Anthropic strikes $13.7B compute deal with Trump-linked Rum Group
Source: The Information, 2026-09-14, by Valida Pau. Article · Raw: RSS Date: 2026-09-14
TL;DR
Anthropic has signed a six-year, $13.7 billion computing agreement with Rum Group, a company with roots in social media and long-standing ties to the Trump administration. It is the latest in a year of cloud deals spanning Google, SpaceX and smaller neoclouds including Nscale (a $45B agreement in August). The aggregate now stands at at least 14.8 gigawatts of committed capacity, costing as much as $517 billion over the next decade, with the Rum deal on top of that.
Why it belongs on the compute-economics thread
Three things make this more than another large number.
1. The counterparty type keeps drifting away from hyperscalers. The sequence runs Google, then SpaceX, then Nscale, now a social-media company with political ties. Each step widens the pool of entities Anthropic is willing to source capacity from, which is what a genuinely supply-constrained buyer looks like. The relevant question for a six-year contract is not price, it is whether the counterparty can actually deliver power and silicon on schedule, and a social-media firm has no track record on either.
2. The political adjacency is priced into nothing and observable in everything. This lands in the same week that Amodei's pacing essay asked for federal coordination on frontier AI, that Anthropic, OpenAI and Google were reported to be quietly discussing a private standards body, and that President Trump publicly rejected a slowdown. A $13.7B commercial relationship with a politically-connected counterparty is a fact about the regulatory landscape as much as about compute.
3. It stress-tests the guarantee structure this wiki already flagged. The 09-12 SemiAnalysis analysis documented roughly $530B of off-balance-sheet guarantees in the Nvidia orbit, including $279B of supply commitments described as primarily memory through fiscal 2029. Anthropic's own guidance to investors runs to roughly $180B of revenue through 2029 against commitments that may reach $517B. Adding $13.7B does not change the ratio much; it changes how many independent parties are inside it.
How this relates to what the wiki already knows
The compute economics page has tracked the circular-financing thread: Nvidia discussing up to $10B into Anthropic's IPO at a roughly $2T valuation, most of which returns as chip orders. This deal is the other end of that circuit, capacity purchased forward against revenue that does not exist yet.
And it is the hardware counterweight to the day's other industry story. Nadella's learning-loop argument (09-14) says the durable enterprise asset is the loop you own, not the model you rent, and that firms should avoid dependence on a single model provider. That may well be right at the application layer and still leave the entire margin downstream of whoever holds 14.8 GW of contracted capacity. Portability of the loop and concentration of the substrate are not in conflict; they describe two different layers, and the value is accruing to the lower one.
Separately, The Information reported the same week that Nvidia's revenue concentration is tightening: three customers accounted for 44% of total sales in the first half of the fiscal year ending in July, up from two customers at 36% last year, against zero customers above the 10% disclosure threshold as recently as fiscal 2023. That is the same structure viewed from the supply side, and it is why Jensen Huang has reason to keep investing in neoclouds and AI firms: he is manufacturing customers.