Four CEOs on the Future of AI: CoreWeave, Perplexity, Mistral, and IREN
- 01The Infrastructure Layer is the Silent Winner of the AI Era
- 02Multi-Model Orchestration is an Emergent and Durable Moat
- 03Power and Land, Not GPUs, Are the Binding Constraint
Episode: Four CEOs on the Future of AI: CoreWeave, Perplexity, Mistral, and IREN Podcast: All In Participants: Aravind Srinivas (Perplexity), Arthur Mensch (Mistral), Daniel Roberts (IREN), Jason Calacanis (Host), Michael Intrator (CoreWeave)
1. Key Themes
The Infrastructure Layer is the Silent Winner of the AI Era
The most durable businesses in AI may not be the model makers but the infrastructure providers who sit between the GPU silicon and the models themselves. CoreWeave's "box" financing structure — where client contracts collateralize GPU debt — has allowed a relatively small company to raise $35 billion in 18 months. The key insight is that infrastructure companies with long-term contracts are de-risked in ways pure-play model companies are not.
"We kind of live above the NVIDIA GPUs, but below the models. And everything in there, all the software, the integration of software and operations and observability and all the things that you need to be able to build a cloud that's purpose built for this one specific use case." — Michael Intrator [00:06:07]
"Within two and a half years of a five year deal, we have paid for everything. The principal has been paid off. The interest has been paid off." — Michael Intrator [00:20:27]
Multi-Model Orchestration is an Emergent and Durable Moat
Rather than betting on one model winning, Perplexity has positioned itself as an "orchestra conductor" across all models. As models specialize rather than commoditize, the value of a neutral orchestration layer increases. This is a non-obvious structural advantage that no single model company can replicate without undermining their own positioning.
"One advantage we have that all these companies you mentioned don't have is the multi-model orchestration. 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." — Aravind Srinivas [00:47:54]
"Dario, CEO of Anthropics, said recently in an interview that models are specializing... Even within coding, Cloud Code and Codex have very different capabilities. Our iOS engineers love using Codex. Our backend engineers love using Cloud Code." — Aravind Srinivas [00:48:42]
Power and Land, Not GPUs, Are the Binding Constraint
The bottleneck in AI infrastructure has quietly shifted from GPU availability to energy and physical land. IREN's early land and power acquisition strategy — locking up 4.5 gigawatts years before demand materialized — is now proving to be the most defensible moat in the industry. Data centers are following renewable energy sources, not traditional real estate logic.
"We've got four and a half gigawatts. For context, that's almost as much power annually as the Bay Area uses in its entirety. So for us, the hurdle or the constraint is really time to compute." — Daniel Roberts [00:22:02]
"In West Texas, where we're located, there's around 45 to 50 gigawatts of wind and solar. The transmission line to export that down to the load centers in Dallas and Houston is 12 gigawatts. So you go and locate to the source of low cost excess renewable energy, monetize it into this digital commodity, export it at the speed of light as tokens." — Daniel Roberts [01:26:40]
2. Contrarian Perspectives
GPU Depreciation Fears Are Manufactured by Short Sellers
The mainstream narrative that GPUs become obsolete in 18 months is being actively promoted by parties with financial incentives to talk down GPU-heavy stocks. The ground truth — 5-6 year customer contracts and rising prices for older-generation A100s — contradicts this narrative entirely.
"My take on the GPU depreciation debate is that it's nonsense, right? It's a debate that is being brought to the forefront by some traders that have a short position in the stock and they're trying to talk down." — Michael Intrator [00:10:36]
"The A100s, the amperes this year, the price has appreciated through the year. Why is that? I think it's because as more installed capacity becomes available, you have new companies that come into existence that have new use cases." — Michael Intrator [00:11:58]
More Compute Creates More Demand, Not Saturation
The conventional worry is that AI compute investment will overshoot demand. The contrarian view, supported by Jevons' Paradox logic, is that as compute gets cheaper and faster, usage explodes rather than plateaus. The current "dial-up" experience of AI image generation today will become real-time, triggering orders of magnitude more usage.
"If we 10x the amount of compute available... and those images take five to 10 seconds, are we going to generate more or less images? This is Jevons' Paradox. You build a couple more lanes. People start to think, well, maybe the distance from Bondi Beach to the central business district is an acceptable commute." — Daniel Roberts [01:30:02]
Open Source Models Are Better for Enterprises Than Closed Models
Against the prevailing assumption that closed frontier models (OpenAI, Anthropic) will dominate enterprise, Mistral argues that open source provides structural advantages in data security, customization depth, and IP leverage that closed models fundamentally cannot match.
