$14 Billion From GPUs — Not Rockets

TD Cowen slapped a Buy on SpaceX this week.

The price target: $200.

SPCX trades near $149. That's 35% up from here.

But the bull case isn't about rockets. And it's not about Starlink. The whole call rests on renting GPUs to AI firms.

Let me explain…

TD Cowen projects AI compute leasing will bring in $14 billion for SpaceX this year. That's 35% of its total sales. By 2027, the firm sees that at $66 billion — 58% of all SpaceX revenue. By 2028… $133 billion.

In other words, two-thirds of SpaceX's income could come from AI, not space.

Sit with that for a moment.

SpaceX posted $7.8 billion in Q2 sales. That was 92% growth year over year. Adjusted EBITDA hit $3.5 billion.

Strong marks for a stock that went public just four months ago at $135 a share.

And TD Cowen says that's the warm-up. They project 62% compound annual growth in revenue through 2031.

Now, I know what you're thinking. A $200 target on a stock that's been public since June? That's bold.

It is. SpaceX lost $4.9 billion last year. Its AI and launch arms still burn cash. Only Starlink turns a profit — 38.6% margin in Q2.

But the thesis isn't about today's books.

It's about where the cash shows up next.

Musk moved up SpaceX's data center plan twice since the IPO. The first draft said 2028. Then it got sooner.

Now TD Cowen says ground-based GPU leasing is the top near-term driver.

Here's why that works.

SpaceX built a web of ground stations for Starlink. Those sites have power. They have cooling. They have fast fiber links.

Plug in GPU racks… and you've got a spread-out AI compute grid built on space gear.

Amazon sold books before it rented servers. AWS made Amazon a tech titan. SpaceX could follow that same arc — with rockets and GPUs.

That's the play.

So the question isn't whether SpaceX builds great rockets.

Monday proved that — Starship reached orbit for the first time.

The question is whether it can rent enough chips to fill a $2.6 trillion price tag.

We'll see…

Your 401(k) gets access to the fastest growing companies

See this document?

It brings sweeping changes to your 401(k) — and yet almost nobody knows about it.

Hidden inside it is what one 40-year trading veteran calls the "biggest opportunity he's seen in years."

It's all right there in Section 3.

That's where Washington opens your retirement account up to a group of companies that have grown as much as 10,000%... 325,000%... and even 600,000%.

These are the fastest growing companies in America, previously off-limits to everyday folks.

But no longer — and yet that's not even the best part…

Because hedge fund legend Larry Benedict has spotted something in this executive order that almost nobody else has...

This executive order is about to transfer as much as $1 trillion of retirement money from one corner of the market to another…

And Larry has found the one ticker sitting in the path of this movement.

Click here to get the ticker positioned to capture a slice of that $1 trillion — free.

Regards,

Lauren Wingfield

Managing Editor, The Opportunistic Trader

P.S. The last time Washington rewrote the retirement rules, Larry's readers had the chance at 188% gains. Now it's happening again — on a far bigger scale. Get the ticker before the money starts moving here.

This isn't a someday business. SpaceX already has a handful of customers leasing its AI compute, and the names are heavyweights — Alphabet's Google and Anthropic among them. In the June quarter, that segment pulled in $2.6 billion, up a startling 247% from a year earlier, putting it at roughly a $10 billion annual run rate already. TD Cowen thinks the business exits this year at a $41 billion run rate. Real customers, real revenue… growing fast. Watch whether more hyperscalers sign on — each one validates the thesis.

Follow the power, not the stock price. TD Cowen's model hinges on SpaceX scaling its compute capacity from about 2.1 gigawatts in 2026 to 6 by the end of 2027 — nearly triple — and all the way to 22 gigawatts by 2031. Gigawatts are the real unit here, because compute leasing is capped by how much power and cooling you can bring online. That ramp is enormous, and it's the single biggest assumption behind the $200 target. Watch the capacity numbers each quarter; they matter more than any single launch.

Hold the bull case up to the light. SpaceX lost billions last year, and only Starlink currently turns a profit — the launch and AI arms still burn cash. The entire valuation leans on a compute business that barely existed 18 months ago scaling into the tens of billions, fast, without a stumble. If the gigawatt ramp slips, or AI demand cools, the math changes quickly. This is a stock priced for a future that has to arrive on schedule. Watch the gap between the projections and the quarterly reality.

Watch SpaceX's gigawatt capacity, quarter by quarter.

Here's why that's the number that matters now. The whole $200 case rests on one idea: that SpaceX can build and fill data-center capacity faster than almost anyone. The unit of that race isn't satellites or launches — it's gigawatts of power online and leased.

The target is steep. From roughly 2.1 gigawatts this year to 6 by the end of 2027, on the way to 22 by 2031. Each step needs power deals, cooling, fiber, and GPUs — all delivered on time.

There's a reason to stay skeptical. Building gigawatts of data-center capacity is hard even for firms that do nothing else. Power is scarce, grid hookups are slow, and GPU supply is tight. A space company doing this on the side is making a bold claim.

But there's a reason it might work. Starlink's ground network already sits on sites with power and fiber — a head start on the hardest part of the problem.

So watch two things: whether the capacity numbers actually climb toward that 6-gigawatt mark, and whether new leasing customers keep signing. Those tell you if the AI story is real or just a model.

Let's keep this simple.

Forget rockets for a minute. Think about renting out apartments.

An AI company like Google needs somewhere to run its chips. Those chips are like tenants — they need space, power, and cooling to do their work. And right now, there aren't enough good buildings to house them all.

SpaceX noticed it already owns a lot of good buildings.

To run Starlink, it built sites all over with three things every data center needs: electricity, cooling, and fast internet lines. Those sites were built to talk to satellites. But the same bones can house racks of AI chips.

So SpaceX can act like a landlord. It rents out computing power by the hour to companies like Google and Anthropic. They bring the demand… SpaceX provides the room and the power.

Here's the part that surprises people. This landlord business could soon out-earn the rockets.

Launching is lumpy and expensive. Renting compute is steady, and demand is exploding as the whole world races to build AI. One analyst thinks it becomes most of SpaceX's revenue within a year or two.

But a landlord is only as good as the buildings it can finish. The whole promise rests on SpaceX adding power — measured in gigawatts — faster than almost anyone has before. That's the hard part, and it's far from certain.

Still, the shift in how to see the company is the point. The rockets get the attention. The quiet rows of humming chips might pay the bills.

Follow the power, not just the launches — that may be where this company's money really lives.