The AI bubble is popping; we just don't know it yet

3 hours ago

Big tech Q2 earnings show wild stock swings, typical of a bubble phase, with investors reacting unpredictably to capex and cash flow reports.
Massive AI data center spending is squeezing free cash flow, especially at Meta, where capex caused a dramatic sell-off.
Amazon's earnings rose due to AWS growth and one-time energy hedging, not core business strength, raising skepticism about AI investment returns.
Chip stocks and hedge funds have seen major crashes, signaling potential bubble deflation.
AI labs like OpenAI and Anthropic generate real revenue, but their valuations rely on speculative future economic replacement by AI.
Companies face a sunk cost fallacy—too deep in AI investments to pull back without market backlash.
Supply chain constraints (power, water, cooling, chip availability) and geopolitical risks threaten AI infrastructure buildout.
Demand for AI is not yet replacing most economic activity; companies are rehiring and finding chatbots less effective than humans.
IT teams should hedge, test small models, and watch expenses instead of going all-in on frontier AI products.
Open-weight and smaller targeted models offer cost-effective alternatives to expensive frontier labs, but prices may rise due to investor pressure.

NVIDIA overstated Blackwell GPU efficiency gains, and AI nonetheless fails to make startups profitable.
Hyperscalers have dramatically increased debt and capital expenditure for AI, with no disclosed AI revenue and growing cash-flow negativity.
AI data center spending creates a vicious cycle: more demand raises hardware prices, requiring more debt, which becomes more expensive.

The author introduces a versioning system for updates on the AI bubble instead of volumes.
The AI bubble is described as a psyop, melodrama, financial crisis, and exploitation of ignorance for trillions in investment.
LLMs are real but don't fulfill promises like replacing 50% of white-collar jobs, as claimed by boosters like Dario Amodei.

The author argues that the multi-trillion-dollar AI capital expenditure boom is unsustainable, driven by reckless spending from hyperscalers on speculative assets, with little to no actual demand outside of subsidized AI labs like OpenAI and Anthropic.
The BIS annual report confirms that hyperscalers' capex commitments outpace earnings and free cash flow, warning that disappointment in returns could trigger a sudden pullback, leading to a protracted investment bust with knock-on effects on borrowers across the supply chain.
OpenAI and Anthropic are deeply unprofitable, with massive losses and compute costs, representing the majority of AI revenue (68% or more) and being subsidized by hyperscalers, making the industry's apparent revenue figures misleading.

LLMs have broken economics because their token-based costs are metered and unpredictable, unlike typical software subscriptions, and users often pay for wasted tokens.
AI companies like OpenAI and Anthropic are unprofitable; they subsidize consumer subscriptions by giving away far more tokens than the subscription cost, and enterprise token billing has led to budget overruns (e.g., Uber).
Most AI services are not differentiated or reliably useful, leading to low adoption and revenue concentration at OpenAI and Anthropic, with many startups unprofitable.

AI is expected to boost profit margins outside the tech sector, but no signs yet.
Current AI company valuations rely on future earnings promises, tied to assumptions about S&P 493 margin growth.
Debates on token costs and marketplaces matter; if costs near zero, revenue may not meet demand for hyperscalers.

NVIDIA overstated Blackwell GPU efficiency gains, and AI nonetheless fails to make startups profitable.
Hyperscalers have dramatically increased debt and capital expenditure for AI, with no disclosed AI revenue and growing cash-flow negativity.
AI data center spending creates a vicious cycle: more demand raises hardware prices, requiring more debt, which becomes more expensive.

The author introduces a versioning system for updates on the AI bubble instead of volumes.
The AI bubble is described as a psyop, melodrama, financial crisis, and exploitation of ignorance for trillions in investment.
LLMs are real but don't fulfill promises like replacing 50% of white-collar jobs, as claimed by boosters like Dario Amodei.

The author argues that the multi-trillion-dollar AI capital expenditure boom is unsustainable, driven by reckless spending from hyperscalers on speculative assets, with little to no actual demand outside of subsidized AI labs like OpenAI and Anthropic.
The BIS annual report confirms that hyperscalers' capex commitments outpace earnings and free cash flow, warning that disappointment in returns could trigger a sudden pullback, leading to a protracted investment bust with knock-on effects on borrowers across the supply chain.
OpenAI and Anthropic are deeply unprofitable, with massive losses and compute costs, representing the majority of AI revenue (68% or more) and being subsidized by hyperscalers, making the industry's apparent revenue figures misleading.

LLMs have broken economics because their token-based costs are metered and unpredictable, unlike typical software subscriptions, and users often pay for wasted tokens.
AI companies like OpenAI and Anthropic are unprofitable; they subsidize consumer subscriptions by giving away far more tokens than the subscription cost, and enterprise token billing has led to budget overruns (e.g., Uber).
Most AI services are not differentiated or reliably useful, leading to low adoption and revenue concentration at OpenAI and Anthropic, with many startups unprofitable.

AI is expected to boost profit margins outside the tech sector, but no signs yet.
Current AI company valuations rely on future earnings promises, tied to assumptions about S&P 493 margin growth.
Debates on token costs and marketplaces matter; if costs near zero, revenue may not meet demand for hyperscalers.

Hasty Briefsbeta