Nvidia on Wednesday refuted a report that the Chinese artificial intelligence startup DeepSeek has been using smuggled Blackwell chips to develop its upcoming model.

The U.S. has banned the export of Nvidia’s Blackwell chips, which are considered the company’s most advanced offerings, to China in an effort to stay ahead in the AI race.

DeepSeek is reportedly using chips that were snuck into the country without authorization, according to The Information.

“We haven’t seen any substantiation or received tips of ‘phantom datacenters’ constructed to deceive us and our OEM partners, then deconstructed, smuggled, and reconstructed somewhere else,” a Nvidia spokesperson said in a statement. “While such smuggling seems farfetched, we pursue any tip we receive.”

Nvidia has been one of the biggest winners of the AI boom so far because it develops the graphics processing units (GPUs) that are key for training models and running large workloads.

Since the hardware is so crucial for advancing AI technology, Nvidia’s relationship with China has become a political flashpoint among U.S. lawmakers.

President Donald Trump on Monday said Nvidia can ship its H200 chips to “approved customers” in China and elsewhere on the condition that the U.S. will get 25% of those sales.

The announcement was met with pushback from some Republicans.

DeepSeek spooked the U.S. tech sector in January when it released a reasoning model, called R1, that rocketed to the top of app stores and industry leaderboards. R1 was also created at a fraction of the cost of other models in the U.S., according to some analyst estimates.

In August, DeepSeek hinted that China will soon have its own “next generation” chips to support its AI models.

WATCH: Nvidia selling H200 AI chips to China is net positive, says Patrick Moorhead

– CNBC’s Kristina Partsinevelos contributed to this report.

Nvidia-backed startup Starcloud trained an artificial intelligence model from space for the first time, signaling a new era for orbital data centers that could alleviate Earth’s escalating digital infrastructure crisis.

Last month, the Washington-based company launched a satellite with an Nvidia H100 graphics processing unit, sending a chip into outer space that’s 100 times more powerful than any GPU compute that has been in space before. Now, the company’s Starcloud-1 satellite is running and querying responses from Gemma, an open large language model from Google, in orbit, marking the first time in history that an LLM has been has run on a high-powered Nvidia GPU in outer space, CNBC has learned.

“Greetings, Earthlings! Or, as I prefer to think of you — a fascinating collection of blue and green,” reads a message from the recently launched satellite. “Let’s see what wonders this view of your world holds. I’m Gemma, and I’m here to observe, analyze, and perhaps, occasionally offer a slightly unsettlingly insightful commentary. Let’s begin!” the model wrote.

Starcloud wants to show outer space can be a hospitable environment for data centers, particularly as Earth-based facilities strain power grids, consume billions of gallons of water annually and produce hefty greenhouse gas emissions. The electricity consumption of data centers is projected to more than double by 2030, according to data from the International Energy Agency.

Starcloud CEO Philip Johnston told CNBC that the company’s orbital data centers will have 10 times lower energy costs than terrestrial data centers.

“Anything you can do in a terrestrial data center, I’m expecting to be able to be done in space. And the reason we would do it is purely because of the constraints we’re facing on energy terrestrially,” Johnston said in an interview.

Johnston, who co-founded the startup in 2024, said Starcloud-1’s operation of Gemma is proof that space-based data centers can exist and operate a variety of AI models in the future, particularly those that require large compute clusters.

“This very powerful, very parameter dense model is living on our satellite,” Johnston said. “We can query, it and it will respond in the same way that when you query a chat from a database on Earth, it will give you a very sophisticated response. We can do that with our satellite.”

In a statement to CNBC, Google DeepMind product director Tris Warkentin said that “seeing Gemma run in the harsh environment of space is a testament to the flexibility and robustness of open models.”

In addition to Gemma, Starcloud was able to train NanoGPT, an LLM created by OpenAI founding member Andrej Karpathy, on the H100 chip using the complete works of Shakespeare. This led the model to speak in Shakespearean English.

Starcloud — a member of the Nvidia Inception program and graduate from Y Combinator and the Google for Startups Cloud AI Accelerator — plans to build a 5-gigawatt orbital data center with solar and cooling panels that measure roughly 4 kilometers in both width and height. A compute cluster of that gigawatt size would produce more power than the largest power plant in the U.S. and would be substantially smaller and cheaper than a terrestrial solar farm of the same capacity, according to Starcloud’s white paper.

These data centers in space would capture constant solar energy to power next-generation AI models, unhindered by the Earth’s day and night cycles and weather changes. Starcloud’s satellites should have a five-year lifespan given the expected lifetime of the Nvidia chips on its architecture, Johnston said.

Orbital data centers would have real-world commercial and military use cases. Already, Starcloud’s systems can enable real-time intelligence and, for example, spot the thermal signature of a wildfire the moment it ignites and immediately alert first responders, Johnston said.

“We’ve linked in the telemetry of the satellite, so we linked in the vital signs that it’s drawing from the sensors — things like altitude, orientation, location, speed,” Johnston said. “You can ask it, ‘Where are you now?’ and it will say ‘I’m above Africa and in 20 minutes, I’ll be above the Middle East.’ And you could also say, ‘What does it feel like to be a satellite? And it will say, ‘It’s kind of a bit weird’ … It’ll give you an interesting answer that you could only have with a very high-powered model.”

Starcloud is working on customer workloads by running inference on satellite imagery from observation company Capella Space, which could help spot lifeboats from capsized vessels at sea and forest fires in a certain location. The company will include several Nvidia H100 chips and integrate Nvidia’s Blackwell platform onto its next satellite launch in October 2026 to offer greater AI performance. The satellite launching next year will feature a module running a cloud platform from cloud infrastructure startup Crusoe, allowing customers to deploy and operate AI workloads from space.

“Running advanced AI from space solves the critical bottlenecks facing data centers on Earth,” Johnston told CNBC.

“Orbital compute offers a way forward that respects both technological ambition and environmental responsibility. When Starcloud-1 looked down, it saw a world of blue and green. Our responsibility is to keep it that way,” he added.

Along with Starcloud and Nvidia’s efforts, several companies have announced space-based data center missions. On Nov. 4, Google unveiled a “moonshot” initiative titled Project Suncatcher, which aims to put solar-powered satellites into space with Google’s tensor processing units. Privately-owned Lonestar Data Holdings is working to put the first-ever commercial lunar data center on the moon’s surface.

OpenAI CEO Sam Altman has explored an acquisition or partnership with a rocket maker, suggesting a desire to compete against Elon Musk‘s SpaceX, according to The Wall Street Journal. SpaceX is a key launch partner for Starcloud.

Referring to Starcloud’s launch in early November, Nvidia senior director of AI infrastructure Dion Harris said: “From one small data center, we’ve taken a giant leap toward a future where orbital computing harnesses the infinite power of the sun.”