Nvidia Just Reportedly Agreed to Pay Almost $13 Billion for the Home of Open Source AI

Nvidia may be about to make the biggest purchase in its history, and the target is not a chip company. It is the website where most of the world’s open AI models actually live. Anthropic just locked in 45 billion dollars of future computing power to keep training its own models. Waymo spent the week publicly arguing that AI by itself cannot drive a car safely, an argument aimed squarely at rivals betting the opposite. And in chess, a 21 year old from Chennai did something no Indian player has ever done before.

Nvidia wants to own the home of open source AI

Nvidia has reportedly agreed to buy Hugging Face, the site where developers upload, share and download open source AI models, for close to 12.9 billion dollars. The Information broke the story, and it has since been echoed by CNBC, Forbes, Bloomberg and TechCrunch, all citing people familiar with the talks.

One thing is worth being precise about. Neither Nvidia nor Hugging Face has confirmed this publicly, and multiple outlets note that no signed contract exists yet, so the deal could still change or collapse before either side announces anything. Take the number as reported, not official.

If it does go through, this would be Nvidia’s largest acquisition ever, more than double what it paid for the networking company Mellanox in 2020, and bigger than the 40 billion dollar bid for the chip designer Arm that regulators eventually killed. Hugging Face turned down a much smaller offer from Nvidia last year, an investment that would have valued the company at only 7 billion dollars. Founded in 2016, Hugging Face is not a chipmaker or a model builder itself. It is closer to a library and a social network combined, the place most AI researchers go first to find or publish a model. Owning it would hand Nvidia a direct line into the software layer that sits on top of the hardware it already dominates, which is exactly the part of the AI stack the company does not yet control.

Anthropic just bought six years of power, in advance

Anthropic signed a deal worth roughly 45 billion dollars with Nscale, a British AI infrastructure company, to rent around 460 megawatts of computing capacity from a data center Nscale is building in West Virginia. The facility is expected to come online at the end of 2027 and will run on Nvidia’s next generation Vera Rubin chips, a system that combines six different chip types working together.

The deal runs six years. It is also not Anthropic’s first this month. Earlier in August, the company signed a separate 10 billion dollar deal with the cloud startup Volta for capacity out of a data center in Norway. Nscale itself had previously worked with Microsoft before that relationship ended, and this agreement is reportedly meant to help anchor Nscale’s own plans to go public.

What is notable here is the shape of the bet. Anthropic is not paying for compute it needs today. It is locking in power and chips years ahead of when the facility will even exist, on the assumption that whoever has guaranteed access to electricity and silicon in 2027 will be the one still standing.

Waymo says there is no shortcut, and it means Tesla

After more than 15 years and over 200 million fully driverless miles, Waymo published an argument this week that amounts to a direct challenge to a chunk of the self-driving industry. Srikanth Thirumalai, the company’s vice president of onboard software, told Axios that “pure end to end systems are not able to meet our safety bar at the scale that we operate.”

End to end systems are the approach where a model takes in raw camera footage and outputs steering and braking decisions directly, with no separate rulebook in between. It is the approach several rivals, most prominently Tesla, have leaned on. Waymo’s argument is that a pure end to end system is a black box: when it fails, there is no clean way to know why, which makes it hard to trust at the scale of a real robotaxi fleet. Waymo instead runs a second, independent AI layer on top of its main driving model, one that checks every proposed move against hard physics and traffic law before the car is allowed to act on it.

The company timed the argument well. The same week, Waymo announced it will bring driverless rides to Munich by the end of 2027, its first market in the European Union, after Germany’s Federal Motor Transport Authority and local officials. Vehicles will first map the city manually, then drive with a trained human behind the wheel before removing them entirely, the same staged process Waymo has used in every city it has entered. As of the announcement, Waymo held no permits yet in Germany’s regulatory database, so the timeline still depends on approvals it has not received.

The prodigy label finally became a title

At the Grand Chess Tour Finals, R Praggnanandhaa came from six points behind on the last day to beat Fabiano Caruana, one of the strongest players in the world, by a final score of 15 to 13. It made him the first Indian ever to win the Grand Chess Tour, one of the sport’s most prestigious annual titles.

Praggnanandhaa, from Chennai, trailed Caruana heading into the final day of matches. He won both rapid games to erase the gap and take the lead, then held an even score through the blitz games that followed, enough to finish two points ahead. He earned 200,000 dollars for first place, and both finalists secured automatic spots in next year’s tour. In the tournament’s third place match, India’s Wesley So swept all four blitz games to beat Vincent Keymer 18 to 12.

Praggnanandhaa became a grandmaster at 12 years and 10 months old, one of the youngest in history, and has spent years being described as a future contender rather than a champion. This week that description stopped being about the future.

Four stories, one thread. Nvidia and Anthropic are both spending unprecedented sums to lock down pieces of the AI stack years before they need them, betting that whoever controls the model hub and the power supply controls the decade. Waymo is betting the opposite kind of caution, that the fastest way to lose the self-driving race is to cut the safety corner AI makes it tempting to cut. And in Chennai, a 21 year old spent a decade being called a prodigy before finally winning the tournament that proves it wasn’t just talk.