Software

AI + ML

DoJ reportedly advances Nvidia antitrust probe

Uncle Sam apparently worried GPU giant may be punishing customers who shop around


Updated The US Department of Justice on Tuesday is said to have stepped up its antitrust investigation into Nvidia, issuing subpoenas seeking evidence for its case against the AI chip giant.

Citing unnamed persons familiar with the matter, Bloomberg reported the legal order comes amid growing concern among watchdogs that Nvidia is engaging in monopolist behavior by making it harder to adopt rivals' products – and punishing customers that attempt to do so.

Nvidia has dismissed the claims, arguing that superior products are the reason for its success. "Nvidia wins on merit, as reflected in our benchmark results and value to customers, who can choose whatever solution is best for them," the GPU giant told Bloomberg.

Nv has previously admitted to prioritizing customers who possess the necessary datacenter capacity to quickly put its hardware to work. The Register has also chatted with Nvidia reps who said they try to prioritize accelerator shipments to operators of prominent AI services.

Nvidia’s accelerator and networking businesses have grown at astounding pace as orgs seek to deploy generative AI services. Training top-end neural networks can require as many as tens of thousands of GPUs, which can cost anywhere from $30,000 to $40,000 apiece. Running inferencing workloads to put models to work also needs accelerators. The amount of hardware you need varies by the size and user demand of the neural nets.

It's estimated that the flurry of investment in GPUs for model training and deployment has helped Nvidia secure more than 80 percent of the AI infrastructure market, despite ongoing efforts by Intel and AMD, at least, to compete with products such as Gaudi3 and MI300X, respectively.

The widespread adoption of these kinds of accelerators – Meta alone is said to be deploying more than 600,000 of the chips – has been reflected in Nvidia's market capitalization, which crested $3 trillion in July. Since then it's slid backwards, resting at a mere $2.65 trillion at the time of writing.

Tuesday's word of these subpoenas came on a day when Nvidia scrip was already slipping further. Shares ended down 9.53 percent at market close. In fact, the market as a whole was hurting on Tuesday for various reasons.

It's not just US authorities that are worried by Nvidia. French regulators are also investigating the Jensen Huang-run mega-corp as part of a broader investigation into the cloud computing sector. The European Union is also said to be digging into the GPU vendor's business practices to determine whether corrective measures are warranted. ®

Updated to add on September 4

In a statement today, Nvidia says that while it has made contact with the US government, it has not been subpoenaed, contrary to what Bloomberg reported.

"We have inquired with the US Department of Justice and have not been subpoenaed," the GPU giant told El Reg. "Nonetheless, we are happy to answer any questions regulators may have about our business."

Send us news
5 Comments

Nvidia and chums inject $160M into Applied Digital to keep GPU sales rolling

Datacenters are the lifeline for its $30B ML-fueled boom

AI-pushing Adobe says AI-shy office workers will love AI if it saves them time

knowledge workers, overwhelmed by knowledge tasks? We know what you need

Nvidia admits Blackwell defect, but Jensen Huang pledges Q4 shipments as promised

The setback won't stop us from banking billions, CFO insists

Buying a PC for local AI? These are the specs that actually matter

If you guessed TOPS and FLOPS, that's only half right

Nvidia's growth slows to a mere 122 percent but it’s still topping expectations

Still growing in China, ramping Hopper prods and predicting Blackwell billions soon

Canadian artist wants Anthropic AI lawsuit corrected

Tim Boucher objects to the mischaracterization of his work in authors' copyright claim

Tenstorrent's Blackhole chips boast 768 RISC-V cores and almost as many FLOPS

Shove 32 of 'em in a box and you've got nearly 24 petaFLOPS of FP8 perf

Benchmarks show even an old Nvidia RTX 3090 is enough to serve LLMs to thousands

For 100 concurrent users, the card delivered 12.88 tokens per second—just slightly faster than average human reading speed

Nvidia's latest AI climate model takes aim at severe weather

That tornado warning couldn't possibly be a hallucination... could it?

LiquidStack says its new CDU can chill more than 1MW of AI compute

So what’s that good for? Like eight of Nvidia’s NVL-72s?

AI firms propose 'personhood credentials' … to fight AI

It's going to take more than CAPTCHA to prove you're real

Copper's reach is shrinking so Broadcom is strapping optics directly to GPUs

What good is going fast if you can't get past the next rack?