1.
The necessity of merging SpaceX and Tesla is whether the two companies can lower the cost of orbital data centers through integration.
Whether SpaceX’s orbital data center and other commercial models can work ultimately comes down to economics.
Analyst Wood Mackenzie estimates that currently building a 1GW orbital data center would cost about $170 billion—more than three times that of an onshore data center of the same scale. Of that, satellite and launch costs account for about 60%.
No matter how many advantages orbital computing has in terms of energy, land, and grid constraints, as long as the unit compute cost remains significantly higher than the ground-based option over the long term, it will be difficult to commercialize at scale.
So the question becomes: when producing the same unit of effective compute, under what circumstances can the full lifecycle cost of putting it in orbit come close to, or even be lower than, doing it on the ground?
2.
SpaceX’s route as described today can roughly be summarized into three things: lowering launch costs, improving the chips’ performance-to-power ratio, and then turning satellites and computing equipment into large-scale industrial products through vertical integration.
Tesla happens to have the other half of the capabilities that SpaceX lacks here.
In November 2025, Musk discussed Tesla’s AI5 chips in a conversation with Ron Baron.
AI5 is mainly aimed at inference workloads for autonomous driving and Optimus. Musk’s stated goal at the time was to achieve, for these kinds of tasks, Nvidia chips’ performance-per-watt by a factor of two to three, while the cost could be only about one-tenth.
On the training side, Tesla will still use Nvidia heavily, so this doesn’t mean AI5 can fully replace GPUs. A more reasonable interpretation is that Tesla wants—on highly fixed inference workloads—to use custom ASICs to achieve a higher performance-per-watt ratio and a lower unit cost per compute.
This line of thinking is especially important for orbital data centers.
In onshore data centers, lower chip power consumption means lower electricity bills and reduced pressure on liquid cooling. But in the orbital environment, for every watt you reduce, the impact propagates along the entire system.
The same computing task, if it only needs half the original power draw, first allows fewer chips and a reduced power supply requirement. Then the solar panel area can be smaller, the distribution equipment and heat sinks can also be reduced, the total satellite weight correspondingly declines, and finally the mass that needs to be launched into orbit—and the number of Starship launch missions—would both decrease.
SpaceX’s later actions also indicate that the company really wants to extend this approach to orbital computing.
In materials submitted to the SEC, SpaceX has already disclosed its Terafab plan. This project doesn’t just consider manufacturing one kind of chip. One line is aimed at Tesla’s vehicle and robot inference; another is specifically for the space environment and the infrastructure for orbital computing.
This year, Intel has also joined the Terafab project, planning to provide chip design, manufacturing, and advanced packaging capabilities.
At present, these collaborations are still mostly in framework form, and there is a great deal of uncertainty about the final scale and how they will actually be implemented.
But the direction is clear: if orbital data centers enter a large-scale buildout phase in the future, SpaceX doesn’t want to rely entirely on external general-purpose GPUs. Instead, it wants to have a set of computing hardware customized for the orbital environment.
3.
The second capability Tesla can provide comes from the energy system.
Orbital data centers don’t need to connect to the terrestrial power grid, but they still require a complete set of power generation, energy storage, power management, and power conversion systems. Solar power is only part of the energy source. How to deliver that electricity stably to the computing chips also requires a large amount of power electronics technology.
Tesla’s long-term accumulation in batteries, power electronics, inverters, energy storage, and thermal management has clear engineering overlap with these needs.
Tesla’s Q2 2026 filings show that in the first half of this year, revenue from selling products such as Megapack to SpaceX alone was about $405 million, and Tesla has also invested $200 million in SpaceX. The industrial synergies between the two companies have already begun to take shape.
4.
What’s even more worth noting may be manufacturing.
Traditional aerospace products typically have limited quantities, very high unit prices, and production methods that are closer to project-based execution.
But if an orbital data center ultimately needs thousands, tens of thousands, or even hundreds of thousands of computing satellites, it won’t be possible to continue using the traditional aerospace industry’s production model.
