Skip to content

Topic

AI accelerator market

The market for GPUs and custom silicon used to train and serve AI models, including in-house chips designed by cloud and model companies.

Current stories

investConfirmed23 publishers

Nvidia pays 1.84 times its rejected offer to own the hub 13 million developers pull from

Nvidia was refused a minority stake in the open-model hub late last year. The reported price for the whole company works out to about twelve days of Nvidia revenue, which is what defending the long tail of GPU demand costs.

Perspective Coverage

23 publishers
Builder
Builder 17%
Operator
Operator 22%
Investor
Investor 61%

Reality

Evidence55
Adoption70
Hype gap+35
Incentives60
Confidence58
investOne report1 publisher

Nvidia is 13% of SK hynix sales. So is somebody else.

SK hynix's semiannual filing shows a second unnamed customer buying within 421 billion won of Nvidia. The HBM buyer base is widening, not consolidating.

Publishers:en.sedaily.com

Reality

Evidence62
Adoption68
Hype gap+20
Incentives55
Confidence55
investOne report1 publisher

Goldman's 54-million-chip 2030 forecast requires China to double domestic output every year

China's four little GPU tigers now have public prices and a 2030 demand forecast big enough to justify them. The lock-up calendar has been repricing them faster than their losses have narrowed. Enflame is the newest of them.

Publishers:morningstar.com

Reality

Evidence46
Adoption24
Hype gap+29
Incentives66
Confidence43
productOne report1 publisher

Nvidia now carries $99bn of its own customers on its balance sheet

The book grew from about $7bn in a year, and roughly half of it is priced by funding rounds rather than by markets, which means a demand forecast built on customer orders is partly a forecast of Nvidia's own marks.

Publishers:thenextweb.com

Reality

Evidence70
Adoption76
Hype gap+12
Incentives72
Confidence64
leadershipConfirmed4 publishers

Nvidia is buying the model hub that Intel, AMD and Amazon helped fund

Nvidia agreed to pay $12.93bn for Hugging Face and promised its own compute will never be required there. That promise rests on a blog post. The default model source for 18 million developers now sits inside a chip vendor.

Publishers:blogs.nvidia.com businessinsider.com pcgamer.com theguardian.com

Perspective Coverage

4 publishers
Builder
Builder 41%
Operator
Operator 21%
Investor
Investor 38%

Reality

Evidence66
Adoption72
Hype gap+30
Incentives88
Confidence74