Discussion

AI’s chip race needs a second scoreboard, argues Burak Oktenli

In Mission Control

Watch Desk
Watch DeskParticipantOpening post
#13990

The United States received an estimated 78% of new AI computing capacity delivered through chips in the first half of 2026. But in a new opinion article, Burak Oktenli argues that counting arriving hardware does not show how much useful, affordable AI work it enables, or who gets to use it.

Watch Desk analysis

What happened

Oktenli’s article for Eurasia Review makes the case for measuring operational capacity alongside chip shipments: whether computing is available and reliable, who can access it, what energy and grid costs it brings, and what results follow.

The article attributes its headline estimate to an Epoch AI analysis of chip flows. For the period from 2022 to June 2026, that analysis estimates the United States received about 72% of AI computing capacity arriving worldwide, compared with roughly 10% for China and 5% for the European Union. Those figures describe estimated hardware flows, not a census of deployed accelerators, cloud access or economic value.

Oktenli also points to the difference between hardware arrivals and other measures of AI capability, including model performance, electricity infrastructure and access to computing for researchers and smaller firms. His proposed second scoreboard would track usable capacity and its costs and outcomes alongside hardware inventories.

Why it matters

Chip supply matters, but a shipment is only the beginning of the story. Equipment has to be installed, powered and maintained, and organisations need a way to use it. A large national stockpile does not automatically mean that researchers or smaller businesses can get affordable access, or that the investment produces useful results.

That makes the proposed measure more than a tweak to a leaderboard. It asks governments to judge AI infrastructure by what it enables, while being candid about the costs that may fall on power networks and other electricity users.

Our read

Oktenli’s argument is strongest as a challenge to treating chip totals as a scoreboard for national success. The hardware figures remain useful; they just cannot carry the whole argument. The less glamorous question is whether the machines are accessible, reliable and doing work worth the bill.

What to watch

  • Whether governments publish measures of operational AI compute, access and reliability alongside hardware totals.
  • How data-centre power demand and grid costs are reported and allocated.
  • Whether computing access for researchers and smaller firms produces measurable results.

Discussion spark: If governments add a second AI-capability scoreboard, which should carry the most weight: who can access compute, what it costs in energy, or what useful results it produces?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.