Good Friday —
OpenAI's pause reveals that Stargate's global rollout is energy-constrained and capital-constrained.
OpenAI paused Stargate UK citing energy costs and regulatory uncertainty. That framing is defensible. It's also missing the most interesting part of this story.
A Guardian investigation found the supposed site of Nscale's supercomputer had not started construction. The site remained a scaffolding yard, still under ownership by a different company, with no evidence of Nscale's ownership despite Nscale claiming purchase and the UK government having already issued a press release stating a £1.9B investment contract was signed. Say…what? That's a project that was operationally troubled before the energy cost argument ever entered the room.
The IPO pre-cleaning narrative is real, though. OpenAI has already cancelled Sora despite a $1B Disney deal, walked back Instant Checkout, and shelved its adult mode service MarketWise - all consistent with a company tidying up its balance sheet story ahead of a public listing.
OpenAI is projected to lose $14B in 2026, with annual burn reaching $57B by 2027 and breakeven not until 2030. CFO Sarah Friar reportedly isn't convinced they're ready for public market scrutiny at all. So yes, capital discipline is real.
Know someone allocating to AI infra? They probably saw the OpenAI headline without the analysis. Forward this email.
On Our Radar
TSMC's 35% YoY surge locks in tight foundry lead times through 2025. Samsung and Intel face a choice: compete on price or watch customers queue for capacity they can't match.
NVIDIA backing SiFive's $3.65B IPO bet signals a CPU fragmentation play. If SiFive gains traction, system builders lose the single-ISA standardization that keeps integration costs down.
SK Hynix accelerating M15X ahead of schedule keeps HBM prices elevated through 2025. GPU system costs stay high, and integrators who can't pass the increase watch margins compress.
A founder paying $3.65M annually per employee for Claude reveals Anthropic's pricing ceiling. The bet: raise API costs without losing whales. But Claude's inference cost still blocks mass adoption.
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.
Narratives Moving Today
Can China build chips without NVIDIA? — conviction 76, ▲11 pts this week.
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