Four CEOs on the Future of AI: CoreWeave, Perplexity, Mistral, and IREN
In a Nutshell
Four AI CEOs at Nvidia GTC detail their pivot from crypto mining to massive AI infrastructure: CoreWeave deploys bleeding-edge Nvidia GPUs via innovative "box" financing for 5-year contracts with hyperscalers like OpenAI, emphasizing inference monetization and 6-year hardware lifespans amid relentless demand constrained by power, memory, and data centers; Perplexity evolves from search to a hybrid AI "computer" with browser/root access, multi-model orchestration, and enterprise tools enabling one-person $1B businesses; Mistral focuses on open, specialized European models via Forge for secure enterprise customization; IREN swaps Bitcoin for AI data centers with 4.5GW renewable power, partnering Microsoft amid talent shortages and Jevons paradox-driven compute explosion. Key takeaways: AI demand overwhelms global capacity, inference drives value, capitalism's boom-bust cycles (e.g., memory, energy) fuel innovation, and hybrid local/server agents lower barriers for entrepreneurship while demanding robust governance.
These notes were generated by AI and may contain inaccuracies.
Interview with four AI CEOs at Nvidia's annual GTC conference: CoreWeave, Perplexity, Mistral, and IREN. Sponsored by the New York Stock Exchange, a modern marketplace for raising capital and building for the future.
CoreWeave, building massive infrastructure for hyperscalers. Michael Intrater, CEO, describes starting as the original hyperscaler by securing GPUs early. Began running an algorithmic hedge fund focused on natural gas. After algorithms were built, downtime led to interest in crypto. Examined Bitcoin mining but preferred GPUs for Ethereum mining due to versatility for other uses. Started company in 2017, mined crypto for first three years, weathered two crypto winters using hedge fund risk management expertise in capital allocation and exposure.
Scaled company and sought other use cases due to crypto volatility. Progression of products: crypto to CGI rendering for animating and rendering images in movies, then batch computing for medical research and science. Moved up the stack in GPU complexity. In 2020-2021, explored GPUs for neural networks. Bought A100s and donated to Luther AI open-source project volunteers. Learned scale parallelized computing from their feedback. Those researchers returned to day jobs demanding similar infrastructure, launching CoreWeave's business.
Recognized scaling laws would drive demand pre-ChatGPT (2020-2021). Compute decommoditizes at scale: anyone can run a GPU, but running world-changing model clusters is different. Focused on scaling delivery to larger clients, accessing capital for sophisticated consumers. Thinks of business above Nvidia GPUs but below models: software integration, operations, observability for AI-specific cloud. Doesn't do general web servers (AWS excels there), solves new AI compute problem.
Sign in to read the full notes
Get access to AI-generated notes, topic timestamps, and more.