OpenAI's Identity Crisis, Datacenter Wars, Market Up on Iran News, Mamdani's First Tax, Swalwell Out
The All-In podcast crew discusses OpenAI's strategic identity crisis and competition with Anthropic, the growing backlash against data center construction across America, and New York Mayor Mamdani's proposed pied-à-terre tax. They also cover the market's resilience amid the Iran conflict, Eric Swalwell's exit from the California governor's race, and play a game show guessing overvalued failed startups.
Summary
The episode opens with a discussion of New York Mayor Mamdani's proposed pied-à-terre tax, estimated at 3.9% annually on second homes valued over $5 million within 15 miles of midtown Manhattan. The hosts argue this will devastate demand for second homes by wealthy non-residents, effectively doubling the cost of ownership over a decade. They compare it to London's stamp duty and non-dom tax changes, which caused high-end real estate to collapse and wealthy individuals to relocate assets to Zurich, Milan, and Lugano. The panel criticizes the tax as counterproductive, noting that wealthy non-residents who own pied-à-terres pay substantial property taxes while using minimal city services — making them net-profitable for the city. They contrast New York's approach with Austin, Texas, where permissive building policies led to falling rents despite surging population growth, arguing that the real solution to housing affordability is supply, not taxation.
The main tech topic covers OpenAI's identity crisis, prompted by a leaked four-page memo from Chief Revenue Officer Denise Dresser, which acknowledged that OpenAI's $30 billion revenue run rate is inflated by approximately $8 billion due to revenue-sharing accounting with AI model providers. The memo also attacked Anthropic's positioning and laid out a pivot toward enterprise and the agent platform layer. Anonymous OpenAI investors were quoted criticizing the company as 'deeply unfocused,' given that ChatGPT has 1 billion users growing 50-100% annually yet the company is chasing enterprise and coding. The hosts analyze the growth rate disparity: OpenAI is growing roughly 3-4x annually, while Anthropic is growing approximately 10x annually, with Anthropic projected to hit $80-100 billion in ARR by end of year. David Sacks argues this divergence stems from Anthropic's laser focus on enterprise coding, where businesses pay metered token costs that scale far better than consumer subscriptions at $20/month with only 3-4% conversion rates. Travis Kalanick frames the dynamic through a network effects lens from his Uber experience, arguing that whoever builds scale through revenue-funded compute will ultimately outcompete those relying on external capital subsidies.
A major thread concerns data center construction being blocked across the United States. Chamath notes that roughly 100 data centers are currently being contested nationwide, with about 40% getting canceled — a figure that has more than doubled year-over-year, representing $162 billion in economic value at risk. He identifies three contributing forces: genuine local concern about electricity rate increases, well-funded 'Doomer' groups like Future of Life Institute astroturfing NIMBYism, and Anthropic itself, which allied with anti-data-center forces as a competitive strategy when it relied on hyperscaler compute rather than owning its own infrastructure. Sacks argues this strategy has now backfired, as Anthropic appears to have hit the limits of third-party compute availability — evidenced by Claude's thinking time being cut by two-thirds before being restored in a newer model. Freeberg adds a populist dimension: the data center has become a physical symbol of wealth inequality, making it a natural target for anti-elite sentiment even when the practical arguments against them are weak or fabricated.
On markets, the hosts note the S&P 500 hitting new all-time highs during the seventh week of military action against Iran. Sacks interprets this as the market pricing in resolution of the conflict based on the Islamabad meeting and presidential statements. Chamath flags that both the Shiller PE ratio and the Buffett Index are near all-time highs, suggesting the market is expensive by historical standards, though a dispersion signal — where only 8-9 companies are hitting highs — makes the picture complex. Travis argues Trump treats the stock market as his personal weather vane, reliably pulling back from policies that threaten it too severely. Chamath says he is personally risk-off and waiting for pending IPOs (SpaceX, Anthropic, or OpenAI) to de-lever.
The Eric Swalwell segment features Freeberg revealing that multiple independent sources had told him about the alleged misconduct allegations as far back as December, yet nothing had been made public. He describes how the simultaneous coordinated release of all allegations suggests deliberate orchestration rather than spontaneous disclosure. Sacks draws a direct parallel to the Biden forced-withdrawal situation, pointing to Nancy Pelosi as the likely central figure in both cases, and argues California Democratic insiders needed to winnow the governor's field to avoid two Republicans advancing through the jungle primary. The episode closes with a game show segment — 'The Price Is Wrong' — where hosts guess famously overvalued failed startups including OpenSea (NFTs at $13B), Clubhouse ($4B), Theranos ($9B), and Quibi ($1.7B).
Key Insights
- David Sacks argues Anthropic is growing at roughly 10x annually while OpenAI is growing at only 3-4x, and claims whoever wins on revenue-funded compute scale will achieve an insurmountable lead within one to two years.
- Chamath contends that OpenAI's $30 billion run rate is inflated by $8 billion due to revenue-sharing accounting with model providers, making apples-to-apples comparison with Anthropic show Anthropic is roughly 20% smaller but growing far faster.
- Sacks argues Anthropic's strategy of allying with anti-data-center Doomer groups has backfired, as the company now needs its own compute infrastructure but has helped make data center approvals politically toxic across the country.
- Chamath reports that approximately 100 data centers are currently contested nationwide, 40% get canceled, and the cancellation rate has more than doubled year-over-year, representing $162 billion in threatened economic value.
- Freeberg claims that multiple independent sources informed him of Swalwell misconduct allegations back in December, and the simultaneous coordinated public release of all allegations suggests deliberate orchestration tied to California Democratic Party efforts to winnow the governor's field.
- Sacks draws a direct parallel between the Swalwell situation and Biden's forced withdrawal, arguing Nancy Pelosi is the central figure in both cases and that the Democratic machine deployed the same playbook of applying private pressure before going public.
- Travis Kalanick argues that Trump uses the S&P 500 as his primary policy weather vane and reliably reverses course on policies that push markets down too far, which the market has now internalized and is pricing in.
- Chamath states he is personally risk-off because both the Shiller PE ratio and the Buffett Index (total US equities divided by GDP) are near all-time highs, and is waiting for SpaceX, Anthropic, or OpenAI IPOs to de-lever his portfolio.
- Sacks argues that consumer AI subscriptions at $20/month with 3-4% conversion rates cannot scale revenue the way enterprise metered token pricing does, which is why Anthropic's enterprise coding focus has driven dramatically superior revenue growth.
- The hosts argue that New York's pied-à-terre tax targets the most price-elastic segment of the real estate market — non-resident second-home owners who could buy anywhere — and will replicate the collapse of London's high-end real estate market that followed similar stamp duty changes.
- Travis Kalanick argues that change management — not technical capability — is the primary bottleneck to AI productivity gains at large enterprises, because complex undocumented processes and organizational bureaucracy resist transformation regardless of the technology's quality.
- Chamath argues that Anthropic may have deliberately withheld its Mythos model not purely for safety reasons but because it was 10-20x more expensive per token than Opus and they lacked the compute to serve it commercially, using the safety narrative as cover while generating a major marketing event.
Topics
Full transcript available for MurmurCast members
Sign Up to Access