Nvidia’s huang crowns openclaw the ‘new pc’ and dangles tokens as engineer bait
Jensen Huang stepped onstage in San José with the swagger of a man who no longer sells chips—he sells gravity. The Nvidia CEO told a packed GTC 2026 hall that every company on Earth now needs an OpenClaw playbook, a doctrine of autonomous agents that, he claims, will do for personal AI what Windows did for the beige box.
The agent in question began life as a weekend side-project called Clawdbot, later nicknamed Moltbot by Reddit memelords. Its creator, Austrian developer Peter Steinberger, was poached by OpenAI last winter, yet the code stayed behind, forked and fertilised by Valley startups until it grew into the open-source beast Huang now worships. He brandished a live demo: a cursor that booked flights, negotiated a refund, then rewrote its own backend while the crowd’s jaws drifted south.
The price of admission: a trillion dollars
Huang paired the sermon with a forecast so large it felt fictional. By 2027, he said, demand for Nvidia’s incoming Blackwell and Rubin AI processors will touch one trillion dollars, a figure that eclipses the GDP of oil states. The chips will chew through inference tasks served by a new engine borrowed from stealth startup Groq, promising latency so low it “feels like thought itself,” according to an engineer in the hallway who had already tested the rig.
But the CEO saved his sharpest hook for the people who will stitch these agents together. Starting this quarter, Nvidia will pay key hires partly in compute tokens—internal currency redeemable for GPU time. A senior engineer might see half her cash salary replaced by tokens worth quadruple their face value in raw flops. Huang’s pitch: take the tokens, spin up 10× more experiments, ship 10× faster. Recruiters at the conference dubbed the scheme “equity 2.0,” except this equity depreciates the moment someone else trains a bigger model.
Security enters through the side door
For all the bravado, Huang conceded that letting agents roam corporate networks is a “software suicide drill” without guardrails. Nvidia’s answer is NemoClaw, a hardened fork that drapes OpenClaw in privacy filters, network micro-tunnels and policy routers that can choke an agent’s internet access mid-thought. Early adopters—JPMorgan, Samsung SDS and a pharmaceutical giant whose NDA still hides its name—will begin pilot tests next month. If an agent misbehaves, a red kill-switch evaporates its container in 300 milliseconds, faster than a human blink.
Outside the convention centre, a line of developers snaked around the block for a 3 a.m. hackathon where the prize was a single Rubin node for a year. One Stanford dropout clutched a sleeping bag and a sign scribbled with marker: “Will trade kidney for tokens.” Security let him keep the sign but confiscated his camping stove.
The after-party raged in a warehouse scented by vape smoke and ambition. A drone hovered overhead projecting a live ticker: 847 billion tokens burned today. Investors huddled in corners, whispering valuations that double every ten minutes. By sunrise, the ticker had crossed the trillion mark, and Huang’s words echoed off the steel rafters: “The new computer is not a box. It’s a colony of minds that never sleeps.” The engineers cheered, then opened their laptops to feed the colony—paid, this time, in the same currency their creations consume.