KDCube
← AI Industry News

Oracle & NVIDIA Ship Agent Builders and Grounding; Daytona Hits $1B

Oracle launched AI Agent Studio for Industry Applications and NVIDIA added an Agent Knowledge Foundation at GTC Berlin, pushing enterprise agents toward governed, data-grounded builders. Meanwhile Daytona raised $60M at a $1B valuation for its agent-sandbox cloud, and Microsoft's Agent Framework 1.22 s...

Lead signals

Highlights

  • Oracle shipped AI Agent Studio for Industry Applications, a no-/low-/pro-code environment to build, test, orchestrate, and manage agent teams across retail, construction, finance, and utilities — U.S. availability planned within 12 months (Oracle).
  • At GTC Berlin, NVIDIA added an Agent Knowledge Foundation to its Agent Toolkit, bundling Nemotron Retriever, a new Auto-Ontology library, and the Kumo structured-data models so agents can use company data without moving it or changing governance controls (The New Stack).
  • Daytona raised a $60M Series B at a $1B valuation to scale its "agent cloud" of isolated sandboxes — the computer every agent needs — with OpenAI and Anthropic among its users (Fortune).
  • Microsoft's Agent Framework python-1.22.0 carries a persisted-state format change that breaks naive rolling deployments — a real pin-and-migrate chore for anyone running it in production (GitHub).

Key Signals

  1. Oracle turns Fusion's agent builder into a cross-industry studio Oct 21

    Building on AI Agent Studio for Fusion Applications, Oracle's new industry edition lets customers and partners assemble agents against existing data, tools, and workflows — e.g., construction agents that flag milestone risks and propose mitigations, or financial agents handling payments and affordability analysis (Oracle). It squarely targets the same builder lane as Copilot Studio, Vertex Agent Builder, and Bedrock AgentCore.

  2. NVIDIA bets open models fix the context gap Oct 21

    The Knowledge Foundation pairs Nemotron Retriever embeddings — NVIDIA claims a 65% search-relevance lift on an OpenSearch setup with negligible added latency — with an Auto-Ontology layer that maps raw data into business concepts (customers, products, orders) and the CUDA-X data libraries (The New Stack). For RAG teams, it's a grounding stack pitched to sit beside existing governance rather than replace it.

  3. The "agent cloud" gets a unicorn Oct 20

    Daytona's $60M round (total funding now $91M) funds expansion beyond its sandbox product into broader compute for training, evaluation, and production agents, with a founder-cited ~25x revenue jump since January (Fortune). The thesis — millions of isolated per-agent computers — echoes the sandboxing push behind Perplexity's SPACE and Cisco/NVIDIA OpenShell.

Why It Matters / What To Watch

  1. Builder surfaces are consolidating around governed, data-grounded agents.
    • Weigh Oracle's studio against your current stack if you already run Fusion data — the no-code-to-pro-code gradient and 12-month U.S. timeline affect buy-vs-build timing (Oracle).
    • Treat NVIDIA's "no data movement, no governance change" claim as a procurement checklist item: verify ontology mapping and retriever relevance on your corpus, not the OpenSearch demo (The New Stack).
  2. Agent infrastructure and runtime upgrades carry operational fine print.
    • If you deploy Microsoft Agent Framework, pin versions: 1.22 runtimes write hosted/merge_mode records older workers can't read, so don't mix versions on one store or downgrade without migrating (GitHub).
    • Watch the agent-cloud capacity race — Daytona's raise signals demand for isolated execution is outrunning in-house sandboxing for teams running many agents (Fortune).

Quick Links