Anthropic Has $750 Billion in Commitments… and Just $17 Billion in Revenue

Anthropic

Across the shows: 39 speakers in other episodes, 22 bullish and 41 bearish calls, past 30 days

• Jeff Snider · Atlas FinancialBearish

  Anthropic's economics don't add up: $17B revenue against $80B paper losses and up to $750B in commitments.

  Snider argues Anthropic's numbers do not reconcile. Leaked S-1 data show about $17B in revenue for the period, growing 45% but slowing from above 50% the prior quarter, against roughly $80B in paper losses in the first half of 2026 and operating losses he estimates at $8-9B for the half-year. Spending commitments reportedly reach about $750B, mostly non-cancellable, over seven to ten years, up from $500B. He doubts the company's claimed positive free cash flow and 11% adjusted operating margin, noting the 'adjusted' label. He also cites the IPO valuation being walked back from $2T to about $1.5T as evidence of the mismatch. He says Anthropic may still succeed but that pulling this off to perfection is unrealistic.

  “...spending commitments that are now maybe 750,000,000,000, the majority of which Anthropic can't cancel. The math just doesn't add up.”at 11:23

SoftBank Group

Across the shows: 2 speakers in other episodes, 0 bullish and 3 bearish calls, past 30 days

• Jeff Snider · Atlas FinancialBearish

  SoftBank's credit default swaps have risen sharply, signalling higher borrowing costs for its AI investing.

  Snider points to SoftBank's CDS, which he says has risen to a high level and exploded higher, meaning the yields it must pay to keep borrowing to invest in AI are rising. He presents this as an early sign of stress in AI financing, showing that the lenders funding AI-linked companies are demanding more compensation for the risk.

  “Even SoftBank. SoftBank has seen its CDS go up to a fair high... SoftBank has seen CDS explode higher. The yields that it has to pay to continue borrowing debt to invest in AI”at 25:40

AI bubble & tech valuations

• Jeff Snider · Atlas Financial

  AI is in a bubble that will have a downside, following the same disruptive-technology pattern as the dot-coms.

  Snider argues AI is in a bubble because every disruptive technology draws all available capital, expectations become hugely positive, and the hype cannot be met in the short run. He says the bubble exists in expectations, asset valuations and debt. Unlike dot-coms, today's AI firms have plausible paths to profit, but their projections are still fanciful and the math does not work. He says the bubble will pop, the long-run payoff comes after a bust, and the timing is unknown. He thinks the pop should come sooner rather than later so less debt is loaded into the cycle.

  “This is indeed a bubble that's going to have a downside.”at 0:40

• Ken Griffin · Citadel

  There are almost certainly bubbles in the current AI moment, as radical economic transformations usually involve bubbles.

  Griffin argues that bubbles are almost certainly present in the current AI moment and that any radical transformation of the economy is generally intertwined with a bubble. He draws a parallel to 1999-2000, when Internet companies were valued on metrics that were hard to justify, and notes that some of those companies had fanciful business models. He also questions how much AI models will be able to charge if they actually solve long-standing problems.

  “there's almost certainly bubbles within this moment in time. ... any radical transformation of our economy is generally intertwined with a bubble.”at 1:28

Frontier AI labs & model race

• Jeff Snider · Atlas Financial

  Large language models are prediction systems with hard scaling limits, so their progress will plateau.

  Snider argues AI models are prediction systems, not intelligence, and that they face real technical constraints. He cites the Chinchilla scaling laws: performance gains from bigger models depend on ever more training data, and the internet's high-quality human text is running out. As models train on increasingly AI-generated content, errors compound, and paying humans for high-quality books and journals has diminishing returns. He says there are reports the plateau in productivity was reached last year.

  “These are prediction models. These are not intelligence. There's no thinking and thought behind them.”at 2:58

Enterprise AI adoption & AI software

• Jeff Snider · Atlas Financial

  Businesses turning AI into profitable productivity gains is not guaranteed, and it is a sticking point in every technology cycle.

  Snider argues that businesses must figure out how to use AI productively and profitably, and that this is no guarantee. He describes it as one of the sticking points in every technology adoption cycle, which leads from a bubble on the way up to a bust on the way down. He lists slow or limited business adoption as one of three pivot points that could bring the cycle to its downside.

  “Businesses have to figure out how to productively use AI in a profitable fashion, which is no guarantee. That's one of the sticking points in every single technological adoption cycle”at 3:12

Credit markets

• Jeff Snider · Atlas Financial

  Credit markets are cracking, and the debt cycle will decide how far the AI bubble runs.

  Snider argues the debt cycle will ultimately determine how the AI boom ends, because the build-out requires astronomical borrowing and lenders will tighten if business adoption or technical progress disappoints. He says the credit cycle is already showing cracks beyond AI: triple-C bond yields reportedly rose above 1,300 basis points, which he had flagged as the next threshold. He ties this to macro pressure from the energy shock. He says that as credit becomes more reluctant to lend, AI funding will be constrained, and that credit is the key place to watch for the regime change.

  “So the credit markets overall are showing cracks to begin with, not just related to AI... just yesterday, triple C's exploded above 1,300 basis points”at 25:44

AI: overall outlook

• Jeff Snider · Atlas Financial

  AI will transform the world, but the payoff is a 2030s or 2040s story, not a 2020s one.

  Snider argues AI will likely change the world and rewrite how people live, so the long-run thesis is not wrong. He says the issue is timing. Technical constraints and business adoption problems mean the real payoff comes after a shakeout, in the 2030s or even the 2040s. Near-term expectations of a singularity-like change are overstated.

  “It's gonna change the world... And that may be true, but it's a timing issue. That's not a 2020s outcome. That's a 2030s, maybe even 2040s outcome.”at 27:23