🕶️ Anthropic just moved into Meta's

Good morning, team. Let's huddle.

Two deals this week say the same thing: access beats ownership, and infrastructure just became the real product. The biggest labs are negotiating compute capacity like it's office space. The smartest capital is backing tools that simulate what used to require fifteen years and a lab.

If you're still pitching on product alone, the conversation has already moved to what powers it, who provides it, and whether you can afford to keep building at all.

Sign up for a free trial of Attio, the agentic CRM.

Ask Attio to build a daily workflow that surfaces the deals that need your attention today, like anything with a stage change, a recent reply, or a new signal in the last 24 hours.

Review your pipeline in Claude, synced live from Attio via MCP.

That's it.

Try Attio Now

When Jeff Bezos, Kleiner Perkins, and Nvidia all converge on the same startup, it's worth asking what they see that the rest of the market doesn't.

CuspAI, a British AI materials startup born out of Cambridge, just raised $450 million at a $2.6 billion valuation. The round was co-led by Kleiner Perkins and NEA, with Lux Capital, AMD Ventures, Glade Brook Capital Partners, Britain's Sovereign AI Venture Fund, and Bezos Expeditions all piling in. Three years ago, this was a research concept. Now it's a platform that simulates novel materials so well that physical lab testing starts to look like a rounding error.

The real signal is the partnership roster. CuspAI launched what it calls an "AI Materials Foundry," a pipeline connecting compute, data, labs, and scientific expertise. Nvidia is providing compute. Meta's Fundamental AI Research team joined. 45 organizations are already plugged in. That's not a startup announcement; that's an emerging standard.

Why it matters for builders: Industrial R&D still burns 70%+ of innovation budgets on physical experiments, with a 15-year average time to market for new compounds. If CuspAI's simulation layer works at scale, every hardware startup designing around legacy materials is building on a foundation that could shift quarterly, not generationally.

Looking ahead… the first public output from that Foundry, specifically which semiconductor or clean-energy compounds get fast-tracked, will tell us whether this is infrastructure or hype.

Meta is reportedly in talks to lease $10 billion worth of GPU capacity to Anthropic. If it closes, it's one of the largest compute transactions ever, and a clear sign that raw infrastructure access has become the most strategic asset in AI.

Think about how fast this shifted. Two years ago, building a frontier model meant pouring billions into your own data centers. Nine months ago, the first major labs started openly valuing access over ownership. Now a $10 billion operational lease is getting negotiated like it's a routine procurement deal.

The second-order effect no one's talking about: mega-leases like this create a two-speed compute economy. Hyperscalers lock in multi-year, multi-billion dollar commitments. Early-stage startups get pushed to the back of the GPU line, stuck paying spot-market premiums that make training runs borderline unaffordable.

For founders, the takeaway is structural. If you're building at the model layer, your fundraising story now needs a credible compute strategy—a hyperscaler relationship, a second-tier data center partnership, or an architecture designed from day one to be ruthlessly compute-efficient. The era of scrappy training runs on borrowed credits is ending fast, and no amount of optimism is going to change the math.

Scale your growth: Join names like HubSpot, Ahrefs, and Deel. Put your product in front of 85,000+ startup founders. Let's talk.