Jensen Huang: Nvidia's Insane Deal With SpaceX Changes Everything!
Jensen Huang just revealed Nvidia's deal with SpaceX — and it is not what anyone expected. "we're going to space. We've already been out in space. Thor is radiation approved and we're in satellites."
Jensen Huang just told a packed room that Nvidia is already in orbit, and that was the easy part. "in the future, we'll also build data centers in space, obviously very complicated to do. So we have we're working with our partners on a new computer called Vera Rubin Space one."
Jensen Huang is ready. But is Elon Musk actually ready too, or is there a catch hiding underneath all of this? By the end of this video, you'll know the real answer, and why it could change how you think about Nvidia's next decade of growth.
AI Generated·May contain errors
Video Summary
AI Generated
NVIDIA and SpaceX have entered a temporary alliance to bring AI computing into orbit. While SpaceX is currently utilizing NVIDIA's high-performance silicon for its AI-1 satellites, Elon Musk is simultaneously developing custom chips via TeraFab to eliminate this dependency and optimize for the unique thermal and power constraints of space.
Jensen Huang just revealed NVIDIA's deal with SpaceX, and it is not what anyone expected.
We're going to space. We've already been out in space. Thor is radiation approved,
and we're in satellite.
Jensen Huang just told a packed room that NVIDIA is already in orbit,
and that was the easy part.
In the future, we'll also build data centers in space. Obviously, very complicated to do so.
We're working with our partners on a new computer called Vera Rubin Space 1.
Jensen Huang is ready, but is Elon Musk actually ready too,
or is there a catch hiding underneath all of this?
At NVIDIA's GTC conference back in March, Jensen Huang
announced a new chip platform called the Space 1 Vera Rubin module.
Compared to the H100, the chip that started this entire artificial intelligence boom,
this new module delivers up to 25 times more compute for satellites doing work in orbit.
Companies like Starcloud, Axiom Space, and Planet Labs signed on immediately.
None of them are household names yet,
but there's one name people expected to see in that lineup, and it wasn't there.
SpaceX.
That's strange, because SpaceX has used NVIDIA hardware on the ground for a decade.
So when Elon Musk stood up just three months later and unveiled his own orbital computer,
a satellite called AI-1, a lot of people assumed he was simply joining NVIDIA's club.
The truth is messier, and honestly, more interesting.
Now, of course, in space, there's no conduction, there's no convection,
there's just radiation. And so we have to figure out how to cool these systems out in space.
Notice what Jensen just admitted there.
NVIDIA, the most dominant chip company on the planet, doesn't have this fully solved either.
Cooling a chip on Earth is simple. You blow air across it, or you run water through it.
In space, there's no air to blow, and no atmosphere to carry heat away.
The only way to get rid of heat is to radiate it directly into the vacuum,
the same way a campfire radiates warmth even though there's no wind.
That's a completely different engineering problem,
and it's the reason every single company chasing orbital computing,
including NVIDIA and including SpaceX, is racing to solve the same physics puzzle at the same time.
Now here's where the SpaceX story gets specific.
Musk's AI-1 satellite carries roughly the compute power of one full NVIDIA
GB300 server rack, the same rack you'd find inside a massive data center on Earth.
And according to SpaceX's own chief financial officer,
the first generation of these satellites will actually run on NVIDIA chips.
So in that sense, yes, there really is a deal.
NVIDIA Silicon is going to space on a SpaceX satellite.
But that's only half the story, and it's the half that makes this whole thing fascinating
instead of just another partnership announcement.
We have now set up a supply chain that can manufacture thousands a week of these systems,
essentially multi gigawatts of AI factories per month inside our supply chain.
And that clip is Jensen talking about NVIDIA's manufacturing scale on Earth,
and it matters here because it shows you exactly what SpaceX is trying to avoid depending on.
Musk doesn't just want to buy NVIDIA's chips, he wants to eventually build his own.
SpaceX merged with Musk's artificial intelligence company, XAI, earlier this year.
And right alongside that merger, SpaceX announced a joint chip factory with Tesla called TeraFab.
