What Could Go Wrong: A Clear-Eyed Look at Terafab and Orbital AI Compute

Most of what gets written about SpaceX's newer business lines falls into one of two buckets: breathless enthusiasm about the future of manufacturing and computing, or dismissive skepticism that treats anything outside the core launch business as a distraction.

Neither is useful if you're trying to actually understand what you own.

This piece isn't a bull case. It's an attempt to lay out, as plainly as possible, what Terafab and orbital AI compute are trying to do, and where the real uncertainty sits — the kind of uncertainty that should inform position sizing and expectations, not headlines.

Two Different Bets, Not One

It's worth separating these out, because they carry different risk profiles even though they're often mentioned in the same breath.

Terafab is a bet on vertical integration into manufacturing — bringing more of the production stack in-house rather than depending on external suppliers. Orbital AI compute is a bet on a physical location decision — that certain categories of computing workload are better served in orbit than on the ground, for reasons of power, cooling, or land constraints.

One is about controlling inputs. The other is about relocating outputs. Conflating them makes both harder to evaluate.

The Case for Terafab

The logic behind bringing manufacturing in-house is familiar from other capital-intensive industries: control over quality, control over timelines, and insulation from supplier bottlenecks that have disrupted aerospace and semiconductor supply chains repeatedly over the past several years.

If it works, it reduces a real vulnerability — dependency on third parties for components that sit on the critical path of every other business line.

The Uncertainty in Terafab

Vertical integration is also one of the most capital-intensive bets a company can make, and it's a bet that takes years to prove out. A few things worth sitting with:

• Manufacturing is not the core competency being tested elsewhere. Launch cadence and reusability are proven. Semiconductor-adjacent manufacturing at scale is a different discipline, with its own learning curve and its own failure modes.

• The payoff is back-loaded. Capital goes in now. Cost advantages, if they materialize, show up years later — and only if the underlying technology roadmap holds.

• Competitive dynamics are not static. Established manufacturers have decades of process knowledge. Closing that gap is possible, but it isn't guaranteed, and it isn't fast.

None of this means the bet is wrong. It means the timeline for knowing whether it was right is long, and the capital commitment in the meantime is real.

The Case for Orbital AI Compute

The pitch here rests on a genuine physical constraint: ground-based data centers are running into limits on power availability and cooling capacity in many of the regions where demand is highest. Orbital placement sidesteps some of those constraints — near-constant solar exposure, radiative cooling in the vacuum of space, and no competition for terrestrial power grid capacity.

If large-scale AI workloads are compute-constrained rather than latency-constrained, an off-planet location becomes a genuinely interesting answer to a real bottleneck.

The Uncertainty in Orbital AI Compute

This is the newer and less-proven of the two bets, and the open questions are more fundamental:

• Nothing at this scale has been done before. There's no long operating history to draw on for orbital data center economics, reliability, or maintenance cadence the way there is for terrestrial facilities.

• Radiation hardening and thermal management in orbit are solved problems for smaller payloads, not necessarily for dense compute clusters at scale. Scaling up introduces engineering challenges that haven't been fully tested.

• The economics depend on launch costs continuing to fall. The entire thesis assumes cheap, frequent access to orbit remains available and gets cheaper — which is itself dependent on the core launch business continuing to execute.

• Regulatory and orbital debris considerations are still evolving. As more entities look at orbital infrastructure, the rules governing it are likely to change in ways that aren't fully predictable today.

This is a bet on a thesis that hasn't been tested at the scale being proposed, made possible by a launch cost advantage that has to keep compounding for the economics to work.

Why Holding Both Uncertainties at Once Matters

The temptation with a position like SPCX is to let the size of the ambition stand in for the strength of the evidence. Big vision, therefore high conviction. That's a mistake in either direction — dismissing the businesses because they're unproven, or assuming they'll work because the team behind them has executed well elsewhere.

The honest position is that Terafab and orbital AI compute are both real attempts at real problems, backed by a company with a credible track record of pulling off things that looked implausible on paper. They are also both, right now, largely unproven at the scale being discussed, with payoffs that sit years out and depend on multiple things going right in sequence.

That's not a reason to panic. It's also not a reason to treat the bull case as settled. It's a reason to size the position and set expectations with the actual uncertainty in mind, rather than the uncertainty implied by a press release or a stock chart.

Where That Leaves You

If you're holding a concentrated position and these business lines are a meaningful part of why you're still in it, it's worth being precise about which bet you're actually making — the manufacturing bet, the orbital compute bet, or both — and how much of your long-term thesis depends on each one landing.

If you want help thinking through how these specific risks map onto your position size and time horizon, Rebellionaire works with investors navigating exactly this kind of concentrated, high-conviction holding. Reach out and let's talk through it.