Everyone is fixated on how fast AI is advancing. Having spent the past few days with the 2026 Stanford AI Index, I think the more important story is a different one.
AI is not simply scaling. It is opening a widening gap between what the technology can now do and what most organizations are actually prepared to govern, absorb, and operate.
Six things are clear to me:
• Compute is compounding in a way few industries have ever lived through. Global AI compute capacity has grown thirty-fold since 2021 and has more than tripled every year since 2022, with Nvidia now behind more than sixty percent of it. This is not a trend line. It is a new cost of doing business.
• The race is no longer one country's story. The U.S. still leads in frontier models, but China is closing the gap fast, and already dominates where AI meets the physical world, installing nearly 300,000 industrial robots in a single year against a few tens of thousands each in Japan and the United States. Whoever leads in models and whoever leads in deployment may not be the same nation.
• The fastest gains are coming from agents. The steepest curves in the entire report are autonomous coding and computer-use. The next decade of value will come from AI doing work, not just drafting it, and that shifts the conversation from productivity tooling to operating model.
• The environmental cost has stopped being a footnote. Training a single frontier model can now generate north of 70,000 tons of CO₂-equivalent emissions, an order of magnitude beyond two model generations ago. Boards will be asked about this. They should be ready.
• Trust is the scarcest input. Public confidence in government's ability to regulate AI ranges from above eighty percent in Singapore to barely thirty percent in the United States, the same country leading the world in investment. Capital is not the constraint. Legitimacy is.
• And the humbling part: today's best models clear the hardest expert-level reasoning challenges, yet still stumble on things a child finds trivial, like reading an analog clock. Capability is advancing rapidly, but not uniformly, and anyone selling uniform progress is selling something.
My conviction after all of this:
The winners of the next decade will not be the organizations with the most AI. They will be the ones that combine capability with governance, human judgment, trust, and operational discipline.
Technology is accelerating. Institutional readiness is the harder problem, and it is the real source of advantage.
Credit: The data and findings above are drawn from "12 Graphs That Explain the State of AI in 2026" by Matthew S. Smith, IEEE Spectrum (April 2026), summarizing Stanford HAI's 2026 AI Index Report.
