article you write, but not the thoughts that you had and actio…”
Kant: RL2L will enable model reasoning much earlier in training
“We think we'll push models to a level of reasoning and thought That will happen far earlier in their training than it does today.”
Kant: AI training will evolve into continuous learning from real-world agent experiences
“And then the fourth stage of training over time increasingly will become learning, continuous learning from real world experiences of these agents. And so those are kind of the four stages that we think training will go to.”
Kant: Foundation models currently cannot perform single-sample learning
“Foundation models today can't do it, and we don't yet have a path In them successfully doing so in a way that, like, really improves it.”
Kant: AI models and AI agents are effectively the same thing
“Models and agents to me are effectively the same thing.”
Kant: External software and agent frameworks will collapse into base models
“But we have a phrase at poolside, which is over time, everything collapses into the models.”
Kant: AI models will equal top human knowledge workers within 36 months
“I think we have a hard time holding the point of view that models will reach the same level of intelligence and capabilities that the world's most capable people in every field have. And when you take a step back, and don't take the next 12 month view, but jus…”
Kant: Most vertical software and agents will disappear when AGI arrives
“A whole bunch of vertical software or vertical agents are probably gone.”
Kant: Linearly adding GPUs to train larger AI models causes exponential training delays
“It is not that I can linearly add more GPUs and train increasingly a more larger model. If I do so, the time it takes becomes exponentially longer.”
Kant: Poolside aims to power all knowledge work globally
“And so we're starting to open up much more broader outside of software development and come back to what we've always said, like we want to be able to power all of knowledge work in the world.”
Kant details Poolside's Laguna model training timeline and 41,000 GB300 cluster
“That's our upcoming family of models, Laguna, where we'll have a small, a medium, and a large. So small is finishing training in a couple of weeks, medium actually starts training this week and large starts training as our 41,000 GB 300 come online.”
Kant: Poolside's Malibu agent matches initial Gemini 2.5 Pro SWE-bench scores
“But if you think about the Malibu agent as a coding agent, for instance, right now, it sits at a level of like sweet bench verified, for instance, where Gemini two and a half pro was when it came out.”
Kant: Poolside operates inside defense SCIFs and air-gapped servers
“So we today operate on workstations that go into SCIFs. We operate on servers that are, you know, on-prem or even as far as air-gapped.”