Palo Alto – March 24, 2026 – WCP is delighted to release our AI tech report From Consumers to Enterprises to LLM Providers: Why AI Agents Are Forcing a Reckoning on Token Economics, authored by Managing Director, Partner Mark Bagley.
A new category of computing is emerging where the primary user of a personal computer is not a human but an AI agent, operating 24/7 on your behalf. The implications ripple across every layer of the technology stack: consumers need affordable always-on agents, enterprises need secure ones, and LLM providers face an existential pricing problem. Agents consume 100x more tokens than humans. The current cloud API pricing
model doesn’t work. The industry is being forced to find other ways.
“What’s your OpenClaw strategy?” : Jensen Huang, Nvidia CEO, GTC 2026 Keynote
Nvidia’s CEO posed that question to a packed GTC audience on March 16, 2026, backed by a projected $1 trillion in chip orders through 2027. The real story isn’t one company’s keynote: it’s the convergence. Nvidia and AMD are building dedicated agent hardware. Apple accidentally built the perfect agent computer. OpenAI hired OpenClaw’s creator. Anthropic constructed an enterprise agent ecosystem. China’s adoption has already
surpassed the US, with government subsidies and street-level installation events. Security professionals called it ‘unacceptably dangerous’: and everyone invested anyway. When this many players converge this fast, the question stops being whether agents will reshape computing and becomes how the economics will work.
Huang declared that OpenClaw spells a new path for the software industry: agents-as-a-service rather than software-as-a-service. If he’s right, every software company on earth needs an answer to his question. But here’s the catch: agents running on cloud LLM APIs at $3-15 per million tokens generate costs that scale linearly with usage: and agents use far more tokens than humans. The economics only work if token costs collapse. Today, the only path to near-zero token costs is open-source models on local hardware. Cloud
providers may eventually close the gap through inference-optimized chips, volume pricing, or new business models: but right now, local open-source is the only option that makes agent-scale workloads affordable. The companies profiled in this report are all, in different ways, racing to solve that equation.
