Interview
Eiso Kant, CTO @Poolside: Raising $600M To Compete in the Race for AGI | E1211
- Strategic Context & Funding: Poolside has secured a $500 million round (totaling $600 million to date), providing the capital necessary to field 10,000 GPUs and enter the race for Artificial General Intelligence (AGI); the company views the current moment as a historical inflection point comparable to the early internet or mobile eras.
- Core Mission: Poolside is focused on building the most capable AI specifically for software development, operating on the belief that human-level capabilities in this domain are a prerequisite for broader AGI and represent the most economically valuable gap to close first.
- Data Strategy: The company argues that existing web-scale code (3 trillion tokens) is insufficient because it lacks "intermediate reasoning" data; Poolside is synthesizing this missing data by generating millions of solutions and capturing the full iteration process (thinking, coding, running, failing) to train models via reinforcement learning.
- Technical Moat: Unlike general LLMs that rely on static text data, Poolside leverages "execution feedback" from deterministic code environments (130,000+ real-world codebases) to provide objective, binary validation (pass/fail) for model outputs, distinguishing its synthetic data generation from purely simulated environments like AlphaGo.
- Compute & Infrastructure: CEO Daniel Diner emphasizes that while algorithms and hardware are "table stakes," compute remains the primary bottleneck; the company's ability to scale is currently dependent on the physical limitations of GPU interconnects, noting that clusters of 32,000+ GPUs are extremely challenging to build and that 100,000+ is the near-future limit.
- Market Positioning: Poolside deliberately avoided investment from major hyperscalers (Google, Microsoft, Amazon) to maintain independence, with the notable exception of NVIDIA, reflecting a strategy to compete as a standalone entity rather than a subsidiary.
- Talent Distribution: Despite being headquartered in London, the company is positioning itself as a global, US-centric operation with significant talent pools in the UK, France, and Israel; they chose this geographic spread to access deep technical expertise in Europe (including the DeepMind and Yandex diaspora) that remains underutilized by US-centric competitors.
- Scaling Economics: The interviewee posits that the industry will move toward "distillation," where massive, expensive frontier models (trillion-parameter) are trained and then distilled into smaller, economically viable models for inference, rather than relying on a single massive model for all user interactions.
- Cost Dynamics: A "price war" is underway among hyperscalers and frontier startups, driven by vertical integration (in-house silicon like NVIDIA, Google TPUs, Amazon Trainium/Inferentia) which reduces reliance on expensive third-party hardware and lowers the marginal cost of inference.
- Geopolitical View: Poolside rejects the notion that China is behind the US in AI progress; instead, they argue China is publishing vast amounts of open research to attract talent and that the West's best strategic move is to remain attractive to global talent, including those from China.
- Personal Philosophy: The CEO frames his motivation not as financial gain but as an inability to find peace without solving the hardest problems, citing a "why" driven by the need to close the gap between human and machine intelligence to unlock global abundance in energy, space, and technology.
- Risk Assessment: The primary failure mode identified is "stumbling" in either the capabilities race or the go-to-market race, as the competitive landscape allows no margin for error in execution or speed; the company must be excellent in both simultaneously.
- Regulatory Stance: The CEO advocates for regulating the end-user application of AI rather than constraining the training compute or research, arguing that such limitations would disproportionately harm small startups while benefiting well-capitalized incumbents.
- Historical Perspective: The interviewee draws parallels to the automotive industry's consolidation, suggesting that while many startups will fail or be acquired, the companies that survive the initial capital-intensive phase will become the foundational platforms for future economic value, similar to how BYD evolved from a battery component maker to a dominant EV manufacturer.