Incredible things are happening in AI. What’s more incredible is that it’s still EXTREMELY early. As in, Day 0.000. Things are only going to get crazier from here.
In case you’ve been living under a rock, Claude Opus 5.5 dropped recently. It’s the current “frontier of intelligence”. Truly mind-blowing. And again: it’s still early. Just wait for what comes next.
The benchmarks are impressive but YAWN. Boring numbers. What matters is what the model can \do\.
Day 0.0 and near-infinite demand
Over a single weekend you can vibe-code a Zelda: Breath of the Wild–style clone with your kids, give a model one prompt plus a mood board and ask for an AI-themed music video; wake up and it’s done. Work that used to take dozens or hundreds of people weeks or months now takes you and an agent a matter of minutes or hours.
And this is day 0.0 of where things are headed. Models keep getting better, faster, cheaper and more capable. AI is improving on every axis ALL AT ONCE. If you want frontier (aka GIGA BRAIN AI), you will pay through the nose but today’s 420 IQ frontier AI will be tomorrow’s budget AI; just as smart, but for a fraction of the cost. Such is the pace of AI progress. The direction is unambiguous. Better. Faster. Cheaper. Indefinitely.
The underlying point: demand for AI will be near infinite for decades. Use of AI will be near infinite for decades. Near-infinite demand and near-infinite use means near-infinite need for AI compute.
That is where SpaceXAI (henceforth SpaceXSI - “SI” for “Super Intelligence” is a necessary and accurate rebrand) enters the picture.
The layer everyone is ignoring
The SpaceXSI team, led by Elon Musk, is competing in the model layer — coding, image generation, language, the lot. They’ve also shipped my favorite product ever, Grok Bot. But what is happening \beneath\ the surface is being completely ignored by most investors (all of whom will regret this).
I once heard a guy with three first names say: he who controls the compute controls the universe.
While every hot take focuses on the visible model layer (the output of having compute for training and inference), SpaceXSI is securing the substrate: COMPUTE.
Regardless of who “wins” in the AI model layer, SpaceXSI has positioned themselves to be the biggest beneficiary of AI success and increasing adoption by owning a huge amount of the underlying AI compute needed by everyone to train models and do inference. First on earth, then OFF it.
It’s worth taking another moment to appreciate the INSANE pace at which AI models are improving, despite it still being day 0.0.
Frontier AI has already exceeded what is needed for 99.999% of use cases.
Once you far exceed the intelligence needed for a class of tasks, more intelligence is pointless. As Musk recently said, you don’t need Newton-level intelligence in your toaster; it would be cruel.
Opus 5.5 is blowing minds at the frontier today. A matter of months from now, the worker-bee grunt models will do equal or better for a fraction of the cost. For most people, most tasks, frontier IQ is unnecessary, and that skew only gets sharper. Cutting-edge science and engineering etc will still want the tip of the spear. But for 99.999% of use cases, the 200-IQ model will be fine; you won’t need to pay through the nose for the 5000-IQ one.
The next point (this REALLY matters), per Musk:
"Hardware is hard. Bringing massive compute online rapidly is incredibly difficult. SpaceX has demonstrated exceptional ability in this regard and will only get better."
Elon is right. And SpaceX has the receipts.
Anthropic and Google are already paying BIG
Opus 5.5, which as I write this is blowing everyone’s minds (and soon will be superseded by something else much better, and so on ad infinitum) is an Anthropic product. Anthropic is currently paying SpaceX on the order of $1.25 billion per month to access a \portion\ of the compute SpaceXSI has spun up on Earth.
Remember the guy with three first names. I think this was his point.
If a company can bring AI compute online more rapidly, affordably, and coherently at massive scale than anyone else on Earth (yes, I said on Earth) and demand for that compute is and will remain near infinite, then that company has near-infinite demand from customers willing to spend billions a month for a slice of it.
You’ve also heard of Google. They’re paying almost $1 billion per month to SpaceXSI for compute too. You’d think Google could just build it themselves, right? Well, they can. But can they do it as fast and affordably as SpaceXSI? No. That’s why they’re paying.
Those are not the only AI compute deals, nor will they be the last. The SpaceXSI compute business on (and then off) earth will be an all time money printer.
Big Rocket
Models are optional. Compute is not.
Whether or not SpaceXSI competes successfully in the model layer (for what it’s worth, I think they will), even if they failed miserably and Grok were a total dud (hard to imagine), it wouldn’t matter for the core thesis. SpaceXSI is clearly intent on building out the lion’s share of AI compute infrastructure: first on Earth, then off it.
If I’m right (spoiler: I am), demand for AI will be near infinite for decades, therefore demand for AI compute will be near infinite for decades. SpaceX can build that compute on Earth faster, cheaper, and more coherently at scale than anyone else. That same company is the only one with meaningful capacity to launch AI compute into orbit, where it can scale at a truly stupid rate. There is, after all, a lot of space in space.
SpaceX is poised to benefit enormously from insane, near-infinite demand for AI (and therefore AI compute) over the coming decades. Meaningful success or failure at the model layer does not meaningfully move that needle.
And it’s not like they’re hiding it. Which makes how slow investors have been on this genuinely baffling. The numbers are public. Almost nobody seems to understand what’s about to happen.
Colossus math for midwits
Musk recently shared goals that put simply, could see SpaceX roughly double their total online AI compute in the next THREE MONTHS. Maybe it takes six months. Who cares? SpaceX could roughly double monetized revenue from compute access in a matter of months.
Guess what? They’ll do it again. And again.
SpaceX is computemaxxing.
Within just a few years their AI compute ARR could be in the multiple hundreds of BILLIONS of dollars.
If you are a giga autist, do the math. AI compute spot rate $30-$50B per GW. This assumes demand pressure doesn’t drive it higher. GW online * spot prices = HOLY FUCKING SHIT.
Ark Invest recently estimated SpaceX could end 2027 with around 7.5 GW online.
Do the math.
No. Actually do the math.
There’s a reason AI compute is absolutely \carrying\ my 20-year SpaceX valuation model (find it on Patreon: https://www.patreon.com/solvingthemoneyproblem or check the pinned post on my X profile). This opportunity is colossal. I don’t think most people are even within an order of magnitude of where this is headed.
Let me say it one more time:
Demand for AI will be near infinite for decades to come.
That means demand for AI compute will be near infinite for decades to come.
Do the math.
He who controls the compute controls the universe.
Do you understand?
- Steven
p.s. Don't get me started on Terafab...
