News

The ARM Architect on AI Bubbles, Europe's Scaleup Gap and the Race to Q-Day

Key takeaways from The Tech Download (CNBC) interview with Amadeus Capital Partners founder Hermann HauserDealroom has a profile for this one. Try Dealroom → (August 2026):

AI is real but valuations will correct. Hauser draws the electricity analogy — AI is a general-purpose technology delivering genuine productivity gains, especially in coding — but expects an internet-bubble-style correction. Some valuations rest on "gigantic circular orders" between Nvidia, OpenAI and Anthropic worth hundreds of billions that "don't seem to work out."

No startup problem, a scaleup problem. Europe produces more startups than the US and comparable university IP output, but loses companies at the scaleup stage. All twelve unicorns Hauser has been involved in ended up in the US.

Backing the scaleup gap with public capital. Via the European Innovation Council, Hauser helped launch the €5bn Scaleup Europe Fund with EQT (€300bn AUM) as manager and a €1bn public cornerstone crowding in 80% market money. The EIC's earlier €10bn fund returns €4 of market capital per euro invested.

The three ARM mistakes. Hauser recounts selling ARM to SoftBank for $32bn, failing to secure a London re-listing, and London fund managers refusing to buy back in at $50bn ("you must be joking"). ARM is now worth around $300bn — the UK's largest company, having created £250bn of value in three years versus a century for the German car industry to reach £200bn.

Fixing Europe's structural drags. Beyond money, Hauser flags the need for a European NASDAQ, and a management-mobility gap versus Silicon Valley. The forthcoming EU Inc. — an EU equivalent of a Delaware corporation harmonising share structures, voting rights and option schemes — is due later this year.

Sovereignty and dependency risks. Hauser names key European dependencies: US EDA duopoly Cadence and Synopsys, ASML's lithography monopoly, and Chinese rare earths. He backs Mistral as Europe's foundation-model option, noting ASML's recent investment and an expected new funding round.

New compute architectures under Amadeus. Citing 99% of energy going to memory-processor data transfer, Hauser has invested in four in-memory compute startups including Fractile (UK) and Semron (Dresden), plus PolyAI (voice AI, FedEx as customer). He sees Nvidia's CUDA moat eroding as developers move up to PyTorch.

Q-Day pulled forward to 2029. Hauser says recent Google and QuEra work suggests cracking RSA-2048 may need ~10,000–26,000 qubits, not a million — moving the quantum-cryptography threat from 2035 to potentially 2029. His quantum bets include error-correction leader Riverlane and networking firm Nu Quantum, both Cambridge.

Read more: CNBC

More top stories