WuBlockchain’s WhiteLine Daily said SemiAnalysis has framed a new way to value SpaceX’s AI infrastructure business: the key metric is not simply how many GPUs it controls, but how much compute it can bring online within a few months. In that view, SpaceX is effectively selling “time-to-compute” — the time it takes from now for large-scale compute capacity to become operational.
Delivery timing, not headline GPU inventory
The report says AI infrastructure deals are often discussed in terms of GPU counts first, but GPUs are not the only scarce input. Data center development takes time across land acquisition, power connection, substations, cooling systems and server hall construction. Those timelines can run longer than chip upgrade cycles, which is why SemiAnalysis puts the focus on time-to-compute.
Under that logic, SpaceX could prepare power capacity and data center facilities first, then buy a newer generation of GPUs closer to the point when power is ready. It could then choose whether to use that capacity internally or lease it out. The value of that model, according to the report, is that energized capacity carries optionality on its own: the company does not have to lock itself into older chips too early, and it can allocate near-term available compute to the highest bidder. SemiAnalysis estimates that delivery timelines for some SpaceX projects could be compressed to about three to five months.
Why customers may pay up for immediate access
The article says Google will pay SpaceX about $920 million per month starting in October 2026 for roughly 110,000 NVIDIA GPUs and other computing resources. It also says Anthropic had previously agreed to buy SpaceX compute for $1.25 billion per month.
Those prices are described as being well above standard long-term cloud contracts because customers may not be able to wait. Their own data centers may still be under construction, while training and inference demand has already arrived. In that setting, what they are buying is less ordinary cloud service than earlier access to compute, similar to paying a rush premium for spot freight capacity during peak periods.
The report also says those prices may not hold over the long run. Google’s agreement, once certain conditions are met, allows either side to terminate with 90 days’ notice. That means current pricing can also be read as shortage-period “capacity insurance,” rather than something that should automatically be projected forward as a conventional long-duration cloud contract.
A $300 billion ARR case depends on several assumptions
SemiAnalysis outlines what the article calls an aggressive scenario in which SpaceX approaches 10 GW of compute capacity by the end of 2027 and generates about $300 billion in annual recurring revenue.
That valuation case depends on several things: whether SpaceX can keep delivery times measured in months, whether NVIDIA can supply enough next-generation GPUs, and whether customers will renew expensive contracts after their own data centers begin operating.
WhiteLine Daily’s conclusion is that SpaceX’s most important AI asset is not just its GPU inventory, but near-term deliverable powered capacity. The report says three figures are worth tracking next: energized GW, revenue per GW and contract renewal timing. Only if those three measures continue to materialize could the short-term premium attached to time-to-compute turn into a lasting business.