Nvidia

Bloomberg explores Amazon’s $8 billion partnership with Anthropic that could advance Amazon’s Trainium hardware and software tools enough for the AWS provider to cut into Nvidia’s stranglehold on the $100-billion-plus market for AI chips:

Trainium2 is the company’s third generation of artificial intelligence chip. By industry reckoning, this is a make-or-break moment. Either the third attempt sells in sufficient volume to make the investment worthwhile, or it flops and the company finds a new path.

$14.8 billion profit in Q1, $16.6 billion in Q2, and now $19.3 billion in Q3 of fiscal 2025 — that’s profit, not earnings. (Earnings were $35.08 billion, up from $30.04 billion last quarter.)

The vast majority is from AI data center, of course — but gaming did have a 14 percent bump. It’s a $3B-a-quarter business, while data center is a $30B one.

From Nvidia’s FAQ:

No, Founders memberships will still have unlimited playtime hours for life, as long as there is no lapse in their membership. 

Also:

Founders can upgrade to Ultimate and return to their prior Founders benefits at any time without penalty, as long as there is no lapse in their payments. 

“It was functional, but the design flaw caused the yield to be low. It was 100% Nvidia’s fault,” Nvidia’s Jensen Huang tells Reuters, effectively confirming The Information’s report from August about why its new flagship AI chips won’t ship in large amounts right away.

He says it’s now fixed, but the timeline stays the same: Q4 for first shipments.

Nvidia has released a new Game Ready driver (566.03) that includes support for Alan Wake II: The Lake House, Call of Duty: Black Ops 6, Red Dead Redemption, and more. The 566.03 release also includes support for 32 new G-Sync compatible monitors. If you use Bluestacks or Corsair iCUE equipment, hold off on this driver as there’s a known high CPU usage issue.

Nvidia’s beta all-in-one app is set to replace GeForce experience later this year, and it might fully launch with the RTX 50-series. Hardware leaker kopite7kimi says it will be released officially together with Nvidia’s next-gen GPUs. Nvidia is holding a CES 2025 keynote in January, where we’re expecting to see the RTX 5090 and RTX 5080.

Nvidia just said it won't begin to ramp production of its new Blackwell GPUs until its fourth fiscal quarter, which begins in November.

Nvidia just reported a record $30 billion quarter, and the vast majority of it ($22B) was data center compute... will it set aside some of its latest and greatest for gamers instead of AI? If so, I wouldn't bet on prices being low.

Nvidia is announcing an updated RTX 4070 with GDDR6 memory today. “To improve supply and availability to meet strong demand, we’re introducing the GeForce RTX 4070 with extra fast GDDR6 memory,” says Nvidia. All of the other specs of the GPU are the same, and Nvidia says “it offers similar performance” in games and apps, despite the decrease in bandwidth from GDDR6X to GDDR6 memory. This updated RTX 4070 will be available worldwide starting in September.

AMD is acquiring ZT Systems, a leading provider of AI infrastructure. AMD is calling it a “next major step” for its AI training and inferencing solutions, in a move that will clearly help it compete with Nvidia’s dominance in AI offerings. ZT Systems will join the AMD’s data center solutions group once the $4.9 billion transaction closes.

Nvidia has partnered up with modding repository CurseForge to support 25 of the most popular WoW Addons — including options for UI customization, combat, action bars, and quest helpers — for Ultimate and Priority subcribers of the cloud streaming service.

The mods are pre-installed ready for launch, always updated to the latest release, and don’t require users to have a CurseForge account.

404 Media reports, with screenshots of Slack conversations and excerpts from emails, on a massive undertaking by Nvidia to scrape online videos for AI training that appears to go well beyond research.

According to the messages, they were attempting to download full-length videos from a variety of sources including Netflix, but were focused on YouTube videos. Emails viewed by 404 Media show project managers discussing using 20 to 30 virtual machines in Amazon Web Services to download 80 years-worth of videos per day. 

Elliott Management, famous for targeting underperforming companies such as Twitter, says Nvidia is in a bubble, in a new letter to investors.

Many of AI’s supposed uses are “never going to be cost-efficient, are never going to actually work right, will take up too much energy, or will prove to be untrustworthy”, it said.

The sell-off at least in part is about Wall Street losing confidence in AI. (I did warn you it was going to be a year of reckoning back in February!) There are a couple of other things going on, with potentially long-term effects on tech, too.

Crypto, a proxy for investors’ appetite for brainless risk, started a plunge that continued to worsen into Monday morning. The cause of all this came from three surprising pieces of economic data that came out last week, causing traders to rethink how they make, or at least don’t lose, money.

Meta’s CEO got a little heated while talking with Nvidia CEO Jensen Huang at the SIGGRAPH conference today in Denver.

The topic turned to Meta’s approach to AI with Llama. Zuckerberg made clear that investing so much in foundational models is strongly influenced by not wanting to relive his history with Apple and the App Store:

“One of my things for the next 10 or 15 years is I just want to make sure we can build the fundamental technology that we’re going to be building social experiences on. Because there have just been too many things that I’ve tried to build and then have just been told, ‘Nah, you can’t really build that,’ by the platform provider that, at some level, I’m just like, ‘Nah, fuck that.’”

The text-to-image generator, which is trained on Getty’s stock pictures, now uses an upgraded version of Nvidia’s Edify AI model, making it faster and more accurate. It also comes with new features to control the “camera settings” used in an AI-generated image, such as depth of field or focal length.

At 4:30PM ET, Nvidia’s CEO will sit down with Wired’s Lauren Goode to talk generative AI. Then at 6PM ET, Huang is scheduled to chat with Mark Zuckerberg — also about AI. You can watch both conversations on YouTube.

Reuters reports that Nvidia is working on a version of its new “Blackwell” chips for the Chinese market, which would be in line with strict US export controls for AI training chips.

Nvidia will reportedly work with Chinese distributor partner Inspur on launching the so-called “B20” chip, which is pitched to compete against domestic offerings from Huawei and Tencent-backed startup Enflame.

It’s a “global outage,” according to an Nvidia status message. “We are working on a fix to bring back the service as soon as possible.”

Nvidia’s other GeForce Now services appear to be operational, so perhaps this issue is tied to everything else going on.

According to The Information, OpenAI is in discussion with Broadcom and other semiconductor designers about developing its own artificial intelligence chip to address shortages in its supply chain and reduce dependency on Nvidia. OpenAI has apparently also hired former Google chip staffers.

Bloomberg previously reported in January that OpenAI CEO Sam Altman was planning to raise billions of dollars to set up a network of chip factories.