In SEC filings, SpaceX has explicitly mentioned a hope that satellite manufacturing will gradually move toward large-scale manufacturing modes similar to the automotive industry.
That is precisely one of Tesla’s most valuable capabilities.
Over the past decade or more, Tesla has been trying to break down complex products into standardized, automated, assembly-line-style manufacturing processes—from batteries and motors to complete vehicles, and then to large-scale casting and machine vision.
Starlink itself is already moving in a similar direction. If, in the future, computing satellites can gradually adopt automotive-industry-style large-scale production, then Tesla’s accumulated expertise in automated factories, supply chains, robotics, machine vision, and manufacturing software could also be integrated into SpaceX’s production system.
Orbital data centers also have a more long-term problem: maintenance.
If a GPU or server fails in an onshore data center, engineers can simply replace it.
Once orbital equipment is damaged, repair costs are much higher. If, in the future, Starship truly achieves low-cost, high-frequency trips back and forth, and then Optimus or other dedicated robots also advance, there should be room to further reduce the costs of orbital maintenance in theory.
But this part is still far from commercialization for now.
5.
If you break down the current orbital data center cost of roughly $170 billion per GW between the two companies, their complementarity becomes even more apparent.
First of all, there’s launch.
In market-circulated calculations, the launch costs for a 1GW orbital data center could reach about $50 billion. Tesla can’t easily solve this directly; the core dependency is still on Starship.
Assuming the implied launch cost today is about $5,000 per kilogram, and that in the future a fully reusable Starship can bring it down to $500 per kilogram, then for a launch mission of the same scale, the theoretical total cost could drop from $50 billion to about $5 billion.
If, in the long term, it becomes possible to get even closer to $100–$200 per kilogram, the constraint that launch imposes on an orbital data center’s economics would fundamentally change.
The second part is chips and computing systems.
This is exactly the area where Tesla, Terafab, and Intel are most likely to play a role.
If, in the future, custom ASICs can simultaneously reduce both the chip cost per unit of compute and their power consumption, the savings would be not only the money spent on purchasing GPUs, but also a simultaneous reduction in power generation, cooling, satellite mass, and launch demand.
Compared with simply lowering the chip selling price, improving the performance-to-power ratio is more significant for orbital computing.
The third component is the satellite platform, solar power, thermal management, and structural manufacturing.
What’s needed here is SpaceX’s satellite industrial ecosystem and Tesla’s capability for large-scale, standardized manufacturing.
So, an orbital data center can ultimately be understood as the sum of several cost components: computing chips, the power system, the thermal management system, the satellite platform, launches, and subsequent maintenance.
SpaceX’s strongest capabilities are concentrated in satellites and launches, while Tesla’s advantages are concentrated in chips, energy, and manufacturing. With these two technology stacks combined, the complementarity is indeed very strong.
This also explains why discussions keep emerging in the market about further integration between Tesla and SpaceX.
6.
As of now, the two companies have not publicly announced a merger, and from an industry perspective, the merger itself in legal terms is not necessarily the most critical issue. The two sides can already achieve a lot of synergy through Terafab, supply-chain cooperation, shared intellectual property, and shared engineering resources.
What really needs to be observed is whether Musk can continuously optimize rockets, satellites, chips, energy, and industrial manufacturing within a single cost curve.
Because what an orbital data center ultimately competes on is not who has the grander technical vision, but a few very real numbers: how much it costs to send cargo into orbit per kilogram, how much AI compute can be produced per watt of electricity, how cheaply each computing satellite can be manufactured, and how much maintenance cost is incurred when equipment fails.
If breakthroughs can be achieved simultaneously in Starship, in-house ASICs, and large-scale manufacturing at the level of the automotive industry, then the current orbital computing cost of about $170 billion per GW would have a chance to gradually move closer to the $40–$60 billion per GW range of onshore data centers.