Price:

AI & TECH

Musk's xAI shifts strategy, becomes Anthropic's infrastructure backbone

Friday, May 15, 2026 · from 3 podcasts
  • Elon Musk leases a 220,000-GPU data center to Anthropic, pivoting xAI to an infrastructure play and minting billions.
  • The deal solves Anthropic's acute compute shortage, letting it chase 'infinite' context windows and agentic workflows.
  • Frontier model builders face a $60B price tag for a gigawatt of power, pushing compute providers toward orbital ambitions.

Elon Musk is no longer competing with OpenAI and Anthropic for the best model. He’s building the power grid their ambitions require.

Musk’s xAI and SpaceX are leasing the ‘Colossus 1’ data center - a 300-megawatt facility with 220,000 Nvidia H100 GPUs - to Anthropic. On All-In, David Sacks framed the deal as the birth of “Elon Web Services,” a direct infrastructure rival to AWS and Azure that generates immediate billions to fund Musk’s own Grok models. For Musk, it’s an admission that building frontier models is a different skill set than building massive hardware.

“This move provides 220,000 Nvidia GPUs and 300 megawatts of power. It also marks the birth of Elon Web Services (EWS).”

- All-In podcast

Anthropic was starving for capacity. The AI Daily Brief’s Nathaniel Whittemore noted the company had the best-in-class coding model but couldn't run it reliably, throttling user tokens and pausing research. This lease doubles Claude Pro rate limits overnight. It resolves a mutual crisis: Musk had unmatched hardware but a stalled model and a dwindling team.

With its compute constraint broken, Anthropic is shifting focus from raw model releases to building an agentic ecosystem. At its recent ‘Code with Claude’ event, the company unveiled ‘Managed Agents’ with features like Dreaming, which lets agents review their own sessions to improve over time, and Outcomes, which uses a separate grading agent to audit work against user-defined rubrics. Boris Cherny, creator of Claude Code, said manual coding has effectively disappeared within Anthropic’s own teams.

The financial scale of this infrastructure arms race is staggering. On FYI, Frank Downing outlined the math: a one-gigawatt data center costs roughly $60 billion - $20 billion for land and power, plus $40 billion for the IT stack and GPUs. If used by the model builder itself, that capacity can drive $30 billion in annual revenue, offering a two-year payback.

“If rented out as infrastructure-as-a-service, a gigawatt generates roughly $15 billion per year. If used by the model builder itself, like OpenAI or Anthropic, that same capacity can drive $30 billion in annual revenue.”

- Frank Downing, FYI podcast

That payoff is why the hunt for power is going orbital. As terrestrial sites grow scarce, SpaceX plans to use Starship to launch modular AI clusters into space. Brett Winton expects this shift to begin around 2028. At $300 per kilogram launch costs, deploying a gigawatt in space could cost $7.5 billion, undercutting the $20 billion terrestrial infrastructure price tag. For companies like Anthropic, getting hardware online fast is more valuable than saving marginal costs later.

The deal crystallizes a new hierarchy: the companies that control the electrons will ultimately govern the tokens.

Source Intelligence

- Deep dive into what was said in the episodes

SpaceX And Anthropic Partnership | The Brainstorm EP 131May 13

  • Anthropic has been supply-constrained on compute, restricting user tokens and cutting off research compute, forcing them to lease Colossus 1 - a 300MW, 220,000 GPU data center - from SpaceX to lift capacity limits.
  • Anthropic previously enforced a policy to cut off competitors like XAI from using its models, but the compute deal has reopened communication between them.
  • SpaceX’s vertical integration from chip fabrication to model inference reduces its risk when leasing out compute, as it can more rapidly rebuild capacity.
  • Building a gigawatt-scale AI data center costs roughly $60B: $19B for the facility, $30B for GPUs, and $11B for other IT equipment like CPUs and networking.
  • A gigawatt of leased compute infrastructure generates about $15B annually; a model provider operating on that scale can generate $30B in revenue.
  • OpenAI’s revenue scaled with compute; they generated $20B on roughly 2 gigawatts early this year, implying about $20B per effective gigawatt for inference.
  • Revenue per watt is increasing as model utility and enterprise willingness to pay rise, creating pricing power for both infrastructure providers and model companies.
  • Space-based AI compute economics hinge on Starship launch costs; at $300/kg, launching a gigawatt costs $7.5B, beating terrestrial facility costs.
  • Manufacturing satellites on a production line offers cost efficiencies over building unique terrestrial data centers, where $5B of the $19B facility cost is labor.
  • Compute scarcity on Earth means AI companies like Anthropic would pay a premium for space-based watts even before launch costs break parity, valuing velocity over price.
  • ARK analysts project SpaceX could begin scaling space-based AI compute in 2028-2029, reaching tens of gigawatts per year in the early 2030s.
Also from this episode: (1)

Energy (1)

  • Small modular nuclear reactors are more likely to scale in the US than gigawatt plants, fitting incremental demand like offsetting retired coal plants.

