Elon's Anthropic Deal, The Next AI Monopoly?, "FDA for AI" Panic, Trading the AI Boom
The All-In podcast hosts discuss Elon Musk's XAI leasing Colossus 1 data center to Anthropic, the emergence of 'Elon Web Services' as a hyperscaler competitor, a potential 'FDA for AI' regulatory framework floated by the White House, and the broader AI-driven economic boom reflected in hyperscaler revenue growth and market highs.
Summary
The episode opens with discussion of Elon Musk's XAI leasing its Colossus 1 data center (220,000+ NVIDIA GPUs, 300+ megawatts) to Anthropic, which had been severely compute-constrained. Chamath had predicted this deal on a prior episode. The hosts frame this as the emergence of 'Elon Web Services' (EWS), positioning XAI as a hyperscaler competitor to AWS, Azure, and GCP. Brad Gerstner estimates the deal could generate $4-5 billion in incremental revenue for XAI, subsidizing Grok training costs while solving Anthropic's supply bottleneck. The hosts connect this to SpaceX's broader 'five-layer cake' business model spanning launch, connectivity, compute, hyperscaling, space data centers, and applications.
David Sachs delivers an extended analysis of Anthropic's growth trajectory — from $10B ARR in January to $30B in March to $44B in April — calling it unprecedented in Silicon Valley history. He projects Anthropic could reach $100B ARR by year-end and potentially $1 trillion by 2027, which he provocatively labels the 'biggest monopoly in human history.' He draws an extended analogy to John D. Rockefeller's Standard Oil, suggesting Anthropic's safety rhetoric could be serving as cover for regulatory capture, similar to how a hypothetical 'Safe Oil' branding could have distracted from monopolistic consolidation. Brad pushes back, arguing it's far too early to call this a monopoly given two startups are still in early revenue stages with significant competition from OpenAI, Google, and others.
The hosts then address reports that the White House is considering an 'FDA for AI' — a pre-release approval regime for AI models — sparked by Anthropic's Mythos model's advanced cyber capabilities. Sachs dismisses much of this as 'fake news' driven by Andrew Ross Sorkin's commentary, noting that senior officials like Kevin Hassett and Scott Bessent do not actually support a formal approval regime. He argues the real issue is cybersecurity hardening over the next 3-6 months as all frontier labs (and eventually open-source models) gain advanced cyber capabilities. The hosts broadly agree on the value of KYC (Know Your Customer) wrappers for frontier model APIs and better government-industry coordination, while opposing any formal pre-release approval system. Sachs frames some safety advocacy as ideologically motivated attempts at regulatory capture.
Chamath argues the underlying cause of growing AI backlash is poor messaging and failure to demonstrate broad societal benefit. He gives tech leadership a 'D minus trending to F' for not investing visibly in American society at large. Jason Calacanis suggests concrete steps like IPO-linked investment accounts for all Americans, increased minimum wage, and universal healthcare as ways to build public goodwill. Brad and Sachs are skeptical of government-driven solutions, preferring market-driven outcomes and pointing to AI's deflationary effects on the broader economy.
The episode closes with a market review, noting hyperscaler revenue acceleration (AWS +28%, Azure +39%, GCP +63%), S&P 500 margin expansion to ~13%, and AI's contribution to ~75% of Q1 GDP growth. Chamath cautions that within roughly 500 days, there will need to be a demonstrable ROI reckoning — enterprises spending on AI tokens must show measurable margin improvements — and notes no evidence yet of AI lifting S&P 500 operating margins in aggregate. Sachs and Brad counter with anecdotal and data-driven evidence of productivity gains, headcount restraint, and improving graduate employment rates.
Key Insights
- Chamath argues that Anthropic and OpenAI's revenue performance is entirely supply-constrained, not demand-constrained — if they had infinite power and compute, their revenues would be even more parabolic.
- Brad Gerstner estimates the XAI-Anthropic data center deal will generate $4-5 billion in incremental revenue for XAI, materially offsetting infrastructure costs and subsidizing Grok development.
- David Sachs draws an extended Rockefeller 'Safe Oil' analogy, arguing that Anthropic's safety rhetoric could be functioning as cover for regulatory capture while the company builds what he calls 'the biggest monopoly in human history.'
- Sachs projects Anthropic will exit 2025 at approximately $100B ARR and could reach $1 trillion in ARR by 2027, which would make it more valuable than the entire Mag7 combined.
- Brad argues that protests against data center construction are not organic local activism but highly organized campaigns similar to those that successfully blocked nuclear reactor construction 30 years ago, and calls for investigation into their funding.
- Sachs contends that the 'FDA for AI' story was substantially fake news originating from Andrew Ross Sorkin's commentary rather than actual White House policy, and notes White House Chief of Staff Susie Wallace issued a statement walking it back.
- Chamath gives tech leadership a 'D minus trending to F' for failure to visibly reinvest in American society, arguing this failure is directly causing the regulatory backlash and political antibodies forming against AI companies.
- The hosts broadly agree that KYC (Know Your Customer) identity verification wrappers for frontier model APIs make sense during preview periods, and note that labs are already tracking API usage and flagging suspicious activity to government.
- Brad argues that electricity costs are actually falling in Texas — where the most data centers are being built — while rising in New York and California that have built no new supply, refuting the narrative that data centers raise consumer electricity bills.
- Chamath warns that within approximately 500 days, there will be a critical reckoning where companies spending on AI tokens must demonstrate measurable margin improvements, and notes there is currently no evidence of AI lifting S&P 500 operating margins in aggregate.
- Sachs argues that the Trump administration's rescission of Biden-era chip export controls and model approval requirements in its first week was a primary catalyst for the current AI boom, alongside energy deregulation enabling AI companies to build their own power generation.
- Brad notes that the Mag5 combined headcount growth over three years has been approximately 3% while their revenues and operating margins have expanded significantly, citing S&P 500 operating margin improvement from 11% in 2023 to 13% in 2025 as early evidence of AI-driven efficiency.
Topics
Full transcript available for MurmurCast members
Sign Up to Access