TLDR
This All In episode recorded at NVIDIA's GTC conference features interviews with four AI infrastructure and model company CEOs: Michael Intrator (CoreWeave), Aravind Srinivas (Perplexity), Arthur Mensch (Mistral AI), and Daniel Roberts (IREN). They discuss the evolution of AI compute demand, business models, specialization vs. general models, and the physical infrastructure challenges powering the AI boom.
For you
PRO members can sign in and request more information — or their own version of this digest, tuned to what they care about.
Sign in to unlockSurprising Insights
GPU prices for older generations (A100s) have appreciated, not depreciated, as new companies find use cases for cheaper compute
CoreWeave's average contract is 5 years and they've paid off principal in 2.5 years, making GPU depreciation debates 'nonsense' per Intrator
Perplexity's enterprise business is growing faster than consumer, with $400/month 'Enterprise Max' tier having positive gross margins on all revenue
IREN uses 100% renewable energy (hydro in BC, wind/solar in West Texas) by locating data centers at stranded energy sources
West Texas has 45-50GW of wind/solar but only 12GW transmission capacity — data centers monetize this stranded energy
Key takeaways
AI compute demand is relentless and exceeds global capacity, with long-term contracts (5-6 years) proving GPU obsolescence fears are unfounded
The winning strategy is specialization: Perplexity orchestrates multiple models as 'Switzerland', Mistral builds verticalized models on open source for enterprise data privacy, CoreWeave focuses purely on AI compute infrastructure
Physical infrastructure (power, land, fiber, skilled trades) is the true bottleneck, not GPUs — companies like IREN secured 4.5GW of power years ahead
New financing models (CoreWeave's 'box' structure) enable massive capital deployment by isolating risk per contract and driving down cost of capital
AI agents and computer-use tools (Perplexity Computer, OpenClaw) are moving from chat to autonomous execution, enabling solo entrepreneurs to run businesses with minimal hiring
Notable quotes
“When we buy infrastructure, we're a success based company... our average contract is five years. So any commentary by anyone either inside or outside of the industry that this stuff becomes obsolete in 16 months or whatever nonsense... it doesn't in any way match up with the facts on the ground.”
▸ Notable quotes · 10:54“The depth of the demand for the service we provide has been relentless and overwhelms the global capacity of the world to deliver enough compute to enable all of the demand for artificial intelligence to be stated.”
▸ Notable quotes · 24:21“We're like Switzerland. We don't have to have one horse in the race. If GPT wins, Gemini wins, Claude wins, Lama wins, it doesn't matter to us.”
▸ Notable quotes · 47:58“AI is inevitable. Robotics is inevitable. The reality is the correlation between human progress and energy consumption is really, really high over a very long time period.”
▸ Notable quotes · 93:57Chapters
0:34 — CoreWeave: From Crypto Mining to AI Hyperscaler: Michael Intrator explains CoreWeave's origins in algorithmic trading and crypto mining, pivot to GPU compute for rendering and research, early donation of A100s to Luther AI, and development of the 'box' financing model enabling $35B capital raise. Discusses GPU lifespan, depreciation myths, and relentless demand.
33:03 — Perplexity: Multi-Model Orchestration and AI Agents: Aravind Srinivas describes Perplexity's evolution from search to browser agent (Comet) to computer-use agent (Computer). Emphasizes accuracy, multi-model orchestration as competitive moat, enterprise growth, hybrid local/cloud architecture, and vision for autonomous business operations.
67:16 — Mistral AI: Open Source Specialization for Enterprise: Arthur Mensch discusses Mistral's partnership with NVIDIA on frontier models, focus on open source for enterprise customization, portable platform for data segregation, forward-deployed engineers working with subject matter experts, and the importance of context engines for enterprise data governance.
79:03 — IREN: Power-First Data Center Infrastructure: Daniel Roberts details IREN's bootstrap from Bitcoin mining to AI compute, 4.5GW power portfolio, 100% renewable energy strategy by locating at stranded wind/solar in West Texas, labor challenges in remote areas, and views on nuclear, custom silicon, and space-based data centers.
Speakers
Jason Calacanis — Host/Interviewer
Michael Intrator — CEO, CoreWeave
Aravind Srinivas — CEO, Perplexity
Arthur Mensch — CEO, Mistral AI
Daniel Roberts — Co-CEO, IREN