HUMAIN agreed to keep Chinese frontier training off its US-approved compute

Divide the roughly $1 billion DataVolt is spending at NEOM by the 100 megawatts it buys, and the build comes to about $10 million a megawatt ^[12]. Carry that rate to the 360-megawatt ambition and the finished site implies something near $3.6 billion ^[13]. Against that, the $5.3 billion AWS alone has committed to data centers in the kingdom is roughly five times the first DataVolt phase ^[14]. The facility is designed to export compute ^[5], so its customers sit outside Saudi Arabia. Taub's essay does not state a price for that compute ^[16].

The condition HUMAIN accepted governs where a Chinese model is trained, not what HUMAIN builds on top of one ^[3]. The MiniMax work shown at LEAP is a deployment relationship ^[2].

Taub's case for leverage is a bidding argument, and he sets out both halves: American restrictions push Riyadh toward Chinese alternatives, and Chinese shortfalls in frontier performance push it back toward Washington ^[6]. Both halves need the performance gap to persist. If Chinese frontier models reach parity, the second push stops and Riyadh is down to one bidder. If Washington's approval conditions reach terms HUMAIN will not sign, the American half stops mattering, because there is no approved capacity left to fence.

The two capitals are also buying different things. Taub argues that Washington now treats chip policy as foreign policy, with access to Nvidia's most advanced processors working less like a commercial transaction than like access to sensitive military technology ^[10], while Beijing uses open models and a stronger domestic stack to reach markets its hardware cannot ^[11].

What Riyadh collects, on Taub's account, is a growing share of the infrastructure other people's models run on, and access it can trade for investment, localization commitments and technology transfer ^[7]. He wrote that the Gulf "spent decades learning to turn control over energy into geopolitical leverage" ^[8]. Taub is the founder and managing partner of Hetz Ventures, an Israeli early-stage firm that invests in data and AI infrastructure ^[9].

The record here supports a narrower claim than a world without blocs: one state-backed champion, one American build, one Chinese model, one written training boundary. Taub's wider read is that the competition produces a handful of swing states playing both sides ^[15]. In my view that position holds only while the frontier gap does, and the half of his mechanism that sends Riyadh back to Washington is the half that expires first ^[6].