In the same week it called on the entire industry to “slow down,” Anthropic unveiled staggering financial figures: over the 11 months ending August 2026, itscompute commitment reached $51.7 billion, covering 14.8 GW—nearly triple the prior estimate (approximately $18 billion, extending to 2029). This is the largest single bet in the AI industry to date ahead of IPO.
Who received the funding
- Amazon + Alphabet: ~11 GW, over $30 billion—undisputed leaders
- Microsoft: Over $30 billion, approximately 1 GW
- SpaceX Colossus: Approximately $4.5 billion
- AMD: A 2 GW order for MI450 chips
Additionally, The Information confirmed that Anthropic is the “anonymous customer” referenced in Rum Group’s six-year deal in Maysville, Georgia For $13.7 billionThe agreement, while Rum Group has yet to close its financing, imposes no upstream financing condition on Anthropic’s payment obligations—a clause only a company supremely confident about its IPO would dare sign.
“Slowing down” and “going all-in” are not contradictory
At first glance, CEO Dario Amodei’s long post last weekend calling for “limits on frontier AI” appears self-contradictory when followed immediately by disclosure of a half-trillion-dollar compute expansion. Yet a close reading reveals Amodei intends to slow down“capabilities”, not“deployment”—and the 14.8 GW of compute falls squarely under deployment. In other words, Anthropic’s strategy is “braking on capability development while accelerating commercialization.”
This is an arms race
Zooming out, this isn’t just Anthropic’s gamble—it’s an industry-wide arms race. In the same week, OpenAI disclosed holding $12.2 billion in committed capital plus a $4.7 billion revolving credit facility; Z.ai’s $5 billion funding round closed today, and ByteDance secured a $29.6 billion three-year syndicated loan. When top players compete for compute using “committed amounts” rather than “already-spent” figures, whoever secures power allocation first locks in their ticket to the next round of competition.
But let’s pour some cold water—517 billion is a “commitment,” not “spent”Most of it is slated for deployment after 2028, and Rum Group’s website does not even exist yet. The real litmus test is whether “nearly profitable for two consecutive quarters” can extend to four. Behind this figure lies a microcosm of the AI infrastructure arms race—more firsthand observations inside AI Dash.