All-In

Four CEOs on the Future of AI: CoreWeave, Perplexity, Mistral, and IREN

with Michael Trendler (CoreWeave), Aravind Srinivas (Perplexity), and others
23 Mar 2026 18 min read 2h 15m
TL;DR

CoreWeave CEO Michael Trendler explains how GPU infrastructure companies are financing massive compute buildouts using structured debt boxes tied to long-term customer contracts, enabling rapid scaling while driving down capital costs by 600 basis points. The conversation reveals that GPU scarcity remains extreme and long-lasting—contracts average 5 years, older chips retain value, and memory is now the real bottleneck, not compute itself.

Key Moments
Michael Trendler
“the the 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.”
Trendler dismisses claims that GPUs become obsolete in 16-18 months, citing CoreWeave's actual customer contracts averaging 5 years.
▶ 10:39
Michael Trendler
“So, what we do is we go out and we find a client. Let's use Microsoft. You brought them up before, right? And Microsoft comes to us and says, "We'd like to buy some thing." And we say, "Okay, great. We're going to sign a contract." Once I have a contract in hand, then what I do is I create something It's not a particularly creative name. It's called the box, yeah, right?”
Trendler explains CoreWeave's innovative financing structure that structures debt around customer contracts and GPU purchases.
▶ 19:03
Michael Trendler
“within 2 and 1/2 years of a 5-year deal, we have paid for everything. The principal's been paid off. Well, the principal's been paid off. The interest has been paid off.”
Demonstrating how the box structure achieves cash flow payoff well before contract maturity, giving lenders confidence.
▶ 20:32
Michael Trendler
“we have dropped our cost of capital by 600 basis points. Wow. It is enormous, right?”
CoreWeave's financing innovation has enabled the company to reduce borrowing costs substantially over two years as lenders gain confidence in the structured deals.
▶ 21:56
Michael Trendler
“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.”
When asked about demand levels and whether constraints like GPU availability could suddenly disappear, Trendler emphasizes sustained, overwhelming demand for four years straight.
▶ 24:21
About the show
›
All-In is a podcast featuring four successful investors and entrepreneurs discussing the latest in technology, business, and culture. This episode, recorded at Nvidia's GTC conference, features interviews with CEOs building the infrastructure and applications powering the AI era, including CoreWeave's GPU cloud platform and Perplexity's AI search engine.
Takeaways
1
The 'box' finances AI infrastructure at scale CoreWeave structures debt by bundling customer contracts, GPU purchases, and data center agreements into discrete cash flow vehicles. This allows the company to borrow against blue-chip customer commitments (Microsoft, OpenAI) rather than equity, achieving 5-year contracts that pay back principal and interest within 2.5 years. The approach has enabled CoreWeave to raise $35B in 18 months and reduce cost of capital 600bps.
2
GPU lifespan extends well beyond 2 years Despite Wall Street speculation that GPUs depreciate in 16-18 months, CoreWeave's customer contracts average 5 years, and older-generation chips (A100s, Ampere) are appreciating in price as new companies find use cases for them. The real constraint isn't obsolescence but secondary-market demand from startups and smaller enterprises priced out of cutting-edge silicon.
3
Memory becomes the new GPU bottleneck While GPU allocation was the constraint in 2023-2024, memory, power, networking, and optical components now limit AI infrastructure scaling. The memory market's cyclical fab-capacity dynamics mean investment decisions made in 2023 can't meet 2026 demand surges, creating structural shortages that favor early infrastructure builders with locked-in long-term contracts.