Anthropic’s $13.7 Billion Bet on Trump-Linked Compute

Anthropic has locked in another massive computing agreement. This one, worth $13.7 billion over six years, comes from an unexpected source. The buyer? The AI company behind Claude models. The seller? RUM Group, the rebranded entity that operates Rumble and hosts Donald Trump’s Truth Social platform.

The deal surfaced quietly at first. On August 24, RUM Group announced a major contract with an unnamed U.S.-based cloud customer for GPU access and related services. Shares jumped more than 8% in premarket trading that day. Reuters reported the basics. The facility sits in Maysville, Georgia. It remains under development. Payments arrive in three tranches. The final one depends on client-approved delivery dates.

But the customer stayed hidden. Until now. On September 13, The Information identified Anthropic as the counterparty. The report sent ripples through tech and political circles alike. And the structure? It includes warrants allowing Anthropic to purchase up to 50.8 million shares of RUM Group for one cent apiece. Those options vest gradually as the compute commitments are met.

RUM Group didn’t build its AI infrastructure from scratch. The company, formerly known for video streaming with a free-speech bent, acquired German AI cloud provider Northern Data in June. That move created a dedicated infrastructure arm. It aims to ride the wave of exploding demand for graphics processing units. The Maysville site represents one of its first major U.S. builds.

For Anthropic the timing feels strategic. The company has spent the past year signing compute deals at a breakneck pace. Its total commitments now exceed $500 billion across the next decade. That figure dwarfs earlier projections of $180 billion in server leasing through 2029. Analysts at Measured AI put the firm’s contracted U.S. capacity pipeline above 15 gigawatts. The tally spans 15 campuses, multiple chip types and five distinct contracting approaches.

Amazon and Google together supply the largest slice. More than 11 gigawatts. Over $300 billion in value. Microsoft contributes another gigawatt through Azure for roughly $30 billion. SpaceX’s Colossus cluster in Memphis adds hundreds of thousands of Nvidia processors. Recent additions include long-term pacts with neoclouds such as CoreWeave, Lambda, Nscale and Fluidstack. One Nscale agreement alone covers 460 megawatts of upcoming Nvidia Vera Rubin chips for about $45 billion.

Yet capacity hunger persists. NVIDIA chief Jensen Huang called Anthropic “compute limited, factory limited” as recently as March. Tom Brown, Anthropic’s co-founder and compute chief, has echoed the pressure. “Access to compute is central to keeping Claude at the frontier and meeting demand from our customers,” he said in a July statement tied to an AMD partnership. That deal could see Anthropic take tens of billions in Instinct MI450 servers. AMD even pledged up to $5 billion in investment.

The RUM pact adds diversity. It brings in a provider with clear ties to the current administration. RUM Group’s connections run deep. Peter Thiel and J.D. Vance have backed related ventures. The company’s anti-censorship stance aligns with Trump-world rhetoric. For an organization that has lobbied for AI oversight and voluntary slowdowns on frontier development, the optics raise eyebrows. Dario Amodei, Anthropic’s chief executive, recently joined calls with Sam Altman and Elon Musk to pace AI progress. Yet the spending continues unabated.

Critics see contradiction. Safety rhetoric meets aggressive expansion. Supporters call it pragmatism. In a market where rivals like OpenAI and xAI race for dominance, falling behind on infrastructure equals falling behind on capabilities. Claude’s enterprise products, including coding assistants and internal tools, require ever-larger clusters. Demand from business customers has pushed annualized revenue toward $65 billion. An initial public offering looms. Strong compute assurances could lift valuation talks.

RUM Group gains validation too. A blue-chip AI customer lends credibility to its pivot. The warrant structure ties the partners together. If Anthropic scales usage, it gains cheap equity upside. RUM collects steady, high-margin revenue from GPU services. The three-tranche payment schedule reduces risk for both sides. Early phases deliver immediate capacity. Later ones hinge on successful build-out.

This agreement fits a pattern. AI labs now pursue heterogeneous supply. They mix hyperscalers, former crypto miners, neocloud startups and even energy-focused players. The goal is resilience against shortages. Power constraints, chip allocation queues and construction delays plague the sector. Georgia’s site offers another geographic hedge. Southern states have courted data centers with tax breaks and faster permitting.

Energy implications loom large. Fifteen gigawatts rivals the output of multiple nuclear plants. Local communities worry about water use, grid strain and emissions. Former EPA officials recently flagged 30 federal actions since January 2025 that they say worsen data-center pollution risks. The Verge covered those concerns. Political alliances may smooth regulatory paths. Or they may invite fresh scrutiny.

Anthropic’s approach stands out. Unlike some peers, it avoids owning most physical plants. Instead it signs long-term leases and managed-service contracts. The RUM deal follows that model. Access without the full capital burden. Yet total exposure remains enormous. Some estimates put the full stack, including chips, power and facilities, near half a trillion dollars over ten years.

Investors appear to approve. RUM shares reacted positively to the initial announcement. Anthropic’s secondary-market valuation has held firm despite broader tech volatility. A successful IPO could provide fresh capital. It might also crystallize the cost of staying competitive. Because one truth dominates. Compute wins races. Models improve only as fast as the silicon and electricity allow.

So the deals keep coming. Lambda. Nscale. Now RUM Group. Each adds another brick to a towering infrastructure moat. Whether that moat supports safety-focused development or simply fuels faster iteration remains the open question. Anthropic insists both goals can coexist. Time, and Claude’s next leaps, will test the claim.

Recent coverage adds color. TECHi detailed the warrant mechanics and tranche math on September 13. TradingKey explored IPO implications the next day. And Measured AI’s exhaustive August analysis of Anthropic’s 15-gigawatt pipeline shows this latest pact as one piece of a far larger mosaic. The mosaic grows weekly.

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top