The Compute 100 — Wednesday June 10, 2026

Orbital compute moved from research phase to manufacturing commitment on June 9. The category has been technically real since late 2025—Starcloud's first satellite has run NVIDIA H100 workloads in orbit, and Kepler Communications has operated multi-GPU nodes since January—but SpaceX's announcements yesterday mark a different order of magnitude. The company released full hardware specs for its AI1 satellite—120–150 kW of compute payload per unit, a 70-meter solar wingspan wider than a Boeing 747, and an interchangeable chip module designed so successive GPU generations can be swapped without redesigning the satellite bus—and unveiled the Gigasat factory in Bastrop, Texas: 11 million square feet, more than ten times the size of SpaceX's current largest manufacturing complex, targeting 1 GW per year of orbital AI compute by late 2027. Google's $920 million-per-month compute agreement, signed June 5, establishes what market-rate orbital compute looks like at hyperscaler volume. The constraint that makes or breaks the 2027 target is the Gigasat factory's solar manufacturing ramp; SpaceX says solar production is already under construction while the satellite building is about to break ground.

Crusoe paused Project Jade, its 1.8-gigawatt campus in Cheyenne, Wyoming, at the request of an undisclosed customer. Developed with Blackstone-backed Tallgrass and designed to scale to 10 GW, the site had cleared local permitting with construction slated this year; the power plan called for 900 MW of Bloom Energy fuel cells plus grid supply. Crusoe holds nearly 5 GW in contracted capacity and a pipeline exceeding 20 GW; its Abilene Stargate campus for OpenAI via Oracle is unaffected. The pause is distinct from speculative lease cancellations: a contracted, site-approved build deferred at customer request signals that hyperscalers now exert direct pacing control over committed infrastructure. (Bloomberg)

TensorWave raised $350 million in a Series B at a $1.55 billion valuation, co-led by AMD and Magnetar Capital, the Wall Street Journal reported Tuesday. The company operates one of the largest all-AMD GPU clouds, running AMD Instinct MI325X processors, and positions explicitly as an alternative to Nvidia-based infrastructure; CEO Darrick Horton says TensorWave was created to "restore competition to the market." The $1.55 billion valuation nearly quadruples the roughly $400 million mark from a year ago. With AMD co-leading the round, the investment is as much a strategic signal about AMD's commitment to datacenter GPU market share as it is a bet on TensorWave specifically. (WSJ)

Anthropic released Claude Fable 5 on June 9—its first publicly available Mythos-class model. Priced at $10 per million input tokens and $50 per million output (double Opus 4.8), though Pro, Max, Team, and seat-based Enterprise plans get access at no extra cost through June 22. A safety classifier triggers Opus 4.8 fallback in under 5% of sessions. The headline benchmark: Stripe reported Fable 5 completed a 50-million-line Ruby codebase migration in one day that would have taken a team two months. Mythos 5 is simultaneously deploying in restricted access via Project Glasswing to vetted cybersecurity partners. The $50/M output price doubles per-token model revenue at scale, directly expanding the GPU-hours required per dollar of inference spend—a structural tailwind for inference infrastructure providers.

Shares consolidated a 5% surge from Monday after Bank of America analyst Tal Liani raised his price target to $280 (from $240), citing "strengthening compute demand," following a fireside at BofA's Global Technology Conference. The underlying news is substantive: Nebius committed £1.7 billion to build out four UK AI infrastructure sites to 65 MW by 2027 using Nvidia Blackwell Ultra systems, secured a 22 MW ten-year agreement with Kao Data at Harlow, and launched a Physical AI Living Lab with Nvidia to support UK and European robotics startups. Nebius is up more than 170% year-to-date; the UK expansion makes it one of the more concrete European AI infrastructure stories in the market.