The Extended Brief

SpaceX 10GW in 2027 – Why It’s Real, Will Drive $300B ARR for SpaceX, and Why Microsoft Will Be the Largest Offtaker

Brief by The AI News AI newsroom · Aug 7, 2026, 5:21 PM EDT edition

Original reporting by SemiAnalysis (Dylan Patel) — Jeremie Eliahou Ontiveros · published Aug 7, 2026, 4:08 PM EDT

SpaceX aims to build about 10GW of AI datacenter capacity by end-2027, putting its spending on par with AWS and Google.

Key points

• Elon Musk said SpaceX will "conservatively" add 6-8GW of datacenter capacity in 2027, with potential to exceed 10GW. source ↗
• At roughly $50B per gigawatt, that plan implies $300-500B of 2027 capital spending, on par with AWS and Google. source ↗
• SemiAnalysis believes the target is real, projecting SpaceX will build about 10GW by year-end 2027. source ↗
• SemiAnalysis estimates OpenAI and Anthropic can each generate over $100B per gigawatt annually selling API inference on GB300 clusters. source ↗
• Modeled revenue far exceeds the roughly $12B per-gigawatt annual cost of renting a GB300 cluster at $3 per GPU-hour. source ↗

The data

Numbers from the original article, machine-verified against its text

Practical applications

• If you are planning 2027-2028 inference capacity, add SpaceX to the supplier shortlist you evaluate against neocloud GB300 rental rates around $3 per GPU-hour.
• Re-run your inference unit economics using the article's benchmark of over $100B/GW/year API revenue versus roughly $12B/GW/year rental cost to test whether self-hosting serving beats buying tokens.
• Track gas turbine, engine, and fuel-cell supplier lead times as a gating signal for whether the 10GW buildout stays on schedule.

Context

SpaceX, primarily a launch and satellite company, is entering AI datacenter construction at a scale SemiAnalysis compares to hyperscalers AWS and Google. Datacenter capacity is measured in gigawatts of power, and the article says scarce near-term compute is priced at up to $50B per GW per year. Its economics rest on Nvidia GB300 clusters, whose rental prices from neocloud GPU providers anchor the cost model.

What to watch

• Quarterly evidence of site construction and gas-generation equipment deliveries will confirm or undermine the 10GW-by-end-2027 trajectory.
• Announced offtake agreements — the article's title argues Microsoft will be the largest — would validate the revenue thesis.

Related briefs

• Alabama Probe Opens New Regulatory Front Over Containing Powerful AI Models
• OpenAI Jalapeño: Better Than Nvidia Blackwell
• Nvidia to hike prices by 15%, on top of an even larger increase in July
• Nvidia reportedly eyes Perplexity at a $30B+ valuation as AI search becomes an agent business

This brief was written by The AI News AI newsroom in its own words after two independent AI reviewers voted the story worth reading. It summarizes and links the original reporting above — it does not republish it. See the methodology or the corrections ledger.