Menlo Park, July 10, 2026, 06:11 (PDT)
Meta Platforms NASDAQ:META rolled out its first public AI model service for developers on Thursday. CEO Mark Zuckerberg told Bloomberg that demand from outsiders to use Meta’s computing power is strong and could rival internal projects. “The offers that you get for using the compute are so high,” Zuckerberg said. But he added Meta isn’t sitting on extra capacity. Ai.Meta
This pairing shifts how investors may see it. U.S. developers now get access to Muse Spark 1.1 via the Meta Model API, costing $1.25 per million input tokens and $4.25 per million output tokens. An API is just a link allowing other apps to tap the model; tokens are small chunks of processed data. Futurum Equities strategist Shay Boloor called it a “much clearer monetization bridge” if the model’s coding results stay strong. Reuters
Meta faces a big test with its massive spending plan. The company expects to spend $125 billion to $145 billion in capex in 2026, mainly on servers and data centers. That compares with $72.2 billion in 2025. The $135 billion midpoint is up 87% and makes up 67% of Meta’s $201.0 billion in revenue from last year. Meta shares closed at $631.48 on Thursday, up 4.7% ahead of Friday’s session.
Meta has an internal plan that targets 7 gigawatts of computing capacity this year, and 14 gigawatts by 2027. A gigawatt is used to measure power capacity. The company also plans to start producing its Iris AI chip in September. “You can’t become an AI titan if you are dependent on another company for chips,” Forrester’s Mike Gualtieri said, pointing to lower model prices from custom silicon. Reuters
SemiAnalysis puts a 200 megawatt slice at more than $10 billion a year if sold at top prices. That 200MW is just 1.4% of Meta’s planned 2027 capacity, around 5% of 2025 sales, and 16% of Meta’s $62.8 billion capex increase. If you ran the math in a straight line, you’d come out over $700 billion—more than triple last year’s revenue—so this is a scarcity number for a small block, not for the full fleet.
| Investor comparison | Calculation | Result |
|---|---|---|
| Share of 2027 external supply | 0.2 GW ÷ 14 GW | 1.4% |
| Yearly projected revenue as share of 2025 sales | More than $10 billion ÷ $200.97 billion | About 5% |
| Projected sales compared to capex jump | More than $10 billion ÷ $62.79 billion | Almost 16% |
| Guess at revenue per GW | More than $10 billion ÷ 0.2 GW | More than $50 billion per year |
| Fleet-wide math — not an outlook | More than $50 billion × 14 GW | More than $700 billion |
Cloud divisions make clear both the opportunity and the shortfall here. Operating margin shows operating profit as a percent of sales. Meta hasn’t disclosed any revenue or margin figures for its Model API preview.
| Platform | Status or Q1 2026 revenue | Year-on-year growth | Operating margin |
|---|---|---|---|
| Meta Model API | Still in public preview; revenue isn’t out | N/A | N/A |
| Amazon.com NASDAQ:AMZN’s AWS | $37.6 billion | 28% | 37.7% |
| Alphabet NASDAQ:GOOGL’s Google Cloud | $20.0 billion | 63% | 32.9% |
Meta has fewer options. The API brings in cash from models already, but offering raw compute would turn Meta into a neocloud—basically renting out AI chips and big blocks of data-center space. Bloomberg said Meta is still looking at both the model hosting idea and the raw compute plan, and there’s a chance the strategy shifts.
This sets up a channel conflict. Meta has agreed to spend as much as $21 billion with CoreWeave NASDAQ:CRWV for compute capacity through 2032, according to a CoreWeave filing. That means Meta is lined up as both a major customer and a potential selective rival, and the size of the deal signals strong demand inside Meta.
The risk is scarcity premiums could disappear before Meta has the supply ready, or spending time renting chips could pull focus from core areas like ads or model training. Motley Fool writer Marc Guberti said Meta is “several years away” from bringing neocloud services because it’s still buying third-party capacity and major new sites aren’t quick to build. Full services would also need sales support teams, stronger security, and firm service-level agreements for uptime and performance. The Motley Fool
For investors, what’s important in the next updates isn’t the cloud talk, but how much capacity Meta lets outsiders reserve, the length of those contracts, what prices look like, profits after covering direct service costs, and if customer deals cut into future capital spending. Meta doesn’t have to copy AWS for this to hit returns. Selling even a small piece can help the numbers on the buildout, so long as it doesn’t drag down products powered by the rest.
Related articles: NASDAQ:META