"If you have open models, you can actually add new parameters. You can make a lot of deeper things that you cannot do with closed models... building on open source technology is a way to save cost, is a way to have better control because you can run the thing on every cloud that you want, on your hardware if you want, you can deploy it on the edge if you want." — Arthur Mensch [00:10:15]
AI Is Not the OS — It Replaces the Paradigm of the OS
The contrarian framing here is that we don't need a new operating system layer. AI itself becomes the operating system by shifting from programmatic instructions to objective-based computing. The entire concept of an OS as a command-execution layer becomes obsolete.
"AI is the operating system. Earlier in the traditional operating system, you execute programmatically. Now you start with objectives, not specific instructions. You come up with a high level objective... So I think models, systems and files and connectors are all coming together. You would think of that as an OS, except you're operating at an abstraction above that where you're thinking in terms of objectives." — Aravind Srinivas [00:42:02]
3. Companies Identified
CoreWeave Specialized GPU cloud provider; the dominant "NeoCloud" for AI compute. Mentioned for pioneering GPU-backed debt financing ("the box"), achieving 600 basis point reduction in cost of capital, and being the first to commercially deploy H100s, H200s, and GB200s at scale. The company raised $35 billion in 18 months.
"CoreWeave, which is a company that many people haven't ever heard of, was able to go out and raise $35 billion in 18 months to build infrastructure at scale." — Michael Intrator [00:20:02]
Perplexity AI-native search and agentic computer product with tens of millions of monthly users. Mentioned for positive gross margins on all revenue (unlike "wrapper" companies), fastest-growing enterprise segment, and multi-model orchestration moat. Has reportedly saved Enterprise Max customers over $100 million.
"Every revenue Perplexity makes has positive gross margins... because we're not just selling tokens. Most of our revenue is recurring because people are paying a subscription fee." — Aravind Srinivas [00:45:55]
Mistral AI French open-source AI company training frontier models with NVIDIA. Mentioned for a unique enterprise deployment model — sending PhD scientists to client sites to train bespoke models on proprietary data without data ever leaving client infrastructure.
"We send the technology. Typically, we send a little bit of scientists because you do need that expertise transfer and that knowledge transfer in between our teams and the vertical experts." — Arthur Mensch [01:12:40]
IREN (formerly Iris Energy) Publicly traded data center company that began as a Bitcoin miner and is transitioning to AI compute. Mentioned for locking up 4.5 gigawatts of power capacity, signing a $9.7 billion contract with Microsoft (representing only 5% of capacity), and using 100% renewable energy since inception.
"We signed a $9.7 billion contract with them late last year. But as I was explaining to you before the show, that's 5% of our capacity." — Daniel Roberts [01:21:35]
Eleuther AI Open-source AI research group. Mentioned as the unlikely origin story of CoreWeave — when CoreWeave donated A100s to Eleuther AI volunteers, it served as paid tuition to learn how to run parallelized AI compute at scale, and those researchers later drove commercial demand.
"I kind of feel like buying those initial GPUs was the tuition we paid to learn how to run this business." — Michael Intrator [00:04:49]
4. People Identified
Michael Intrator Co-founder and CEO of CoreWeave. Former algorithmic hedge fund manager focused on natural gas. Mentioned for pioneering the GPU-backed "box" financing structure that has become an industry standard, reducing CoreWeave's cost of capital by 600 basis points, and his prescient early bet on GPU compute versatility starting in 2017.
"We have dropped our cost of capital by 600 basis points. Wow. It is enormous. Right. And so you're seeing a company that is driving its cost of capital down towards where the hyperscalers borrow, which will enable us to be able to be competitive with them over time." — Michael Intrator [00:21:48]
Aravind Srinivas Co-founder and CEO of Perplexity. Mentioned for building a profitable (positive gross margin) multi-model AI company at scale with ~400 people competing against trillion-dollar incumbents, and for a clear product vision of AI as the universal computer operating through objective-based orchestration.
"Speed is our moat. One of the things that big companies cannot do is move at the speed we do, serve customers at the speed and quality." — Aravind Srinivas [00:51:35]
Arthur Mensch Co-founder and CEO of Mistral AI. Mentioned for building the leading European open-source AI company with a differentiated enterprise go-to-market — deploying PhD scientists on-premise at client sites (including ASML and HSBC) to train models on proprietary data with zero data egress.
