Anthropic and Elon Musk have not, historically, gotten along — Musk has sparred publicly with the company and with xAI's chief rival ambitions. Which makes it a genuinely strange detail that Anthropic's own confidential IPO paperwork shows the company has committed to paying Musk's SpaceX up to $84.5 billion for computing capacity through 2029. A Wall Street Journal profile that circulated widely on October 10, 2026 explains how that happened, and who inside Anthropic made it happen.
Quick summary: Anthropic agreed, in a deal first disclosed via SpaceX's own IPO filing in May 2026, to pay SpaceX $1.25 billion a month through May 2029 for exclusive access to more than 220,000 Nvidia GPUs at xAI's Colossus 1 data center near Memphis, Tennessee — originally built for xAI's own model training. Anthropic's own confidential IPO prospectus, dated around September 29, 2026 and reviewed by Reuters, showed that total commitment has grown to up to $84.5 billion — nearly double the original figure. A Wall Street Journal profile picked up widely on October 10, 2026 reports that Anthropic cofounder and chief compute officer Tom Brown — described as a Republican and the company's main liaison to both the Trump administration and to Musk — brokered the deal's expansion, and separately helped end a June 2026 standoff in which the White House had required Anthropic to shut down two of its models over safety concerns.
The original deal: renting out Musk's own training cluster
The arrangement traces back to a cloud-services agreement Anthropic and SpaceX entered into around May 2026. Under it, Anthropic agreed to pay $1.25 billion a month through May 2029 for exclusive access to Colossus 1, a data center near Memphis that xAI originally built to train its own models — including more than 220,000 Nvidia GPUs. That works out to close to $45 billion across the full term of the contract, and it was first made public through SpaceX's own S-1 filing as the company prepared for its IPO (Source: DataCenterDynamics).
The number nearly doubled by this fall
Anthropic is separately preparing for its own IPO — one reportedly targeting a valuation as high as $2 trillion. A confidential prospectus tied to that process, dated around September 29, 2026 and reviewed by Reuters, disclosed that Anthropic's total committed spend with SpaceX/xAI for compute has grown to as much as $84.5 billion — nearly double the original roughly $45 billion figure, with the monthly rate still described as around $1.25 billion (Source: BigGo Finance). What exactly changed — additional data center capacity, an extended term, or both — wasn't fully detailed in the reporting available at publication; this piece will follow up if a clearer breakdown becomes public.
Who is Tom Brown, and why did this need a broker
The Wall Street Journal profile, summarized widely on October 10, 2026, centers on Tom Brown, an Anthropic cofounder who now serves as its chief compute officer. Brown is described in that reporting as a Republican and as Anthropic's primary point of contact with both the Trump administration and with Musk personally — a notable role inside a company whose public identity has leaned heavily on AI-safety caution. Reporting describes Brown meeting Musk at SpaceX in March 2026 and arguing that Anthropic was less ideologically opposed to Musk's views than Musk apparently believed, a conversation that reportedly helped pave the way for the compute arrangement. Separately, Brown is credited with helping resolve a tense standoff from June 2026, in which the White House had required Anthropic to shut down two of its models over safety concerns, by convincing officials at the Commerce Department, the Pentagon, and the White House that the company had since tightened its safeguards (Source: Shopifreaks).
Why this is surfacing now
None of this is really a new partnership forming — the core SpaceX compute arrangement has been public since May. What's new is the scale of the number revealed through IPO-preparation disclosures, and the first detailed account of the relationship-building and political navigation behind it. Both are a fairly ordinary part of any company's pre-IPO disclosure process: prospectuses require naming major contractual commitments, and a company the size Anthropic is aiming to be — a roughly $2 trillion target valuation — has an enormous compute bill to account for regardless of which chip supplier it uses. The specific optics here — a safety-focused AI lab paying its most vocal corporate critic's company tens of billions of dollars, brokered partly through political relationships — are what make it a distinct story from a routine compute contract, not the dollar figure alone.
01How much is Anthropic actually paying SpaceX?
Roughly $1.25 billion a month through May 2029 for compute capacity at xAI's Colossus data center. The total committed figure, per Anthropic's own IPO prospectus reviewed by Reuters, has grown to up to $84.5 billion — up from an original estimate of roughly $45 billion when the deal was first disclosed in May 2026.
02Why would Anthropic pay a company tied to Elon Musk, given the public friction between them?
Reporting attributes the deal partly to Anthropic's need for compute capacity at scale and partly to relationship-building led by cofounder Tom Brown, who reportedly convinced Musk that Anthropic wasn't as ideologically opposed to him as Musk believed. The business logic is straightforward: SpaceX/xAI's Colossus cluster has GPU capacity Anthropic needs, regardless of the companies' public rivalry.
03What was the June 2026 safety standoff mentioned in the reporting?
Per the Wall Street Journal's reporting, the White House required Anthropic to shut down two of its models over safety concerns in June 2026. Tom Brown is credited with helping resolve that standoff by convincing the Commerce Department, the Pentagon, and the White House that Anthropic had tightened its safeguards since.
Whatever you make of the politics behind any single lab's compute deals, the practical question for most people is simpler: is the model you're paying for still worth it. LumiChats lets you compare Claude against 40-plus other leading models side by side, so you can judge the actual output rather than the corporate relationships behind it.
