Analysis
Three acquisitions closed or were agreed within days of each other, and together they say something about AI infrastructure exits that no single deal shows on its own: Nscale agreed to buy Anyscale, the commercial steward of the open-source Ray framework, for roughly $1.65 billion. Okta agreed to buy AI identity-security startup Permiso for roughly $200 million. And IonQ completed its $1.8 billion acquisition of chip foundry SkyWater Technology after a deadlocked FTC let the deal proceed unconditionally.
Each deal looks different on the surface -- orchestration software, identity security, semiconductor fabrication -- but the underlying logic is identical: an incumbent with existing infrastructure, distribution or capital is buying a specialized capability outright rather than waiting for that target company to reach an independent public listing, or building the capability in-house. Nscale gets Anyscale's engineering team and enterprise customer base (Coinbase, Runway, Bedrock Robotics) instead of building LLM orchestration internally. Okta gets a working AI-agent-sandboxing product that only publicly launched months earlier, rather than spending a year or more building comparable tooling from scratch.
“Nscale gets Anyscale's engineering team and enterprise customer base (Coinbase, Runway, Bedrock Robotics) instead of building LLM orchestration internally.”
For founders in AI infrastructure and security specifically, this is worth internalizing early: the fastest, most reliable path to a real return in this cycle increasingly runs through being acquired by a company that already has the customer relationships and balance sheet to close quickly, not through an independent IPO process that can take years and depends on public-market appetite holding up. Permiso went from public launch to acquisition in a matter of months -- that's not a company that ran out of independent options, it's a company that had a strategic buyer with obvious synergies show up fast.
The risk for the broader ecosystem is concentration: if the largest, best-resourced infrastructure and platform companies keep absorbing the most capable specialized startups before they reach real independent scale, fewer AI infrastructure companies actually make it to a public listing on their own, and the category's public-market comparables stay thin. What to watch: whether this M&A pace continues through the rest of 2026, and whether any of the three acquirers (Nscale, Okta, IonQ) eventually bundles its acquired capability into a broader platform pitch ahead of its own IPO.
