A live model of the Western AI compute market, 2026–2040. Silicon, power, and datacenters (including orbit). Not the software on top. Three dials, 2,197 pre-solved worlds, 5,000 futures. You see the fleet, tightness, the dollar market, and the year orbit undercuts the ground.

We modeled the AI compute market from 2026 to 2040 as one connected system: what a delivered unit of compute costs by build route, how demand responds as prices fall, and where new capacity actually gets built. The model treats orbital datacenters as a build route inside the same framework as terrestrial, tests every assumption across its full range, and stress-tests the story under thousands of simulated futures. This Premium article is a presentation of the key takeaways.

An interactive research brief on how radiation affects AI compute in low Earth orbit. At 550 km, a modeled 1,000-satellite fleet sees tens of thousands of daily glitches, but error correction and recovery software keep downtime to about 8 minutes a year — while shielding mass drives nearly all of the modeled ~$82k annual radiation cost per satellite.

The sharpest technical objection to SpaceX's orbital AI plan is not power or heat. It is that modern AI models are designed to run across hundreds of interconnected chips... so we model how a typical 72 GPU NVDA rack would perform as a node in comparison. Forcing a frontier model into one satellite forgoes up to 44% of achievable output. The penalty is real but not permanent, better software, slimmer models, and models designed to fit the satellite could reduce it to near parity over time.

The sharpest technical objection to SpaceX's orbital AI plan is not power or heat. It is that modern AI models are designed to run across hundreds of interconnected chips... so we model how a typical 72 GPU NVDA rack would perform as a node in comparison. Forcing a frontier model into one satellite forgoes up to 44% of achievable output. The penalty is real but not permanent, better software, slimmer models, and models designed to fit the satellite could reduce it to near parity over time.

SpaceX prices its IPO this week. At $135 a share the deal carries roughly $1.77 trillion of equity value and raises about $75 billion, the largest listing on record; SPCX begins trading on NASDAQ on June 12. Our first article opened up the model and walked through how it works, but it held back on numbers. This piece puts the IPO opening price on the page.

We have been assembling the SpaceX ‘Gigamodel’ for several months. It is a ~400-variable undertaking, projecting the three engines that matter, launch, Starlink, and AI. We are confident it is the deepest independent financial model of SpaceX available. We are continuing to fine-tune it and will soon run the Monte Carlo simulation to derive valuation output ranges. (This is to follow in the next few days, ahead of the listing.)