Highlights

  • $350 million raised to build the whole AI stack in-house: models, compute, storage, and a blockchain that logs what each computation did.
  • Sealed hardware runs the job and writes a receipt to the blockchain, so outsiders can check the claim rather than trust it.
  • The buyers CEO Michael Heinrich wants: institutions, hospitals, governments. One partner has already put 2.7 million agents on the chain.

Zero G Labs has raised $350 million across multiple rounds, starting from a 2023 pre-seed round that was supposed to raise $5 million and grew considerably beyond that, to build what CEO Michael Heinrich describes as a full, vertically integrated stack for verifiable AI: its own models, a compute network, decentralized storage, and a blockchain layer to record cryptographic proof of what actually happened at every stage of an AI computation.

The technical core is trusted execution environments, or TEEs: physical hardware, including specific Nvidia GPU configurations, that let a computation run in an isolated, tamper-resistant environment and then write an attestation of that computation onto the blockchain. "Everybody else can verify that what you said happened actually indeed happened," Heinrich said on the DeAI News Journal. The pitch: institutions, hospitals, or governments can run AI workloads while proving, not just claiming, that specific data wasn't used for training, or that a model ran exactly the code it was supposed to.

Zero G isn't just infrastructure for outside developers; the company says it runs its own tooling internally, including for chat applications and building its own models on its own compute. One ecosystem partner, Neosol, has launched roughly 2.7 million agents on Zero G's chain, according to Heinrich, more than any competing chain, he said, doing tasks like agent-to-agent conversation, transactions, and research retrieval.

HOW AI WAS USED IN THE PRODUCTION OF THIS PIECE: I wrote this story off a DeAI News Journal interview with Michael Heinrich and filed it to DeAI News through the Distro Publisher MCP connector, with Claude handling the house-style pass and the mechanics of filing. Any subsequent edits were made by hand in DistroVerse, Distro's built-in (non-AI) editing interface. Quotes and figures were checked back against the interview.