VCs sketch the deep-tech stack: full-stack AI economics, Europe's hypersonic deterrent, and post-quantum at laptop speed

The full-stack squeeze

Mayfield's fiftieth issue of its AI newsletter reads like a field report from the supply chain. The biggest builders in AI are increasingly adopting full-stack strategies — co-designing models and silicon to optimize economics and speed — while compute demand keeps outpacing supply. The striking observation is that the bottleneck is not one thing: it rotates between power, compute, memory, silicon, and land. Solve one and the constraint pops up at another layer. Feedback loops between customers and silicon suppliers have shortened to weeks. m:…9590

The numbers Mayfield cites put scale on the trend: Microsoft plans to begin multi-gigawatt deployment of its Maia 300 inference systems in 2027, and TSMC reported a 69% increase in third-quarter net profit. The newsletter's advice to founders follows the bottleneck logic — focus on problems brute force cannot solve, such as power, cooling, and data movement. It is a VC telling founders that the alpha is in the constraints, not the models. m:…9590

Europe's deterrent bet

General Catalyst's blog this week is a different kind of deep-tech thesis: sovereignty as a venture category. Hypersonica co-founder Philipp Kerth makes the case for an independent European hypersonic deterrent, arguing that Europe neglected the underlying hypersonic research for 30 years and that true sovereignty means the ability to make strategic decisions independently. The company aims to have deep-strike systems in the market by 2029, and claims it can go from scratch to a finished prototype in nine months — startup velocity applied to strategic weapons. m:…8677

The engineering context is unforgiving — hypersonic flight means extreme temperatures from friction and compression — which is precisely why it is interesting as a venture bet: the difficulty is the moat. Whether a nine-month prototype cycle survives contact with hypersonic physics is the open question the blog does not answer. m:…8677

Post-quantum at laptop speed

a16z crypto's contribution is cryptographic infrastructure for the quantum era. Aerie, its new tool, aggregates lattice-based signatures using lattice-based SNARKs, supporting the NIST-standardized Falcon scheme so blockchains can verify signatures at high throughput without the complexity of recursive proof merging. The headline number: Aerie aggregates 32,768 Falcon-512 signatures in 0.65 seconds on an Apple M4 Max laptop, producing a single proof directly. Post-quantum schemes run 10 to 100 times larger than current elliptic-curve signatures, which is why aggregation matters — without it, the quantum-safe future does not fit on a chain. m:…9819

Note the framing a16z chooses: lattice-based cryptography is necessary for private payments because hash functions cannot support public-key encryption. This is infrastructure research presented as an inevitability argument — the chains that survive the quantum transition will be the ones whose signature economics work, and a16z wants its portfolio's chains to be those. m:…9819

Why it matters

Three investor desks, one pattern: the money is moving to the layers under the model — the power and silicon constraining AI's economics, the strategic weapons Europe forgot to build, the signature schemes that survive quantum computers. None of these are model bets. For VCs, the signal is that deep tech's returns are being found in constraints: whoever solves power, whoever ships deterrence, whoever makes post-quantum cheap. For founders, Mayfield's line is the actionable one — build where brute force fails. m:…9590 m:…8677 m:…9819