←21 Oct 2026 · Vinod Bijlani

I flew into Taiwan thinking about chips. I flew out thinking about tokens.

The industry is solving the compute problem. Enterprises are now facing the consumption problem, and that needs a governed token factory.

Agentic AIAI economicsToken factory

Just back from the mecca of semiconductors.

TSMC is working on 1.4 nm chips, with the A14 process planned for volume production in 2028. NVIDIA and AMD are bringing rack-scale systems to market. I flew into Taiwan thinking about chips. I flew out thinking about tokens.

Over four days in Taipei I had around twelve customer conversations across technology, telco, financial services and the AI ecosystem. I expected to talk about GPUs and AI factories. But most customers already have some AI capacity. Some own GPUs. Some rent them. Many buy tokens from the public cloud.

The next problem is running it at scale:

  • Who is consuming the tokens?
  • Which models and agents are allowed?
  • What does each useful outcome actually cost?

That's when it clicked. The industry is solving the compute problem. Enterprises are now facing the consumption problem.

And it only grows from here: thousands of agents, millions of prompts, billions of tokens. Capacity is no longer the gap. Control is.

That takes an operating model that can meter consumption, govern models and agents, secure data access and measure ROI in real time. That is the governed token factory: the layer that turns compute into business outcomes, with visibility, accountability and control.

This trip didn't just show me where AI is made. It showed me where AI gets hard.

In your view, what's the real gap in enterprise AI today: compute, consumption, governance or ROI?

Read moreForbes: Every CIO should consider a hybrid token factory ↗Forbes: Which AI workload goes where? ↗
GPU plus model is not a token factory: ten capabilities a token factory needs
Infographic · tap to enlarge

Invite Vinod to speak, write or advise.

Keynotes, panels and executive briefings on AI factories, sovereign AI and AI economics. Send a message on LinkedIn.