Anthropic is spending money like a nation-state at war — and it still might not be enough
Anthropic signed a deal with SpaceX to use all the compute capacity at their Colossus 1 data centre: over 300 megawatts and 220,000 NVIDIA GPUs. That’s just the latest in a spending spree that now includes commitments totalling hundreds of billions of dollars across multiple providers. The AI compute arms race has moved past ‘aggressive’ and into ‘genuinely concerning’ territory.
Let’s run the numbers, because they’re borderline absurd:
- SpaceX deal: 300+ MW, 220,000 GPUs from Colossus 1 (available within the month)
- Amazon deal: Up to 5 GW agreement, nearly 1 GW by end of 2026
- Google/Broadcom deal: 5 GW agreement, coming online 2027
- Microsoft/NVIDIA deal: $30 billion of Azure capacity
- Google Cloud commitment: $200 billion over five years — over 40% of Google’s entire cloud revenue backlog
- $50 billion US infrastructure investment with Fluidstack
- Lightspeed-led round: Potential $50 billion valuation — could exceed OpenAI
- Losses this year: $11 billion. Profitability? Not until 2028.
Oh, and Alphabet itself is investing another $40 billion in Anthropic ($10B now, $30B on performance milestones).
The SpaceX angle
What makes the SpaceX deal notable isn’t just the scale — it’s who the partner is. Elon Musk owns SpaceX. Elon Musk co-founded OpenAI and is currently suing OpenAI in a high-profile trial. Now SpaceX is providing compute for OpenAI’s biggest competitor.
The politics are messy. The compute needs are desperate enough that Anthropic partnered with a company whose founder is simultaneously trying to sue their main rival into oblivion.
The circular money problem
Anthropic’s spending creates a closed loop that looks increasingly fragile:
- Google invests $40 billion in Anthropic
- Anthropic commits $200 billion to Google Cloud (more than 40% of Google’s backlog)
- Google buys chips from NVIDIA/Broadcom
- Everyone announces record AI capex
- The Financial Stability Board warns about private credit exposure to AI
The UK-based Financial Stability Board said AI accounted for more than a third of private credit deals in 2025. They warned of a “sharp correction in asset valuations” — financial regulator speak for “this might all be house of cards.”
Anthropic projects $18 billion in revenue in 2026, $55 billion in 2027, $102 billion in 2028, and $148 billion in 2029. The growth rate baked into those numbers is unprecedented. If they miss by even 20%, the debt-fuelled infrastructure stack starts looking shaky.
Usage limits raised — for now
For actual users, the compute deals mean something tangible: Anthropic is raising usage limits for Claude Code and the Claude API. Pro, Max, Team, and Enterprise plans all get higher rate limits. Peak hours limit reductions on Claude Code are gone. API rate limits for Claude Opus models have been increased substantially.
But the longer question is: can Anthropic sustain this? Their own timeline says profitability is two years away. That’s two years of burning through $11 billion per year, financed by a complex web of hyperscaler credit, private debt, and round after round of fundraising.
NZ angle: data centre implications
For New Zealand, the Anthropic/SpaceX deal underscores something important: the global scramble for compute is so intense that companies are turning to space exploration firms for data centre capacity. The Datagrid AI factory near Invercargill — 280 MW, 6% of NZ’s national demand — suddenly looks like a modest facility by comparison.
If Anthropic is willing to partner with SpaceX for immediate GPU access, what does that mean for smaller data centre markets like New Zealand? It suggests that as the hyperscale players lock up the Tier 1 data centre capacity, secondary markets become increasingly attractive for overflow workloads — but also that the power demands will keep growing, and NZ’s renewable energy advantage becomes a critical bargaining chip.
🔍 THE BOTTOM LINE
Anthropic’s SpaceX deal and $200 billion Google Cloud commitment are signs of an industry running at max speed toward an uncertain destination. The company will lose $11 billion this year, has no clear path to profitability before 2028, and is financing its expansion through a web of hyperscaler credit, private debt, and circular deals where the same companies invest in each other’s infrastructure. When the Financial Stability Board is warning about AI private credit exposure, it’s worth paying attention. The compute arms race might produce the most useful AI ever built, or it might produce the most expensive financial correction in tech history.
❓ FAQ
Q: Why would SpaceX sell compute to Anthropic when Elon Musk is suing OpenAI? A: Money. SpaceX doesn’t care about the OpenAI lawsuit. They have a data centre with GPU capacity and Anthropic will pay top dollar.
Q: Is Anthropic’s spending sustainable? A: Their own financials show $11B losses in 2026 and 2027 with no profit until 2028. This is fuelled by venture capital, hyperscaler investment, and private debt. The FSB is worried.
Q: How does this affect me as a Claude user? A: Short term: higher rate limits, better service. Medium term: depends on whether Anthropic survives the infrastructure cost crunch.
Q: Will this drive up data centre costs globally? A: Yes. The competition for GPU capacity is pushing up power, land, and construction costs everywhere, including New Zealand.
📰 SOURCES
- Anthropic Blog — “Higher usage limits for Claude and a compute deal with SpaceX”
- Bloomberg — “Anthropic, SpaceX Sign Deal to Boost AI Computing Power”
- NBC News — “Anthropic and SpaceX announce major partnership”
- The Information — “Anthropic commits $200B to Google Cloud”
- Computing UK — “Anthropic commits to $200bn spending on Google Cloud”
- TechCrunch — “Anthropic and OpenAI both launching enterprise AI joint ventures”
- The Guardian — “Global finance watchdog warns over private credit fuelling AI boom”
