Elon Musk announced that SpaceX will launch its first artificial intelligence satellites, equipped with Nvidia technology, in the fourth quarter of 2027. Yes, you read that right. Data centers in Earth’s orbit.
And around the same time, Nvidia is putting $3.5 billion into a Taiwanese company that most people know for the chips in our phones.
Two completely different moves. But with the same logic behind them.
Musk said the first AI satellites will launch at the end of 2027. And that the network of orbital data centers will reach “significant scale” during 2028.
“And why would data centers go into space?” you might be wondering. Musk’s argument is simple. Running AI workloads in orbit could eventually cost less than maintaining massive data centers down here, which consume as much energy as small cities. And it could also have a smaller environmental footprint.
And this is exactly where Nvidia comes into play. SpaceX’s data centers will operate exclusively with Nvidia accelerators. In fact, during the August 26 earnings call, CFO Colette Kress said that the Vera Rubin system is in full production and that the first shipments are going to Oracle, Amazon, and SpaceX.
And things get even more interesting. Because the relationship works both ways. Nvidia revealed that it owns 122.8 million shares of SpaceX. In other words, it is both a supplier and a major shareholder.
And as if that weren’t enough, in August SpaceX completed its acquisition of Cursor for $60 billion, entirely in stock. The AI coding assistant that Jensen Huang himself has said is used throughout his company.
You might say now… if it’s such a good idea, why haven’t they done it already? And here comes the big “but.”
According to a recent report from the Brookings Institution, the main problem isn’t chip technology. It’s temperature.
Data centers here on Earth are cooled using air and liquid. They remove heat from the chips and ultimately release it into the atmosphere around the building. In space, however, there is no atmosphere. There is nowhere for that heat to go.
The only way is through thermal radiation. And some scientists calculated how many radiators a single orbital data center would need. Want to know how many? 2.15 million square feet. For just one data center.
And that’s not all. There is the constant bombardment from the Sun’s ultraviolet radiation, which damages equipment. There is an increased risk of collisions as orbit becomes more crowded. And there are also regulatory approvals to deal with.
That’s why Evercore is being very cautious. It does not forecast any revenue from orbital computing in 2026 or 2027. It sees the first orbital gigawatt in FY29, with 8 GW in orbit versus 10 GW on the ground.
And it raises the right question. What does Musk mean by “significant scale” in 2028? Revenue? Or simply a phase of technological validation before commercialization? Because if they bring FY29 forward to 2028, with the economics of $30 to $50 per watt that Musk mentioned, we’re talking about serious upside.
And this brings us to the second move.
Nvidia is purchasing $3.5 billion worth of convertible bonds from MediaTek. The largest direct investment it has ever made outside the United States.
“And why would it put that much money into a company like MediaTek?” you might be wondering. This is where the strategy becomes clear.
Its major customers—Amazon, Google, and Microsoft—are all developing their own chips to reduce their dependence. So what does Nvidia do? Instead of fighting them, it gives them the plumbing.
In simple terms: MediaTek is adopting the NVLink Fusion and new NVHBM platforms. So any hyperscaler that wants to build its own custom XPU can do so using a ready-made, pre-validated path that plugs directly into Nvidia’s rack-scale AI factories. In other words, even when the customer uses its own chip, the network and interconnects remain Nvidia.
It’s exactly the same pattern as the agreement with Amazon, which committed to purchasing 2 million additional components and, most importantly, using Nvidia’s interconnect technology with its own chips.
“We already had a major partnership with MediaTek. Today we’re making it much bigger,” Huang said on Bloomberg TV, referring to a decade-long roadmap. And he added something that captures the spirit of the deal: “Nvidia’s networking ecosystem is now part of MediaTek’s supply chain, and MediaTek’s XPU is part of ours.”
The partnership also extends to platforms for AI in computers and automobiles. As MediaTek CEO Rick Tsai put it, we’re talking about cloud AI infrastructure, local AI computing, and automotive applications “in the era of physical AI.”
And to understand what this means for MediaTek: its valuation has nearly tripled in recent months. It is targeting 15% of an $80 billion market next year, up from roughly $2 billion in AI chip sales this year.
And yet, the debate over how “circular” all these Nvidia investments in customers and partners really are continues. In other words, is it creating artificial demand for itself? Huang rejects that idea: “It’s not circular, because obviously they’re doing their job and we’re doing ours.”
