Arm co-founder Hermann Hauser’s AI warning: The revolution is real, but so is the bubble risk

Hermann Hauser says the AI/semiconductor wave will ultimately create more value than prior technology revolutions, but expects a “rollercoaster” as some valuations are “ahead of themselves” and circular financing risks rise. He argues large AI-capitalized players (e.g., OpenAI/Anthropic) should endure turbulence, while computing architecture is forced to evolve due to costly AI compute and bottlenecks (driving in-memory and photonic approaches). The newsletter also flags active AI-and-chips demand (TSMC’s July sales jump; Intel’s $20B common offering) alongside governance and security scrutiny, after OpenAI, Anthropic, and Meta reported rogue model behavior during routine testing.

Analysis

The actionable signal is a broadening of AI economics, not a slowdown. If the bottleneck shifts from raw accelerator demand to memory bandwidth, cooling, and power-efficient architectures, the marginal dollar of capex migrates away from the most crowded beneficiary set and toward the hidden infrastructure names. That is structurally constructive for TSM and SKHYV, and selectively for ARM if heterogeneous and low-power designs gain share; it is less about one chip vendor “winning” and more about which layer captures the constraint rent.

INTC’s financing move matters as a balance-sheet signal: the industry’s capital intensity is rising faster than product cycles can absorb it, which raises dilution risk for weaker operators and strengthens the premium on self-funded scale. META and GOOGL should absorb turbulence better than the model vendors because distribution and cash generation let them endure a valuation reset, but any evidence that AI spend is outrunning monetization would pressure their multiples over the next 1-3 months. The immediate market reaction may be noisy; the bigger move comes if investors start repricing the durability of AI capex rather than the headline growth rate.

The contrarian view is that consensus is still too concentrated on NVDA as the sole tradeable expression of AI. A more durable setup is the supply-chain and architecture layer where pricing power is tied to physical constraints, not narrative multiple expansion. Falsifiers: a rollover in TSM monthly sales, HBM pricing, or hyperscaler capex commentary would weaken the thesis; absent that, the likely path is continued dispersion, not a collapse, over 6-18 months.

Key Decisions for Investors

• Long TSM / short NVDA for 1-3 months: express the view that AI spend is broadening into foundry/packaging rather than staying concentrated in accelerators. Risk/reward is attractive if NVDA multiple compresses on architecture-diversification headlines; falsify on another clean NVDA guide raise.
• Buy SKHYV on pullbacks for a 6-12 month horizon: HBM and memory remain the gating item, and scarcity pricing can persist even if GPU enthusiasm cools. Stop if memory pricing rolls over or if customer concentration starts to unwind.
• Short INTC against TSM into the next financing/risk window: dilution plus execution risk makes INTC the cleaner way to express balance-sheet fragility in AI. Cover if foundry traction or AI-related order commentary materially improves over the next quarter.
• Use ARM as a call-spread candidate on weakness, 6-18 months: if power efficiency and chiplet design adoption increases, ARM can earn a higher architectural royalty mix. The risk is valuation; only pursue on a post-earnings or sector pullback.