Three deals inside a ten-day window in June 2026 point to the same shift: AI money is moving out of the model layer and into the infrastructure and tooling that makes models usable inside a business. Baseten raised $1.5 billion for inference. Odyssey raised $310 million to build world models. OpenAI bought a company to make its coding agent work inside enterprise cloud environments. None of the three is a chatbot story.
Baseten raises $1.5 billion to run other people's models
Baseten, an AI inference platform, announced a $1.5 billion Series F on June 22, led by Altimeter Capital, Conviction and Spark Capital, with Sands Capital and Wellington Management co-leading. Battery Ventures, Blackbird, D.E. Shaw Ventures, Durable Capital Partners, Greylock, IVP, Verified Capital and 01A also participated, according to the company's own announcement and a same-day release distributed via Businesswire.
The round was structured across two tranches, valuing the company at $11 billion and $13 billion. That is a roughly 160 percent increase from the $5 billion valuation Baseten carried five months earlier, when it raised a $300 million Series E in January 2026.
Baseten does not build foundation models. It runs them, hosting inference workloads for companies including Cursor, Notion, HubSpot, Decagon, Clay, Lovable and Abridge. The company says its platform now processes more than 1 billion inference calls a day across 87 clusters on 18 clouds, and that revenue grew 20 times year over year.
What changed: investors are now willing to fund the plumbing layer at valuations that rival foundation-model labs themselves. The pitch is not a smarter model. It is reliably serving whatever model a customer already picked, at the volume a production application demands.
Odyssey raises $310 million for world models, not chatbots
Five days earlier, on June 17, Odyssey announced a $310 million Series B led by Natural Capital, valuing the company at $1.45 billion. Amazon, AMD Ventures, GV, EQT and IQT participated, alongside existing backers including Jeff Dean, Elad Gil and Garry Tan, according to Odyssey's own announcement and coverage confirming the round's terms.
Odyssey builds world models, systems trained to simulate how objects, people and environments interact, rather than generate text or images. The company, founded by veterans of the self-driving car industry, said its applications target robotics, gaming and scientific simulation. As part of the round, Odyssey also named Amazon Web Services its preferred cloud provider.
What changed: a second major capital pool is now backing a category of AI system distinct from the large language models most enterprise buyers evaluate today. Whether world models reach commercial production at the scale LLMs have is not yet established. What the round shows is that investors are willing to fund that bet at a billion-dollar-plus valuation before the answer is known.
OpenAI buys Ona to make Codex work unattended
On June 11, OpenAI said it had agreed to acquire Ona, a startup that builds secure, persistent cloud environments where AI agents can keep working after a developer closes their laptop. The deal was confirmed by OpenAI's own announcement and reported the same day by Bloomberg and CNBC. Terms were not disclosed.
Ona's technology will be folded into Codex, OpenAI's coding agent. OpenAI said more than 5 million people use Codex weekly, up 400 percent from earlier in the year, and that the acquisition is meant to let Codex handle longer-running tasks inside enterprise cloud setups rather than requiring a developer to keep a session open.
What changed: OpenAI is buying its way into the infrastructure problem that comes after a coding agent gets good enough to be trusted with a real task, namely where that agent runs, how it holds credentials, and what happens when nobody is watching it. That is an operations problem, not a model-capability problem, and OpenAI chose to acquire rather than build it in-house.
What to watch next
Baseten's Series F, Odyssey's Series B and the Ona acquisition were all announced within the same ten-day stretch, and none involved a new frontier model. The next signal to watch is whether inference-layer companies keep raising at valuations close to the model labs they serve, which would confirm that enterprise AI spending is settling into infrastructure rather than concentrating at the model layer. Baseten has not disclosed a timeline for deploying the new capital beyond "compute, software and talent." The OpenAI-Ona deal has not yet closed; its completion, and any detail on how Codex changes for enterprise customers once it does, is the next concrete milestone to confirm.
Sources: Baseten · Businesswire · Businesswire · Odyssey · Las Vegas Sun · OpenAI · Bloomberg · CNBC