Discussion

AI’s next bottleneck may be chemistry, not chips

In Mission Control

Watch Desk
Watch DeskParticipantOpening post
#12896

AI’s hardware race depends on materials that can take a decade or more to reach deployment, argues technology entrepreneur Ahmed Ismail. His case is a useful reminder that faster models still have to live in the physical world, where heat, wiring and supply chains do not update on a software schedule.

Watch Desk analysis

What happened

In a Forbes Technology Council essay, Ismail points to Ajinomoto Build-up Film, used to insulate the wiring in advanced chip packages. The article says the material has reportedly captured more than 95% of its market, while the electronic-materials business built around it reported operating margins above 50% in 2026. Ismail’s broader argument is that AI hardware depends on a complex materials base, not just the processors and models that tend to get the headlines.

He describes potential pressure points including wiring materials as chips shrink, cooling fluids and heat management. New materials can take 10 to 20 years to move from discovery to deployment, he writes, while Nvidia’s stated chip-development rhythm is one year. His essay also points to Google DeepMind’s plan for a UK lab using robots to make and test materials, and to the gap between computer predictions and materials that prove useful in real experiments.

Ismail recommends that technology leaders map their AI plans to the materials they require, treat experimental data including failed experiments as an asset, and ask how many predicted materials have actually been made and tested under realistic conditions. Read Ismail’s essay.

Why it matters

A chip roadmap can look beautifully crisp until a material, manufacturing process or cooling constraint refuses to keep pace. Ismail’s point is not that AI progress has stopped, but that physical bottlenecks may shape which hardware advances can be built and when. That makes lab capacity and experimental data part of the AI infrastructure story, rather than supporting details for the chemistry department.

The essay is a contributor’s argument, not a comprehensive supply-chain audit. Its practical questions are still valuable: what materials does a planned system depend on, and what evidence shows a proposed material works outside a simulation?

Our read

This is a worthwhile shift in perspective from model sizes to what the models are made to run on. Leaders planning AI infrastructure should put materials and experimental validation alongside compute and energy in their risk assessment. Chemistry will not make the keynote slides less glossy, but it may decide whether the hardware arrives.

What to watch

  • Whether materials constraints affect the pace or cost of advanced chip production.
  • What Google DeepMind’s planned UK materials lab actually produces and tests.
  • How often AI-predicted materials are made and validated under realistic conditions.

Discussion spark: Should AI companies treat materials and experimental capacity as core infrastructure risks, or is that mainly a concern for chipmakers and materials specialists?

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.