Good Thursday —
Meta's $21B compute deal with CoreWeave trades hyperscaler flexibility for guaranteed capacity till 2032
The $21B figure brings CoreWeave's total contracted backlog with Meta to $35B. That's real, and the demand signal is unambiguous. But look at what else happened today: CoreWeave is raising $3B in convertible notes due 2032 and $1.25B in senior notes due 2031. The DDTL 4.0 facility it closed two weeks ago explicitly priced against Meta backlog as collateral. Interest expense hit $388M in Q4 2025 alone, with total debt near $29.8B.
This deal is as much a CoreWeave credit event as it is a Meta infrastructure story. The contract is the collateral narrative that makes the debt raise coherent.
The thing people will get wrong: this isn't Meta choosing external over internal. Meta has its own gigawatt-scale data center construction underway and entered a $27B joint venture to fund the Hyperion campus, covering 80% of costs through Blue Owl Capital. It's also deploying MTIA v2, its in-house accelerator. The company's CFO described third-party cloud spend as bridging constraints while it scales its own footprint. CoreWeave gets 2032 on the contract. The real question is how much of it converts. If Hyperion and MTIA v2 scale faster than Meta's inference demand grows, the back half of that commitment gets renegotiated or underutilized.
Know someone allocating to AI infra? They probably saw the CoreWeave headline without the analysis. Forward this email.
On Our Radar
Google keeps buying Xeons while building custom silicon.Intel holds the inference fallback position. If IPU roadmap slips, that fallback becomes permanent.
Appeals court upheld the block. Anthropic's government revenue path narrows, pushing it deeper into commercial while Palantir and Scale face less competition for defense contracts.
Another model on top of $100B in committed capex. The question isn't whether Muse Spark is good. It's whether on-device visual models can justify that spend.
OpenAI is anchoring on ads to justify capex. One analyst already called it unsubstantiated. If ads don't scale, inference margins compress. Watch quarterly engagement as the leading indicator.
Regime Snapshot
Compute (CRS): 65 — scarcity. More buyers than available capacity. Lead times extending.
Memory (MRS): 76 — Tight. Demand outstripping supply. No relief until new capacity comes online.
This is exactly the kind of structural shift the terminal tracks in real time — which companies are exposed, how the constraint map is shifting, and what the regime history says about timing.
We’re opening the private beta to a small group of investors and analysts. If you want the full picture, not just the daily snapshot:
— Teng