Microsoft (NASDAQ: MSFT) could emerge as a dominant force in AI inference, matching frontier labs like OpenAI and Anthropic in revenue potential, according to semiconductor research firm SemiAnalysis.
The firm’s proprietary models estimate that frontier AI companies can generate more than $100 billion in annual revenue per gigawatt of compute capacity, underscoring the fierce race to secure data-center power.
AI inference refers to the process of running a trained AI model in the cloud to generate real-time responses, predictions, or outputs for end users.
SemiAnalysis argues that Microsoft is uniquely positioned to capture these economics, citing the company’s access to OpenAI models without bearing the burden of training costs.
“Having full access to OpenAI models, they can generate the same revenue and margin per MW, while paying none of the training costs,” SemiAnalysis wrote in its analysis.
The April 2026 renegotiation of Microsoft’s deal with OpenAI proved pivotal, removing the old 20% revenue share arrangement that had previously weighed on Microsoft’s returns.
“Satya [Nadella] nailed the negotiations with OpenAI: the deal reworked in April 2026 dropped the old 20% revenue share from the equation. Put simply, Microsoft has a giant incentive to procure as many MWs as possible, as fast as possible.”
Microsoft Azure’s revenue is currently growing at approximately 42% year over year, but SemiAnalysis suggests that figure could exceed 100% growth by next year as more capacity comes online and is directed toward inference workloads.
Microsoft has already signed 10 gigawatts of data-center contracts year-to-date, representing over $300 billion in total contract value, with SemiAnalysis expecting additional agreements to be announced for capacity coming online over the next two years.
SpaceX is emerging as a surprising but formidable competitor in the compute market, with CEO Elon Musk stating during the company’s first-ever earnings report that SpaceX “conservatively” expects to build and deliver an additional 6 to 8 gigawatts of capacity next year, with potential to exceed 10 gigawatts.
SemiAnalysis noted that its analysis of potential sites, power equipment, and construction timelines suggests SpaceX’s target is achievable, pointing to the company’s ability to build at remarkable speed.
The company constructed the 300-megawatt Colossus 1 facility in just 122 days, while its Southaven power plant expanded from 495 megawatts in February to 1.7 gigawatts by July, according to the report.
SemiAnalysis estimates SpaceX could reach approximately $300 billion in annual recurring revenue even if only half of its incremental 2027 capacity is monetized, with the remainder consumed by xAI and Cursor for model training.
On retail sentiment platform Stocktwits, sentiment was rated “neutral” for MSFT and “bullish” for SpaceX as of Monday, reflecting divergent investor enthusiasm between the two compute heavyweights.
MSFT stock has gained over 28% since its earnings report on July 20, while SPCX has climbed 15.5% since SpaceX’s earnings report on August 12, reflecting growing investor confidence in the AI infrastructure buildout.
