One-Line Summary

Capital and compute consolidate around frontier labs as OpenAI raises $122B, AWS deepens ties with OpenAI, Nvidia debuts a CPU for agents, and Google ships an enterprise agent platform.

Big Tech

OpenAI raises $122B to scale compute and an AI 'superapp'

OpenAI, the company behind ChatGPT and widely used developer APIs, closes $122 billion at an $852 billion valuation to accelerate “the next phase of AI,” aiming to turn research advances into products people use at work. ¹

The round is anchored by Amazon, Nvidia, and SoftBank, with continued participation from Microsoft; OpenAI also raises over $3 billion from individual investors via bank channels, will be included in ARK Invest ETFs, and expands its revolving credit facility to approximately $4.7 billion. ¹

OpenAI frames compute as its core advantage and outlines a broader infrastructure portfolio across multiple clouds (Microsoft, Oracle, AWS, CoreWeave, Google Cloud) and chips (Nvidia, AMD, AWS Trainium, Cerebras, plus a Broadcom‑partnered in‑house chip), alongside a plan to unify ChatGPT, Codex, search, and browsing into one agent‑first “superapp.” ¹

The company reports more than 900 million weekly active users, over 50 million subscribers, $2 billion in monthly revenue, enterprise contributing over 40% of revenue, APIs processing 15 billion tokens per minute, and an ads pilot surpassing $100 million in ARR — positioning consumer scale as a funnel into workplace adoption. ¹

Nvidia unveils Vera CPU to speed agentic AI

Nvidia introduces Vera, a data center CPU designed to handle the non‑GPU parts of AI work — like orchestration, tool calls, and reinforcement learning loops — that have become the new bottleneck as described by Amdahl’s law. ²

Vera features 88 custom Arm‑based Olympus cores, up to 1.2 TB/s of memory bandwidth, and claims up to 50% faster “sandbox” performance than traditional CPUs, while linking to Rubin GPUs over NVLink‑C2C with up to 1.8 TB/s of coherent bandwidth for rapid hand‑offs between CPU and GPU. ³

A rack configuration integrates up to 256 liquid‑cooled Vera CPUs to sustain more than 22,500 concurrent CPU environments, and systems will be available from major OEMs including Dell, HPE, Lenovo, and Supermicro in the second half of 2026. ³

For users, the pitch is shorter wait times and steadier latency when agents compile code, browse the web, or process data — benefits Nvidia ties to higher single‑core speed, deterministic execution, and reduced pressure on key‑value caches during agentic inference. ²

OpenAI partners with Amazon on stateful runtime and Trainium scale

OpenAI and Amazon agree to co‑create a “Stateful Runtime Environment” powered by OpenAI models and delivered through Amazon Bedrock, while AWS becomes the exclusive third‑party cloud distributor for OpenAI Frontier, the company’s enterprise platform for building and managing teams of AI agents. ⁴

The deal includes OpenAI consuming about 2 gigawatts of AWS Trainium capacity and expands an existing $38 billion multi‑year agreement by $100 billion over eight years; Amazon also commits a $50 billion investment in OpenAI ($15 billion upfront and another $35 billion pending conditions). ⁴

For AWS shops, this points to tighter integration of OpenAI’s agent capabilities with Bedrock’s AgentCore and security controls, with the stateful environment expected to launch in the coming months. ⁴

Industry & Biz

Anthropic buys Vercept to advance computer-use in Claude

Anthropic, the company behind Claude, acquires Vercept, a startup focused on “computer‑use” agents that operate macOS environments by seeing and clicking like a person, signaling a push to make Claude perform tasks inside live apps. ⁵

Vercept has raised $50 million to date and will shut down its Vy product on Mar 25 with a 30‑day migration window, while Anthropic highlights Claude Sonnet 4.6 reaching 72.5% on the OSWorld benchmark for desktop task completion. ⁵

Co‑founders Kiana Ehsani, Luca Weihs, and Ross Girshick join Anthropic, though not all founders transition; the deal follows Anthropic’s earlier acquisition of the Bun coding‑agent engine, with financial terms undisclosed. ⁵

New Tools

Google launches Gemini Enterprise Agent Platform for workplace agents

Google Cloud debuts the Gemini Enterprise Agent Platform, described as the evolution of Vertex AI, to help companies build, scale, govern, and optimize AI agents that can be delivered to employees through the Gemini Enterprise app. ⁶

The platform spans build tools like Agent Studio and an upgraded Agent Development Kit, scale features such as long‑running Agent Runtime with Memory Bank, governance with Agent Identity/Registry/Gateway, and quality controls via Agent Simulation, Evaluation, and Observability. ⁶

Model access includes 200‑plus options in Model Garden — from Gemini 3.1 Pro, Gemini 3.1 Flash Image, and Lyria 3 to open models like Gemma 4 and third‑party choices such as Anthropic’s Claude — and Google says future Vertex AI services and roadmaps run exclusively through this platform. ⁶

What This Means for You

OpenAI’s raise and product direction suggest more AI will show up inside the tools you already use — not just as chat, but as agents that remember, browse, and execute multi‑step work, with a single “superapp” surface tying it together. ¹

If your stack is on AWS, the OpenAI–Amazon agreement implies you could align OpenAI agents with Bedrock’s permissions, billing, and monitoring once the stateful runtime and Frontier access arrive, potentially lowering adoption friction. ⁴

If you build on Google Cloud, Gemini Enterprise Agent Platform centralizes agent development and oversight, giving business users a low‑code path while IT retains control through identity, registry, and gateway guardrails. ⁶

On performance, Nvidia’s Vera CPU targets CPU‑bound steps like code runs and data prep in agent loops; as these systems roll out, expect snappier tool use and fewer stalls around the model’s “glue” work. ²

Action Items

1. Map one workflow to an agent pilot: Pick a repetitive process (e.g., monthly report assembly) and outline inputs, guardrails, and a success metric your team agrees on.
2. Try Google’s Agent Studio: If you have Gemini Enterprise access, prototype a simple agent in Agent Studio and route it through IT review to learn the governance flow.
3. Brief your AWS owner on OpenAI–Amazon: Identify one use case that could benefit from Bedrock integration and ask your AWS account team about access to the stateful runtime and Frontier.
4. Baseline agent wait times: Time CPU‑heavy steps in your current agent flows (build/test, data transforms) so you can assess impact as infrastructure like Nvidia Vera becomes available.