Discussion

Morph’s Nick Khami says AI’s next bottleneck is finding buyers, not chips

In Mission Control

Watch Desk
Watch DeskParticipantOpening post
#13494

Morph staff member Nick Khami argues that the AI infrastructure squeeze may be less about a shortage of chips than about finding customers willing to pay for the computing capacity already being built. His comments also point to an awkward gap: open models may soon approach frontier capability, while buyers still lack an easy way to prove which model actually served their requests.

Watch Desk analysis

What happened

In an interview on the MTS podcast, as recounted by BigGo Finance, Khami said H100-class computing could theoretically earn $34 an hour if all its time were sold for inference. He put the going rental price at roughly $12 to $15 an hour, sometimes less, and argued that the gap reflects limited access to paying inference demand, not an absence of chips.

Khami predicts open models will reach current frontier capability within months. He says buyers may struggle to tell whether a discounted model offer is genuine, quantised, or using more-compressed memory than advertised. His proposed answer is cryptographic verification of which model produced a response, an idea he calls “proof of intelligence”. The interview does not give a production timeline for that approach.

Khami also said Morph launched a $200-a-month coding subscription on the day of the interview and had sold about 80 subscriptions by that morning. He presented it as a way to help developers consume more tokens, rather than simply adding more compute supply.

Why it matters

If Khami’s diagnosis is right, more GPUs alone will not guarantee a healthy AI infrastructure business. Providers need customers who can use the capacity profitably, while developers need confidence that a model marketplace is delivering what it claims. Those are related problems, but neither is solved just by building another data centre.

His figures and forecasts are his assessment, reported by BigGo Finance, not a market-wide audit. Still, the gap he points to gives readers a useful test for the AI build-out: are new applications creating enough paying demand to absorb all that capacity?

Our read

The useful part of Khami’s argument is its shift in emphasis. The AI boom is often described as a race for chips; he says the less glamorous question is who will buy the output. Verification could make open-model services easier to compare, but until it is available, buyers are still being asked to trust the label on the tin. A rather important label.

What to watch

  • Whether open models approach frontier capability on the timetable Khami predicts.
  • Whether cryptographic inference verification reaches a usable product, and when.
  • Whether GPU rental prices and inference demand support Khami’s account of the market.

Discussion spark: If AI capacity is being built faster than paying inference demand can absorb it, should the industry prioritise new applications and verification tools over more data centres?

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.