"Our technology is a set of services, a set of training tools, a set of data processing tools that I can take and that I can put on the infrastructure of my customers. So suddenly, the flow of data doesn't go. There's no data flow coming back to Mistral because everything stays there." — Arthur Mensch [01:12:10]
Daniel Roberts Co-CEO and co-founder of IREN (alongside his brother). Former Australian entrepreneur who built one of the most defensible positions in AI infrastructure by developing owned data centers years before demand emerged, securing 4.5 gigawatts of renewable-powered capacity. Mentioned for strategic foresight, community-first hiring approach, and 100% renewable energy commitment.
"We set about to build out large scale data centers. Yes, the first use case was Bitcoin mining. But as we said to our seed investors, use that to bootstrap the platform, generate cash flow, layer in higher and better use cases over time as they emerge." — Daniel Roberts [01:19:30]
Sarah Fryer CFO of OpenAI. Mentioned in passing for sharing a remarkable data point illustrating the deflationary power of AI compute: the cost per million tokens fell from $32 to $0.09.
"She was talking about the cost of a million tokens when ChatGPT3 came out. And it was $32 and change. And now a million tokens cost $0.09." — Michael Intrator [00:31:08]
5. Operating Insights
Build the "Box" Structure to Scale Capital-Intensive Businesses
CoreWeave's SPV-style "box" financing — where client contracts, hardware, and data center leases are bundled as collateral — is a replicable playbook for any capital-intensive business with long-term contractual revenue. The key discipline: only accept counterparties with strong balance sheets and contracts of sufficient length to amortize all costs. This de-risks lenders, reduces cost of capital over time, and protects against contagion between deals.
"Once I have a contract in hand, I take my contract with Microsoft and I put it in the box... and the first thing it does is it pays the data center, it pays the power bill, it pays the interest and the principal. And then whatever's left flows back to us." — Michael Intrator [00:19:41]
Use AI Tools to Compound Product Shipping Velocity
Perplexity is using their own AI computer product internally — non-engineers are shipping code via Slack bots, board memos are being generated in one shot, press briefing dossiers are auto-generated. This internal dogfooding loop creates a compounding advantage: the product improves faster because the team itself accelerates with it.
"AI coding tools have made it much faster for us to ship things, which is honestly one of the reasons why we built computer because now even non-engineers are shipping code here by just bringing a slack bot and asking it to fix bugs. The iteration has just been like exponential." — Aravind Srinivas [00:52:06]
Route to Customers That Cannot Complain About Early-Stage Quality
CoreWeave's origin story contains a tactical gem: donating GPUs to open-source researchers meant SLA expectations were zero, giving the team space to learn operational excellence without commercial consequences. For early-stage infrastructure or deep-tech companies, finding "forgiving" early customers (academic labs, open-source projects, nonprofits) is a legitimate and underused go-to-market strategy to pay tuition at low cost.
"They can't really get pissed at us if we're not very good at it initially... and that began to really give us an understanding of what was necessary to run scale parallelized computing." — Michael Intrator [00:03:51]
6. Overlooked Insights
The "Context Engine" is a Billion-Dollar Unsolved Enterprise Problem
Arthur Mensch briefly mentioned a concept called a "context engine" — a semantic mapping layer that understands what enterprise data exists, who owns it, and what access permissions apply, so that AI agents can't inadvertently surface compensation data to engineers or expose confidential HR information. This was mentioned almost as a footnote, but it is actually the core unsolved problem blocking enterprise AI adoption at scale. Any company that cracks enterprise-grade, role-aware, semantically-aware data access control for AI agents will be foundational infrastructure. No major player has solved this yet.
"What you actually need, which is hard to do, is what we call a context engine. So a mapping of where the data sits that comes with a certain number of metadata that is telling you that this data is actually not accessible to this part of the company... So that's hard. It's actually hard. You need to rethink entirely the way your IT systems are being connected." — Arthur Mensch [01:18:15]
API Access to Walled Gardens Is a Massive Untapped Revenue Stream Nobody Is Monetizing
Jason Calacanis raised — and Aravind agreed with — the idea that LinkedIn, Reddit, and the New York Times should offer authenticated, rate-limited, paid API access to AI agents acting on behalf of subscribed users. This was treated as a casual wishlist item, but it represents a structural new revenue category for every platform with proprietary data behind a login wall. The fact that LinkedIn charges $50/month for premium but doesn't let AI agents act on that subscription is a gap that either these platforms will eventually monetize or that a middleware company will exploit. First mover here has significant leverage.
"LinkedIn, I already pay LinkedIn like 50 bucks a month. They should just let the $50 a month one work with computer... I think fundamentally giving users a choice and setting it up as a win-win for both the business and the user is where the world should head to." — Aravind Srinivas [01:06:01] and [01:06:18]