CVAI London: European… | Newcomer Newsletter Summary

1. Key Themes

European AI & Defense Sovereignty Is Becoming an Investment Category

Geopolitical anxiety about U.S. reliability is directly driving capital formation in Europe — both in AI infrastructure and defense tech. Cohere's Aidan Gomez is merging with a German model provider to build multi-democracy AI capability, while Sequoia co-led a €500M defense round.

"We want to build this alliance to help create capability beyond just one democracy." — Aidan Gomez "Unfortunately, there's nothing like conflict at the doorstep to really wake you up as a government. Also I think governments realize that they've been under-investing in defense since before the war." — Luciana Lixandru

The AI Bubble Is Widely Acknowledged — But Not Seen as Imminent

A majority of surveyed AI insiders believe a bubble exists, yet the consensus is that it won't pop soon. The market is running on belief, not distress signals.

51% of respondents said "yes bubble, not bursting this year," compared with just 5% who said "yes bubble, yes about to burst."

Compute Scarcity Is Worsening and Accelerating Hoarding Behavior

GPU availability is critically constrained, with reservations required 6–18 months in advance. Companies believe the crunch will get worse, creating a self-reinforcing race to lock up infrastructure.

"Basically all GPUs are essentially sold out, pretty much everywhere. People are kind of making reservations 6 to 18 months in advance and even for those it's a competitive process." — Josh Tobin "There's a lot of people in the industry right now that believe that over the next couple of years the compute crunch is going to get even worse than it is today, and so it's creating this race dynamic where these companies are buying up everything that they can get their hands on."

Agent Identity and Interaction Design Remains Unsettled

The AI agent space lacks a dominant paradigm. Startups are still debating whether users want persistent agent identities or fluid, capability-driven interfaces — a signal that the UX and product layer is wide open.

"The industry is still sorting out whether humans will mostly interact with agents with unique identities and capabilities or with a more fluid, hive-mind-like AI with capabilities that can be turned on and off."

Vertical Domain Depth Is the Defensible Strategy for AI Founders

The big labs are concentrating on coding, leaving audio, video, vision, and consumer verticals underserved. Founders who build around unbearable customer pain in a specific domain have a durable edge.

"While OpenAI and Anthropic pour gobs of money and research into AI coding tools, Tobin said that this focus is leaving audio, vision, video, and consumer applications genuinely underserved." "Customers buy a product when they have unbearable pain and you have convinced them that only your software product can put an end to their pain. Everything else is just getting lucky." — R. Martin Chavez

2. Contrarian Perspectives

OpenAI Is the Most-Wanted Short Among AI Insiders

Despite its dominant public profile and trillion-dollar valuation narrative, a plurality of AI practitioners — the people closest to the technology — would bet against OpenAI at its current price.

"When asked what company they would most like to short at its present valuation, a plurality of 33% answered OpenAI, with Perplexity next at 25%."

Safe Superintelligence Is More Attractive Than Most Application-Layer Unicorns

SSI — a company with no product, no revenue, and near-zero public profile — ranked third among companies insiders would want to own at today's valuation, ahead of most application-layer companies. This suggests sophisticated insiders are betting on fundamental model research over near-term commercialization.

"Safe Superintelligence came in third with 8%" of respondents choosing it as the private unicorn they'd most want to own — beating out the majority of well-known AI application companies.

The Compute Bubble Signal Is Hidden in Plain Sight

The GPU market shows none of the slack that typically precedes a bubble pop. This makes the "bubble" framing intellectually interesting but practically irrelevant for near-term capital allocation — the physical supply constraint is real, not speculative.

"If it's a bubble, there are no signs that I've seen that it's like bursting right now. Basically all GPUs are essentially sold out, pretty much everywhere." — Josh Tobin

3. Companies Identified

Cohere Enterprise AI model provider; Canadian company Why mentioned: CEO Aidan Gomez is merging with a German model provider to build a multi-democracy AI capability stack, positioning as a sovereignty-safe alternative to U.S. hyperscalers.

"We want to build this alliance to help create capability beyond just one democracy."

Stark Defense European defense startup making kamikaze drones and other weapons Why mentioned: Sequoia co-led a €500M round — cited as a direct investment thesis tied to European governments waking up to defense underinvestment.

"Unfortunately, there's nothing like conflict at the doorstep to really wake you up as a government."

Anthropic AI safety-focused frontier model lab Why mentioned: Named the top private unicorn insiders would want to own at today's valuation, with 54% of survey respondents choosing it — framed as the consensus bull bet among practitioners.

"54% said Anthropic" when asked what private unicorn they would like to own at today's valuation.

ElevenLabs AI voice/audio model company Why mentioned: Cited repeatedly as a model case study for successfully carving out a defensible vertical niche (voice) while the large labs focus elsewhere. Second most popular unicorn pick at 13%.

"ElevenLabs landed many plaudits throughout the day as one company which successfully carved out its market for voice models."

Safe Superintelligence (SSI) Research-stage AI lab focused on superintelligence safety Why mentioned: Ranked third (8%) on the list of unicorns insiders would want to own — a surprising result given its lack of commercial product.