The whole purpose of that factory is to build a new processor called the D3,
designed specifically to survive radiation in orbit and run hot without needing heavy
cooling systems. If that chip succeeds, SpaceX would no longer need NVIDIA hardware in its
satellites at all. So the deal isn't a long term marriage, it's closer to a bridge loan.
SpaceX is using NVIDIA's chips right now because NVIDIA is the only company that actually has
working AI silicon ready today. But Musk built the AI1 satellite with what's called an interchangeable
payload, meaning the compute hardware can be swapped out for a different chip later
without redesigning the whole satellite. That single design choice tells you everything about
Musk's real intention. He's not committing to NVIDIA, he's keeping the relationship temporary
on purpose. What you see here on the left, on this side, is tokens per watt. Tokens per watt
is important because every data center, every single factory by definition is power constrained.
This is exactly why this rivalry matters financially, not just technically. Every chip
NVIDIA sells gets judged on one number, how much useful work it can squeeze out of a fixed amount
of electricity. On Earth, that electricity is limited by the power grid. In space, it's limited
by how much sunlight a satellite's solar panels can capture, and how much heat its radiators can
dump overboard. Whoever builds the most efficient chip for that environment effectively writes the
rulebook for an entirely new category of data center. And right now, SpaceX is betting that
its own custom chip can eventually out-efficient anything NVIDIA builds for the specific conditions
of orbit, even though NVIDIA currently has no real competition on raw performance. So let's go
back to the question from the start of this video. Is Elon Musk actually ready to run his AI satellites
on NVIDIA chips long-term, or is there a catch hiding underneath all of this? Here's the catch,
laid out plainly. SpaceX's own filings ahead of its recent stock market listing admitted the company
can't currently secure enough chips to build this satellite network at the scale it wants. That's
not a small detail. That's SpaceX, one of the most resourced companies in the world, publicly stating
it is dependent on outside chip makers for its biggest new venture. NVIDIA is almost certainly
one of those outside chip makers, at least for now. Meanwhile, Musk is pouring billions into
TeraFab specifically to escape that dependency as fast as possible. There was a monkey king,
token king. Well, we take all of our software, as I told you, we vertically integrate,
but we horizontally open. This is the part most people miss when they hear that NVIDIA is
untouchable right now. NVIDIA's advantage isn't just the chip. It's the chip plus the software,
plus the cooling systems, plus the networking, all designed together as one package. That's
incredibly hard to copy on Earth, and it's even harder to copy in space, where every gram of weight
and every watt of power matters 10 times more. Musk knows this. It's exactly why his own AI1 design
still relies on NVIDIA hardware for now, even while his engineers race to build an alternative.
For NVIDIA, this means something simple but important for investors. Even a company as
ambitious and well-funded as SpaceX cannot walk away from NVIDIA chips today. Every satellite
SpaceX launches using NVIDIA hardware in the near term is essentially additional revenue for a
company that already dominates ground-based data centers, just in a brand new market that barely
existed two years ago. And if TeraFab ends up delayed, which is common with first-generation
chip factories, that dependency could last years longer than Musk is currently projecting.
So here's the real picture. NVIDIA isn't locked into an exclusive partnership with SpaceX,
and SpaceX isn't loyal to NVIDIA either. What's actually happening is a temporary alliance between
two companies who need each other today, while one of them is quietly building the tools to no
longer need the other tomorrow. That's not a typical Silicon Valley friendship. That's a
countdown clock, and right now NVIDIA is still the only company that can supply chips fast enough to
keep that clock from running out early. Here's what actually happened. At NVIDIA's GTC conference
back in March, Jensen Huang announced a new chip platform called the SpaceOne Vera Rubin module.
Compared to the H100, the chip that started this entire artificial intelligence boom,
this new module delivers up to 25 times more compute for satellites doing work in orbit.
Companies like StarCloud, Axiom Space, and Planet Labs signed on immediately. None of them are
household names yet, but there's one name people expected to see in that lineup, and it wasn't
there. SpaceX. That's strange, because SpaceX has used