# data center revenue — X 热门讨论 (2026-09-29 20:14 UTC)
## @marsrepublica (samuel) · 09-29 16:29 · ♥60 ↻6 💬2 Anthropic’s $2T+ IPO just got announced, and it’s even bigger news for SpaceX.
Anthropic is one of the fastest-growing companies in history: roughly $9B ARR at the end of 2025, $47B in May, and $65B+ by July. Investors expect $100–120B by year-end, as the world’s leading AI company at the frontier of RSI and approaching AGI.
That growth rate is the entire story.
The IPO’s core sell depends on Anthropic maintaining that growth and lead while outrunning its burn. Unlike SpaceX, whose infrastructure dominance gives investors confidence to back its AI ambitions, Anthropic’s moat is purely software in a hypercompetitive market where rivals are only weeks to months behind and could overtake it quickly. Maintaining its lead and growth is therefore existential to a $2T valuation.
The problem: sustaining anything close to this growth requires an unprecedented amount of compute.
Energy and infrastructure constraints already leave vast data-center projects waiting, while political opposition is rising through populist, bipartisan movements gaining traction across states to pause projects. Major AI labs already face these constraints.
And we’ve already seen what happens when Anthropic hits a compute limit.
Earlier this year, Anthropic couldn’t serve its latest model and was near dead halt until a rapid SpaceX deal for 325K GPUs from Colossus saved them.
This will only get worse in the coming era. Good-enough models are reaching mass distribution, while RSI is increasingly dependent on compute gains.
Anthropic is already attacking the problem from every direction.
Amazon has agreed to provide up to 5GW of new capacity, including nearly 1GW of Trainium2 and Trainium3 capacity by the end of 2026. Anthropic also has a 5GW agreement with Google and Broadcom beginning in 2027, a strategic partnership with Microsoft and NVIDIA involving $30B of Azure capacity, and a $50B American AI infrastructure investment with Fluidstack.
But there is a difference between having agreements for future capacity and having a company that can deploy enormous amounts of compute immediately.
That is where SpaceX becomes different.
SpaceX has been deploying compute at extraordinary speed: about 400MW+ per quarter, potentially 880MW+ in Q4 and up to 1GW+ per quarter in a fully locked-in scenario for its 10GW 2027 build goal.
By December, it could have 500.8K–720.8K GB300s, potentially reaching 634K–854K by April if its six-month undisclosed client deal isn’t renewed.
Elon has also said SpaceX is targeting roughly 10GW of nameplate compute by the end of 2027, with projects totaling 20GW at the power-and-cooling level, although he does not expect all 20GW to come online.
That is exactly the deployment velocity Anthropic needs.
I expect Anthropic to take some SpaceX’s Q4-to-Q1 extra capacity and sign a multi-GW deal covering the rest of the year, paying per deployment as each tranche of compute comes online.
SpaceX can finance the main GPU cost through NVIDIA × bank financing deals, which become increasingly viable as banks and investors recognize compute as the actual AI moat, while covering build costs with its own cash on hand ($80B+ generally estimated by year-end).
The structure is straightforward:
Anthropic commits to the demand.
SpaceX finances and builds the infrastructure.
Compute comes online in stages.
Anthropic pays as each deployment becomes operational.
SpaceX gets a massive contracted customer to support its build and generate unprecedented revenue, and Anthropic avoids the hard problem of building compute itself, which starting this late would be a waste for its goal vs the short terrestrial window as it won’t matter long term as orbital is the future and guess who provides that? SpaceX!
Expect a major deal soon! Bookmark this. https://x.com/marsrepublica/status/2104971879039525091
## @danielisdizzy (Daniel) · 09-29 19:04 · ♥54 ↻1 💬9 A deal between $IREN and $GOOGL will happen.
$GOOGL is expected to reach ~32 GW of data center capacity by 2030, up from ~5 GW today.
Roughly 25% of that capacity is expected to come from leased infrastructure.
No doubt a meaningful portion of Google’s leased capacity will come from $IREN.
$IREN is one of the few companies building AI infrastructure at the scale hyperscalers need.
$IREN has made it clear that it doesn’t want to rely exclusively on hyperscalers and frontier AI labs. It also wants to serve AI developers and enterprises and become a full-stack AI cloud platform.
But $IREN still needs massive hyperscaler deals.
These multi-billion-dollar contracts can provide significant upfront payments and, more importantly, give $IREN highly creditworthy counterparties backing long-term contracted revenue.
That makes the cash flows more bankable and allows $IREN to raise debt at a lower cost to fund its massive expansion.
Customer diversification builds the platform.
Hyperscaler deals finance the scale.
$GOOGL x $IREN is coming. https://x.com/danielisdizzy/status/2105010821487001837
## @JCOOPsolutions (JCOOP) · 09-29 19:04 · ♥36 ↻0 💬0 $ZONE | $CBRS POWERING THE NEXT WAVE OF AI
Zone Frontier’s 10-year agreement with Cerebras Systems could put ZONE at the center of a pressing AI challenge: securing the power and space to run next-generation compute.
▪️ $800M in projected contract value over the initial term. ▪️ 100% pre-leased Minnesota data center campus. ▪️ 55 MW of planned utility capacity at full buildout. ▪️ Zone expects to own nearly 80% of the project and begin generating revenue in Q1 2027.
If Zone delivers, this could be a blueprint for bringing major AI customers online at scale. Watch the buildout and the infrastructure behind AI may be the story to follow.
Communicated - Disclaimer: https://t.co/jGsGLvPr7x https://x.com/JCOOPsolutions/status/2105010758731907381