"Surprisingly, Safe Superintelligence came in third with 8%."

Dust Agent orchestration startup Why mentioned: Co-founder Stanislas Polu used the company's experience to surface key unresolved questions about agent identity, skills frameworks, and human-AI interaction design.

"The company sees skills — specific, definable abilities — as perhaps the most useful framework for putting models to work."

Recursive AI foundation model lab pursuing recursive self-improvement Why mentioned: Raised $650M; CTO Josh Tobin provided the most direct first-person account of the GPU scarcity problem and its competitive implications.

"The lab is betting it can go all the way to recursive self-improvement with $650 million in funding."

OpenAI Dominant AI lab Why mentioned: Named the top short candidate by 33% of surveyed insiders — notable as a bearish data point from practitioners.

"A plurality of 33% answered OpenAI" when asked which company they'd most like to short at its present valuation.

Perplexity AI-powered search engine Why mentioned: Second most popular short candidate at 25%, suggesting valuation skepticism among insiders.

"Perplexity next at 25%" on the most-wanted-short list.

Harvey AI legal tech company Why mentioned: Third on the short list at 8%, indicating some valuation concern despite its high profile in vertical AI.

"Harvey came in third on the short list with 8%."

4. People Identified

Aidan Gomez CEO of Cohere; co-inventor of the Transformer architecture Why mentioned: Articulating a European AI sovereignty strategy through a merger with a German model provider; living in the UK and actively building cross-democracy AI alliances.

"We want to build this alliance to help create capability beyond just one democracy."

Luciana Lixandru Partner at Sequoia Capital Why mentioned: Representing Sequoia's thesis on European defense and AI sovereignty; co-led the €500M Stark Defense round.

"Also I think governments realize that they've been under-investing in defense since before the war."

R. Martin Chavez Former Goldman Sachs CIO; Vice Chairman at Sixth Street; Alphabet board member Why mentioned: Offered a practitioner's framework for AI product-market fit, grounded in his Goldman sales experience.

"Customers buy a product when they have unbearable pain and you have convinced them that only your software product can put an end to their pain. Everything else is just getting lucky."

Josh Tobin Co-founder and CTO of Recursive Why mentioned: Provided the most data-rich, on-the-ground account of GPU scarcity and the hoarding dynamic forming among foundation model companies.

"Basically all GPUs are essentially sold out, pretty much everywhere. People are kind of making reservations 6 to 18 months in advance."

Stanislas Polu Co-founder of Dust (agent orchestration startup) Why mentioned: Surfaced the unresolved product design question at the heart of the agent space — skills vs. identity — based on direct user behavior observations.

"Many users prefer the concept of an agent with an identity and a particular set of skills."

Tom Hulme Partner at GV (Google Ventures) Why mentioned: Co-led Recursive's $650M funding round alongside Greycroft, backing the recursive self-improvement thesis.

"The lab is betting it can go all the way to recursive self-improvement with $650 million in funding co-led by Tom Hulme at GV and Greycroft."

5. Operating Insights

Pick Verticals the Labs Are Ignoring, Not the Ones They're Crowding

The most actionable advice for founders: don't compete with OpenAI and Anthropic on coding. Target the genuinely underserved modalities — audio, vision, video, consumer — where big lab attention is sparse. ElevenLabs is the proof point.

"This focus is leaving audio, vision, video, and consumer applications genuinely underserved."

Lead with Pain, Not Technology

R. Martin Chavez's framework — derived from Goldman's top sales executive — is a forcing function for product positioning: if you can't articulate the unbearable pain your product eliminates, you don't have a product, you have a lottery ticket.

"Customers buy a product when they have unbearable pain and you have convinced them that only your software product can put an end to their pain. Everything else is just getting lucky."

Secure Compute Early — Treat It as a Strategic Asset, Not a Variable Cost

For any AI company building at the infrastructure or model layer, GPU access is a competitive moat. The window to secure capacity at reasonable terms is narrowing, and waiting is a strategic mistake.

"People are kind of making reservations 6 to 18 months in advance and even for those it's a competitive process."

6. Overlooked Insights

The "Skills" Framework May Be the Winning Agent Architecture

Buried in the Dust conversation is a potentially important product design insight: framing agent capabilities as discrete, transferable "skills" rather than monolithic agent personalities may be the more scalable and enterprise-friendly architecture — but it's in tension with what end users actually want emotionally.

"The company sees skills — specific, definable abilities — as perhaps the most useful framework for putting models to work. But many users prefer the concept of an agent with an identity and a particular set of skills."

Cohere's Cross-Border Model Merger Could Establish a New Category: Allied-Nation AI

The merger with a German model provider is more than a business development move — it's an attempt to create a sovereignty-compliant AI stack that European governments and enterprises can adopt without depending on U.S. hyperscalers. If this model proves viable, it could define a new vendor category in regulated industries and government procurement.

"We want to build this alliance to help create capability beyond just one democracy." — Aidan Gomez