Google Pays SpaceX $920 Million a Month to Rent Compute Power, How Did Neocloud Suddenly Become the Best Business?
TechFlow Selected TechFlow Selected
Google Pays SpaceX $920 Million a Month to Rent Compute Power, How Did Neocloud Suddenly Become the Best Business?
Everyone wants to be NeoCloud.
Navigating Web3 tides with focused insightsAuthor: degentrading
Compiled by: TechFlow
TechFlow Editor's Note: Everyone wants to become a neocloud (new compute cloud), from xAI and Meta to SoftBank, all jumping in to sell compute power. Author degentrading reverse-engineers a real contract where Google pays SpaceX: 110,000 GB200s, rent per card per hour $11.6, a 100MW data center pays back in less than a year. This article breaks down the cost structure and books of neocloud step by step, answering a core question: why building data centers and selling compute power is the best business right now.
「xAI and Google reach a partnership.」
「Meta signals willingness to sell its surplus compute infrastructure.」
「SoftBank plans to provide 10GW scale AI compute power in the United States.」
No matter where you look, it seems everyone wants to become a neocloud and start selling compute power. Why is this happening?
Let me walk you through the neocloud books first, starting with a real case.
On June 5, 2026, Google announced it would pay SpaceX $920 million per month to rent compute capacity from xAI data centers.
This batch of compute power includes 110,000 Nvidia GPUs, plus CPUs and other memory components, essentially a fully configured data center (Colossus 2).
The GPUs here are Nvidia's GB200 NVL72.
Calculating based on this amount, the unit price per GPU is 920 million / 110,000 / 720 = $11.6 per hour.
Let's hypothesize a 100MW data center, filled with GB200.
Figure Note: Author's calculation of construction costs for a 100MW data center, approximately $50 billion per GW
My estimate is about $50 billion per GW. Calculated this way, the construction cost for 100MW is about $5 billion.
This number aligns with industry estimates. Note that this is the construction cost for GB200. The more advanced the chip, the more expensive it is; for example, for a GB300 construction plan, compute costs are 20% higher.
Jensen Huang saying the cost per GW is $10 billion is actually very realistic—if you factor in the compute cost of Rubin, that's twice Blackwell. Power and other costs are also rising.
We also know each GB200 consumes about 1200 watts, so 100MW converts to approximately 83,333 GPUs.
Applying the economic model of that xAI contract, revenue is 83,333 × 11.6 × 365 × 24 = $8.467 billion per year.
The cost structure is as follows:
Electricity cost per MW per year is $740,000, infrastructure maintenance per MW per year is $250,000, labor per MW per year is $300,000.
Figure Note: 100MW data center books calculated based on xAI contract rent ($11.6 per card per hour), payback period less than one year
Thus we see a crazy scenario: payback period less than one year.
At this point I have to say, this SpaceX contract is ridiculously generous. Is Google facing a severe compute shortage?
If we switch to a less crazy, more normal assumption, such as a price of $6 per hour.
Figure Note: Calculation after rent drops to $6 per card per hour, payback period extends to about two years
The payback period extends to about two years.
I estimate the long-term contracts signed between neoclouds and hyperscalers (hyperscaler) are priced at approximately $4 per hour. Hyperscalers then resell this compute power at higher prices, such as AWS at over $12.
Purely for fun, I turned the construction of a single data center into an estimated income statement, looking like this.
Figure Note: Estimated income statement for single data center construction
Of course, we can argue that GB200 will be worthless after five or six years. But the A100 was released in 2020 and is still being rented out at $1 to $2 per hour. This card rented for about $3 in 2021, and hyperscalers are still renting it at $2 to $3.5 per hour.
Gavin Baker believes the depreciation cycle for GPUs should be extended to about 10 years. I agree with him, especially after agent AI and inference demand pick up.
Ultimately, the current compute market is pricing in a severe shortage.
Anyone who tells you otherwise is covering their eyes.
The reason SpaceX rents out compute power is because at the price it rents out, it makes a fortune (also because Colossus 1 and 2 interconnectivity encountered problems). When Musk signed this contract, he retained the right to terminate the lease early, saying: 「If compute power becomes particularly tight, I said we might need to take it back at some point.」
TL;DR: Everyone wants to become a neocloud now because compute power is particularly tight, and it prints money.
From the facts at hand, being able to build data centers and sell compute power is the best business right now.
This is also why hyperscalers are willing to prepay neoclouds to lock in compute supply.
Emphasizing again, a neocloud's valuation should equal the sum of the net present value of all compute contracts it can secure. Therefore, financing capability, accurate depreciation schedule, and execution capability are the three most important levers affecting valuation.
To those who open their mouths and say this is all circular financing (call it whatever you want), you need to know: Anthropic's gross margin on inference services now exceeds 70%. Those frontier labs that consume the most compute power can switch to positive cash flow at any time, as long as they are willing.
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
@degentradingLSD