How to Build an AI Native Team with Mike Cannon-BrookesMay 9

  • Anthropic is developing future models with higher judgment, 'infinite' context windows, and multi-agent coordination, with research head Diane Penn suggesting infinite context could enable continual learning.
  • Anthropic CEO Dario Amodei revealed the company saw 80x annualized growth in revenue and usage during the first quarter of 2026.
  • Anthropic and SpaceX announced a partnership granting Anthropic full use of XAI's Colossus 1 data center, containing 220,000 Nvidia GPUs operating at 300 megawatts.
  • Elon Musk explained his decision to lease Colossus 1 after meeting with Anthropic's team and concluding their work was 'good for humanity'.
  • Nathaniel Whittemore argues Elon Musk's pivot from model builder to compute provider represents AI play 3.0, aligning with his strength in scaling known-but-hard infrastructure projects.
  • Chamath Palihapitiya predicted power constraints would force AI labs to negotiate for compute, creating an opportunity for SpaceX's excess capacity.
  • Anthropic will dissolve XAI as a separate company, integrating it as SpaceX AI, and will initially raise API rate limits and eliminate peak hour reductions for Claude Code users.
Also from this episode: (6)

AI & Tech (5)

  • Anthropic's Code with Claude event centered on agent harnesses and workflow tools rather than a major model release, reflecting the shift from model to ecosystem competition.
  • Anthropic launched Dreaming, a scheduled memory review system for agents that extracts patterns from past sessions to improve performance over time.
  • Anthropic's Outcomes feature uses a separate grading agent with a user-defined rubric to automatically evaluate and iterate on agent outputs, improving document quality by 8.4% for Word and 10.1% for PowerPoint.
  • Managed Agents now support multi-agent orchestration, allowing a lead agent to delegate tasks to specialists with shared context and auditable execution traces.
  • Anthropic released a suite of 10 predefined financial service agents for Claude Finance, including pitch builders and market researchers, alongside a cookbook for customization.

Coding (1)

  • Claude Code creator Boris Cherny disavowed the term 'vibe coding', stating his company uses coordinated AI agents over Slack with automated testing and no manually written code.

Elon's Anthropic Deal, The Next AI Monopoly?, "FDA for AI" Panic, Trading the AI BoomMay 8

  • Elon Musk leased Colossus 1, his H100-powered data center, to Anthropic, providing them with over 220,000 Nvidia GPUs and over 300 megawatts of energy.
  • Chamath Palihapitiya argues AI revenue growth for Anthropic and OpenAI is constrained entirely by compute and power supply, not demand. Brad Gerstner estimates this deal will generate $4-5B in incremental revenue for SpaceX this year.
  • David Sacks cites Anthropic's revenue growth as unprecedented, tripling from roughly $10B ARR to $30B ARR from January to March 2024, then accelerating to $44B ARR in April. He argues this trajectory could make it the most powerful monopoly in history.
  • Sacks draws a parallel between Anthropic's safety-focused branding and John D. Rockefeller, suggesting effective altruism rhetoric could distract from the creation of a historic monopoly, using the hypothetical example of 'Safe Oil'.
  • Brad Gerstner points to hyperscaler revenue growth as proof of the AI boom's ROI: AWS grew 28% to a $150B run rate, Azure grew 39% to $108B, and Google Cloud grew 63% to $80B.
  • Jason Calacanis advocates for tech winners to give back, suggesting companies like Nvidia or SpaceX allocate 1-5% of IPO shares to a public investment fund, and calls for a sector-led, gradual increase in the minimum wage to boost consumer spending.
  • Brad Gerstner credits Trump administration policies - rescinding Biden-era chip/model approval rules and promoting 'drill baby drill' energy policies - for unleashing the AI boom and enabling the concurrent blue-collar construction surge.
  • David Sacks identifies organized activist groups, funded by unknown parties, as the primary force protesting new data center construction, comparing their tactics to those that halted nuclear reactor construction decades ago.
Also from this episode: (5)

Regulation (1)

  • The hosts discuss reports that the White House is considering an 'FDA for AI' to vet new models for safety, a move reportedly catalyzed by Anthropic's Mythos model. David Sacks dismisses this as fake news, stating senior Trump officials do not support an approval regime.

Models (1)

  • Sacks argues the real regulatory need is for cyber defense, as models like Mythos and OpenAI's equivalent give sophisticated hacking capabilities. He supports KYC for preview API access but opposes pre-release government approval of models.

AI & Tech (2)

  • Chamath Palihapitiya is skeptical that AI-driven productivity gains have yet materialized in broader economic data, noting a lack of evidence for lifted S&P 500 operating margins. He predicts a reckoning on ROI within roughly 500 days.
  • Gerstner argues against top-down government solutions for healthcare and minimum wage, stating market-driven abundance from AI will solve these problems more effectively than increased regulation.

Business (1)

  • David Sacks counters that unemployment remains at historic lows (~4.2%) despite efficiency gains, and a Wall Street Journal article shows it's getting easier for recent college graduates to find jobs, contradicting AI job-loss narratives.