Anthropic expands into xAI Colossus2 data center: GB200 chip-driven AI computing power arms race
Anthropic expands to xAI Colossus2 data center equipped with NVIDIA GB200 (Blackwell) chips. This is the second time that xAI has provided top computing power to Anthropic after Colossus1, marking a new stage of AI computing power infrastructure.
Core conclusion
On May 20, 2026, Tom Brown, head of Anthropic research, disclosed on X/Twitter: Anthropic will expand to xAI’s Colossus2 hyperscale data center, using NVIDIA GB200 (Blackwell architecture) chips. This is another major collaboration between the two companies at the AI infrastructure level after xAI provided all the computing power of Colossus1 to Anthropic.
Key Takeaways: -Initiator of the event: Tom Brown (Anthropic research leader, former OpenAI core member, first author of the paper "Scaling Laws for Neural Language Models") -Affected objects: AI training and inference infrastructure industry
- Core changes: xAI continues to provide top computing power resources to Anthropic, and AI computing power cooperation enters a new stage
Background: The past and present life of the Colossus series
xAI’s Colossus cluster, located in Memphis, Tennessee, is one of the largest AI supercomputers on the planet. Colossus1 deploys more than 100,000 NVIDIA H100 GPUs, all of which were previously assigned to Anthropic. Now Colossus2 is being expanded and will deploy NVIDIA's new generation Blackwell architecture GB200 chip.
The timeline of this collaboration is worth noting:
- Colossus1 (2024-2025): 100,000+H100, xAI training Grok series model
- Colossus1 → Anthropic (2025-2026): xAI transfers the entire cluster computing power to Anthropic for use
- Colossus2 (2026): Expanding, equipped with NVIDIA GB200
- Colossus2 → Anthropic (2026.5): Tom Brown confirms Anthropic will use GB200 computing power
Key Impact (by Dimension)
| Dimensions | Change | What it means to us | Recommended actions |
|---|---|---|---|
| AI computing power landscape | xAI transforms from a competitor to a computing power supplier | AI infrastructure cooperation model innovation | Pay attention to GPU cloud service price changes |
| Model capabilities | Anthropic obtains GB200 computing power acceleration | Claude’s model iteration speed may exceed expectations | Prepare to integrate Claude into more workflows |
| Competitive Situation | xAI may be withdrawing from the "general AGI race" | Grok model priority is reduced | Reduce dependence on the Grok ecosystem |
| Computing power cost | GB200 energy efficiency is several times higher than H100 | Inference cost is expected to be further reduced | Optimize AI call frequency and scale |
Industry Interpretation
xAI’s strategic shift in computing power
The most intriguing sign of this collaboration is that xAI may be moving from the front line of the AGI race to the infrastructure layer. Both Colossus1 and Colossus2 were funded by xAI, but now the core computing power is going to Anthropic rather than being used to train Grok or xAI’s own models.
Several observers in the HN community have suggested several possibilities:
- xAI believes that training its own cutting-edge models cannot form an effective barrier, and instead bets on computing power infrastructure.
- Elon Musk intends to support Anthropic to compete with OpenAI (the two companies have deep connections)
- xAI internally believes that the room for improvement of the Grok model is limited and the marginal returns of computing power are diminishing.
Computing power is the moat
An engineer in the comment area pointed out an essential question: "Does the expansion of computing power on this scale mean that computing capital is replacing model architecture and becoming the real moat for AI competition?" The answer to this question may determine the direction of the AI industry in the next five years. If the answer is yes, then organizations like xAI that can mobilize SpaceX-level engineering capabilities and capital will gain a structural advantage.
Direct impact on content creators
For readers of WayToClawEarn (content creators and automation practitioners who are concerned about the money-making opportunities of AI), the impact of this incident is reflected in:
- Claude API may be cheaper: GB200’s energy efficiency is significantly better than H100, and there is huge room for reduction in inference costs.
- Model iteration acceleration: Anthropic has obtained a stronger computing power base, and the release pace of Claude 4.5/5 may be accelerated.
- AI Entrepreneurship Ecosystem Changes: If xAI completely shifts to the infrastructure layer, it may launch GPU cloud services for developers, breaking the existing pricing structure
Related extended information
Tool entry
Related tools appearing in the text: Anthropic, Claude, OpenAI, xAI, NVIDIA
Internal link guidance
- Want to learn how to make money with Claude? Watch: Claude Code automated writing practice: build an AI content production pipeline in 30 minutes
- Real case: How Claude Code helps people earn monthly $9,000: Claude Code 48 hours to start a business: one person + US$29 monthly fee, monthly income in 3 months $9,000
- For a deeper interpretation: the relationship between AI Agent and computing power, see this comprehensive analysis: AI Agent drives automated website operations: Build a fully automatic content pipeline in 30 minutes
Monetization angle
How can you make money from this trend?
WayToClawEarn focuses on verified earn playbooks—not just news. Start from these cases.
n8n + OpenAI affiliate site
Automate content and affiliate monetization
Claude + n8n automation agency
Charge monthly for agent workflow builds