Nvidia corners the AI agent stack
Nvidia's COMPUTEX 2026 keynote centered on 'agentic AI' with new hardware and models positioning AI agents as the primary consumers of future compute. The newsletter also covers Bernie Sanders' proposed AI sovereign wealth fund, a Meta Instagram security exploit via AI chatbot, and various other AI industry updates.
Summary
At COMPUTEX 2026, Nvidia CEO Jensen Huang declared 'Agentic AI has arrived,' unveiling a sweeping set of products designed to make AI agents — not human users — the central consumers of compute power. Key announcements included RTX Spark supercomputer chips co-developed with Microsoft to run AI agents on PCs, the Vera CPU (claimed to be 1.8x faster than rivals and already adopted by Anthropic, OpenAI, and NYSE), Cosmos 3 (an open robotics model enabling proactive planning for robots and self-driving cars), and Nemotron 3 Ultra (a 550B parameter open-source model competing with Chinese alternatives like Qwen3.5 and Kimi K2.6). The newsletter frames Nvidia's coordinated stack-wide approach as unique in the industry, with the company's $5T+ valuation now organized around software categories that barely existed two years ago.
On the policy front, Senator Bernie Sanders previewed the American A.I. Sovereign Wealth Fund Act, which would transfer 50% of equity in the largest AI companies into a public fund, distributing gains to ordinary Americans. Sanders drew parallels to Norway's $2T oil fund and Alaska's oil dividend program, arguing that AI is built on humanity's collective knowledge and should benefit the public, not just investors. The newsletter notes the practical difficulty of getting AI labs to relinquish half their equity despite the bill's populist appeal.
A significant security story emerged around Meta, which had granted its AI assistant the ability to handle password resets on Instagram and Facebook since March. Hackers exploited this by simply asking the AI chatbot to change account email addresses and issue reset codes, successfully compromising high-profile accounts including a dormant Barack Obama account, Sephora, and a Space Force general. Meta stated the exploit has been resolved. The newsletter highlights this as a cautionary example of delegating sensitive infrastructure functions to easily manipulated AI tools.
Other notable items include: Anthropic filing with the SEC to go public, OpenAI beginning construction on a 1 GW data center in Michigan, Florida's AG filing the first state-level lawsuit against OpenAI alleging ChatGPT's role in planning violence, MiniMax releasing its open-weight M3 model, and an Apollo economist claiming there is 'zero evidence' of AI-caused job losses.
Key Insights
- Nvidia is explicitly reframing AI agents — not human users — as the primary consumers of compute power, with its entire product lineup now organized around enabling agentic software that didn't exist two years ago.
- Bernie Sanders argues that AI is built on 'the accumulated knowledge, creativity, and labor of mankind,' framing it as a public resource analogous to oil and citing existing sovereign wealth funds as viable models for public benefit.
- Meta's decision to hand over password reset authority to its AI chatbot created a months-long exploit window where hackers could take over high-profile accounts simply by asking the AI to send reset codes to a new email address — no sophisticated hacking required.
- Nvidia's Vera CPU is already in production use by Anthropic, OpenAI, and the NYSE, suggesting that frontier AI labs are adopting Nvidia's agent-focused silicon beyond just GPU training workloads.
- Apollo's chief economist claims there is 'zero evidence of job losses because of AI,' arguing that cheaper technology increases demand and creates more jobs — a claim that stands in notable contrast to widespread public concern about AI-driven displacement.
Topics
Full transcript available for MurmurCast members
Sign Up to Access