Anthropic-SpaceX vs OpenAI Stargate vs Google-Broadcom (May 2026)
Anthropic-SpaceX vs OpenAI Stargate vs Google-Broadcom (May 2026)
Anthropic announced its 300+ megawatt Colossus 1 lease with SpaceX on May 6, 2026 — instantly the largest immediately-available frontier compute deal of the year. It joins OpenAI’s Stargate program and Google-Broadcom’s 5GW Anthropic deal as the megadeals reshaping AI compute in 2026. Here’s the cheat sheet.
Last verified: May 10, 2026
The three at a glance
| Deal | Anthropic-SpaceX Colossus 1 | OpenAI Stargate | Google-Broadcom (Anthropic) |
|---|---|---|---|
| Announced | May 6, 2026 | 2025, expanding through 2026 | 2025-2026 |
| Capacity | 300+ MW, 220K+ GPUs | $500B+ multi-year program | 5 GW (online 2027) |
| Hardware | NVIDIA H100/H200/GB200 | Mixed; custom + NVIDIA | Custom Broadcom silicon |
| Online | Now (operational) | Phased 2026-2028 | 2027 |
| Location | Memphis, TN | US sites (TX, NM others) | TBD US sites |
| Customer | Anthropic | OpenAI | Anthropic |
| Provider | SpaceX (post-xAI merger) | Oracle, SoftBank, others | Google + Broadcom |
What each deal actually is
Anthropic-SpaceX Colossus 1: instant frontier capacity
The Colossus 1 data center in Memphis, Tennessee was originally built by xAI to train Grok. After the SpaceX-xAI merger, xAI moved its training to the larger Colossus 2 facility, leaving Colossus 1 immediately available. Anthropic leased the entire facility in May 2026.
What Anthropic gets:
- 300+ megawatts of operational compute capacity, today.
- 220,000+ NVIDIA GPUs — H100, H200, and GB200 accelerators.
- Immediate availability at a moment when Anthropic’s other compute commitments (Amazon, Google-Broadcom, Microsoft, Fluidstack) deliver capacity over the next 6-24 months.
The customer-visible effects landed within hours of the announcement: Claude Code rate limits doubled across Pro/Max/Team/Enterprise plans, peak-hour reductions for Pro and Max accounts ended, and Opus API rate limits rose considerably.
The forward-looking note: Anthropic and SpaceX both mentioned exploring multi-gigawatt orbital AI compute capacity — using Starship launch cadence to put data centers in orbit. Best-case timelines put first pilot capacity in the late 2020s; this is R&D, not 2026 capacity.
OpenAI Stargate: $500B+ multi-year buildout
Stargate is OpenAI’s purpose-built AI infrastructure program. The headline number — $500 billion over multiple years — sums commitments from OpenAI, Oracle, SoftBank, and others to build dedicated AI data centers across the US.
The shape:
- Phased delivery through 2026-2028+.
- Multiple US sites including Texas, New Mexico, others.
- Mixed hardware — NVIDIA at first, custom silicon and accelerator partnerships expanding.
Stargate is the structural answer to “where does the next decade of OpenAI compute live.”
Google-Broadcom (Anthropic): custom-silicon megadeal coming 2027
Anthropic’s deal with Google + Broadcom commits 5 gigawatts of compute capacity coming online in 2027, using custom Broadcom-designed silicon. It’s the structural counterpart to the immediate-capacity SpaceX deal — long-tenor, custom hardware, locked through the late 2020s.
Combined with Anthropic’s $30B Microsoft-NVIDIA Azure deal and $50B Fluidstack US infrastructure agreement, Anthropic now has multi-gigawatt frontier compute committed through the end of the decade.
What this means for AI customers
Rate limits are easing. May 6, 2026’s doubled Claude Code rate limits are the leading indicator. Expect more capacity-driven announcements through 2026-2027 as Microsoft, Amazon, and Google-Broadcom capacity comes online.
Pricing has more room. Once supply constraints ease, frontier model pricing has more room to compete. Expect competitive pressure on Claude Opus 4.7, GPT-5.5, and Mythos pricing in H2 2026.
Concentration risk is real. Memphis (Colossus 1), Texas (Stargate), Tennessee/Oregon (Google sites) are increasingly single points of failure for global AI infrastructure. Production workloads need multi-provider fallbacks; uptime SLAs deserve scrutiny.
Power grid stress is now AI policy. 300 MW is the size of a small city’s load. Multiple deals at this scale will draw regulatory attention on grid capacity, water for cooling, and climate impact. Expect 2026-2027 policy moves on AI compute siting.
Decision implications for builders
You’re building on Claude (API or Cursor / Claude Code / Cline / OpenClaw). → The May 6 capacity unlock is real. Plan to use the higher rate limits. Set timeouts and retry logic that take advantage of the new headroom.
You’re building on OpenAI (GPT-5.5, GPT-Realtime-2). → Stargate’s phased delivery means capacity expansions through 2026-2028. Watch for OpenAI rate-limit announcements; they’ll come on a Stargate-site cadence.
You need multi-provider redundancy for production. → This is the right time. Frontier providers are racing on capacity, but each individual site is a SPOF. Architect provider routing now (LiteLLM, OpenRouter, in-house router) before the next regional outage.
You’re cost-optimizing. → The capacity buildout will pressure prices in H2 2026. Don’t lock multi-year pricing today; quarterly contracts give you re-negotiation room as supply expands.
What changed in May 2026
- May 6: Anthropic announces 300+ MW Colossus 1 lease with SpaceX.
- May 6: Claude Code rate limits doubled; peak-hour reductions removed; Opus API rate limits raised.
- May 6: Joint orbital data center messaging from Anthropic and SpaceX.
- Pre-May: OpenAI Stargate phased delivery continues. Anthropic’s Google-Broadcom 5GW deal targets 2027 online.
What to watch next
- Capacity-driven pricing announcements from Anthropic, OpenAI, Google through Q3 2026.
- Stargate site openings — each new Stargate site means OpenAI rate limit and pricing flexibility.
- Power-grid permitting fights — Memphis, Texas, Oregon utility hearings will shape future siting.
- Orbital compute pilots — SpaceX-Anthropic announcements through 2027.
- Concentration regulation — does the FTC or DOJ scrutinize single-customer-takes-entire-facility deals like Colossus 1?
Related reading
- Anthropic $900B funding round
- Cloudflare AI-first restructuring 600% explained
- Grok after SpaceX merger: orbital data centers
Last verified: May 10, 2026 — sources: Anthropic.com news (higher limits with SpaceX), Forbes, AIBusiness, MorningStar/MarketWatch, MLQ.ai, x.ai compute partnership note, Anthropic $50B US infrastructure announcement, Anthropic Google-Broadcom partnership announcement.