Defense AI, travel agent AI, AI inference infrastructure, and a SpaceX compute deal joined the cohort this quarter. The mix is broader than 2024–25.
Five big AI rounds closed in the same quarter, and they didn't all chase the same thing. That's the read underneath the wire coverage: in Q2 2026, private AI capital stopped concentrating on foundation models and started funding the rest of the stack at the same time.
For most of 2024 and 2025, the biggest AI rounds by dollar volume were foundation-model checks. OpenAI, Anthropic, xAI, Mistral. The capital flowing into private AI tracked the model layer. Q2 2026 broke that pattern. Anthropic's $65B Series H, announced in late May at a $965B post-money valuation, is still the year's biggest private AI round; it was led by Altimeter Capital, Dragoneer, Greenoaks, and Sequoia Capital. Anthropic also confidentially filed a draft Form S-1 with the SEC on June 1, the standard step to keep an IPO option open without committing to one. Earlier in July the company said run-rate revenue crossed $47B, a self-reported figure that has not been independently audited.
The other four major rounds in the cohort didn't go to model makers. Anduril's $5B Series H is funding defense AI: manufacturing capacity, autonomous systems, and next-generation military products. The category had been a slow burn; the round reset the bar. Three more rounds, tallied by Benzinga: Long Lake closed $2.25B while pursuing an acquisition of American Express Global Business Travel, a wager that AI workflows can rewire a legacy services business. Baseten raised $1.5B for its enterprise AI inference platform, the layer that serves trained models to paying customers at production load. Reflection AI's $2.5B reportedly paired with a SpaceX partnership to support an AI compute buildout inside SpaceX.
That mix: foundation models, defense, services AI rollup, inference infrastructure, and applied AI for a major incumbent. It is the broadest single-quarter AI round set the private market has produced. The individual releases came company by company; the cohort pattern only shows up when the deals are read together.
For two years, private AI capital tracked the model layer: whoever could train the smartest model drew the round. Q2 2026 distributes capital across five layers at once, which means the allocators are hedging the bet. If a foundation-model company stumbles at the next benchmark cycle, it's no longer a portfolio-level event for LPs writing AI checks, because the same quarter's deals also sit in defense, inference, and services.
There's a second-order effect on the public side. A $47B run-rate business is large enough to clear the disclosure thresholds public investors demand, and an S-1 on file means the IPO option is open, not imminent. A successful Anthropic listing would reset price discovery for every private AI company left standing. The non-foundation-model ones will trade off their own pipelines: Anduril against defense backlogs, Baseten against Replicate, Modal, and the in-house serving stacks, not against any headline benchmark score.
Each of the new layers is still small. Anduril has shipped products, but defense AI as an investable category is early. Baseten competes with Replicate, Modal, Together, and a stack of in-house serving platforms, none of them dominant. Long Lake's Amex GBT pursuit remains an unclosed deal. Watch Q3: another defense round or another inference round at this size would confirm the Q2 cohort as a structural broadening rather than a one-off.