• Avoid treating Anthropic as a straightforward IPO buy: its projected compute costs may exceed $74 billion in 2027, while revenue growth and usage remain uncertain; compare full costs with revenue before investing.
• Treat OpenAI as high risk if it goes public, and scrutinize its funding runway and cash burn rather than relying on a valuation below $1 trillion, which is only a scenario.
• Be cautious on SpaceX: the speaker sees it as overvalued and warns that computing contracts could be canceled in H1 2027, but provides no price target.
• For Broadcom (AVGO) and Alphabet (GOOGL/GOOG), monitor whether Anthropic TPU leases begin in 2027; the discussion offers no specific buy recommendation.

Anthropic (pre-IPO; no ticker stated)

• The amended prospectus reportedly showed Q3 2026 revenue of just under $17 billion, equivalent to an annualized run rate of about $68 billion.
• Anthropic posted a small operating loss in Q3 after recording an operating profit in Q2. The speaker attributed the reversal to rising payments under its SpaceX computing deal.
• The filing’s disclosed computing commitments were $5 billion for 2026, $28 billion for 2027, and $43 billion for 2028. The speaker noted that these figures exclude several obligations, including the SpaceX deal, future data centers, and TPU leases that have not yet begun.
  • The speaker’s revised estimates, which include additional obligations, put computing costs at $74 billion in 2027 and more than $100 billion by 2029.
  • Anthropic reportedly spent about $19 billion on computing in the first nine months of 2026 and was approximately at break-even over that period.
• The speaker argued that Q3 may be a near-term revenue high, citing Ramp usage data: Anthropic’s index recovered to 98.4 after the launch of Claude Opus 5.5 but remained below its July peak of 100.
• Overall sentiment was bearish. The speaker questioned whether revenue can grow fast enough to cover the company’s rising computing costs and commitments.

Takeaways

• For prospective IPO investors, compare revenue growth and operating results with the full expected cost of compute—not only the commitments disclosed in the filing.
• Track usage and revenue after model launches. The speaker’s bearish thesis depends partly on the view that the earlier “token maxing” surge has faded.
• The transcript gives no IPO price or valuation target for Anthropic.

OpenAI (private; no ticker)

• The speaker said OpenAI’s reported funding valuation fell from $1.4 trillion to $1.25 trillion and that it raised $11 billion against a reported $30 billion target.
• The speaker interpreted the lower valuation and smaller-than-targeted raise as evidence that OpenAI needs funding and that investor appetite has weakened.
• The speaker said OpenAI had hoped to IPO in the second half of 2026 but may be unable to do so, and suggested that a 2027 IPO could require a valuation well below $1 trillion if Anthropic’s IPO underperforms.
• Sentiment was extremely bearish, with cash burn and the difficulty of raising additional capital identified as key concerns.

Takeaways

• If OpenAI becomes publicly investable, assess its funding runway, cash burn, and ability to meet its own projections alongside its valuation.
• The transcript’s possible sub-$1 trillion valuation is the speaker’s scenario, not a confirmed target or outcome.

SpaceX (ticker not stated)

• The transcript says SpaceX went public in June at an IPO price of $135. Its stock reportedly opened around $160, rose to about $200, and later fell to $136.
• The speaker described SpaceX as overvalued and said its Q3 revenue benefited substantially from short-term computing contracts with Anthropic and Google.
• The speaker expects those computing contracts could be canceled in the first half of 2027, potentially reducing SpaceX revenue and putting pressure on its stock price.

Takeaways

• Consider how much of SpaceX’s reported revenue depends on temporary computing contracts, and watch for contract renewals or cancellations in 2027.
• The transcript provides specific historical prices but no price target. Its expectation of a potential stock-price decline is the speaker’s view.

Broadcom (AVGO)

• Broadcom is involved in supplying TPUs, alongside Google, for leases to Anthropic.
• The speaker said those leases have not yet begun because the TPUs have not been delivered, so they are not included in Anthropic’s disclosed commitment table. They are expected to begin in earnest in 2027.

Takeaways

• For investors following AVGO, the discussion points to AI-chip demand and financing-backed TPU deployments as areas to monitor.
• The transcript does not provide a view on Broadcom’s valuation, earnings impact, or stock-price target.

Alphabet / Google (GOOGL, GOOG)

• Google is identified as a TPU supplier alongside Broadcom, with TPUs expected to be leased to Anthropic through special-purpose vehicles financed by third-party lenders.
• The transcript says the leases have not yet begun and are expected to start in earnest in 2027.
• Google is also named as having short-term computing contracts with SpaceX; the speaker argues that these contracts may be canceled in the first half of 2027.

Takeaways

• Monitor whether TPU deliveries and leases begin as expected, and how much of related demand is sustained rather than temporary.
• The transcript does not offer a Google stock recommendation or price target.

NScale and Lambda (private data-center providers)

• Anthropic reportedly has tens of billions of dollars in spending commitments with NScale and Lambda for data centers that have not yet been built.
• The speaker said they do not view either company as credible and doubted that the data centers would be built, so they believe Anthropic may not have to pay those amounts.

Takeaways

• These commitments are a source of uncertainty in estimating Anthropic’s future computing costs: the speaker treats them as less likely to become actual expenses, but the transcript does not establish that they will be canceled or avoided.
• The transcript provides no investable ticker or valuation information for either provider.

AI computing infrastructure

• The discussion highlights the scale of data-center and compute spending required by major AI companies, including long-term facility commitments, TPU leases, and contracts with cloud and data-center providers.
• The speaker’s central concern is that infrastructure costs may rise faster than AI companies’ revenue, particularly if current usage growth does not continue.

Takeaways

• When evaluating AI companies or infrastructure suppliers, distinguish between announced commitments, costs already being incurred, and contracts that depend on future construction or delivery.
• Pay attention to whether customer demand and revenue growth keep pace with the large, multi-year costs described in the transcript.