Key Highlights

  • An analysis puts AI's 2030 Revenue need at USD 3 trillion a year to justify planned data centers.
  • Anthropic's prospectus shows a run-rate near USD 81 billion at October's end, against about USD 50 billion reported for OpenAI.
  • Anthropic's first-half revenue reached USD 16.3 billion, up nearly 14 times, while compute costs took 58% of third-quarter revenue.
  • A tally of 41 local Data Center ballot measures in 10 states found results skewing almost unanimously against development.

The Arithmetic of the Buildout

Anthropic has set a price range of USD 70 to USD 78 a share for its Nasdaq listing, and coverage of the filing puts the implied valuation at roughly USD 1.3 trillion, and up to USD 1.4 trillion at the top of the range. OpenAI has been reported to be leaning toward waiting until next year for its own listing.

An analysis of the planned data center buildout finds that AI needs around USD 3 trillion in annual revenue by 2030 to offer a conventional rate of return. Estimates of the required revenue vary among analysts, because they rest on assumptions about returns and costs. Suppliers are also financing demand: Nvidia's filing for the quarter ended July 26 shows USD 47.9 billion of non-marketable Equity securities, mostly stakes in private companies, up from USD 3.8 billion a year earlier. The spending side is visible in company results too: Alphabet alone reported USD 51.4 billion of Capital spending in the third quarter and maintained a 2026 range of USD 195 billion to USD 205 billion.

Where Revenue Stands

Anthropic's prospectus amendment reports a run-rate of approximately USD 81 billion as of Oct. 31, up about USD 10 billion from the end of September. The company defines run-rate as usage fees over the trailing 28 days multiplied by 13, plus monthly subscription fees multiplied by 12. The filing says the figure is not a substitute for revenue under accounting rules, that holiday periods can move it, and that the company anticipates its growth rate declining over time.

Reported revenue is smaller. Anthropic took in USD 16.3 billion in the first half of 2026, compared with USD 1.2 billion a year earlier, and about USD 17.0 billion in the third quarter. OpenAI's annualized revenue was reported to be near USD 50 billion, lower than earlier reports had suggested.

That money sits at the top of a funnel. Models run on chips designed by Nvidia and made by Taiwan Semiconductor Manufacturing, and the chips need data centers and power.

An Unusual Business Model

Selling AI has so far been lucrative for its leaders, yet the Economics differ from conventional software. Software had high switching costs, network effects and near-zero cost to serve an extra customer. AI models offer the reverse: switching can take one line of code, one chatbot resembles another, and every query runs on a scarce chip in an expensive data center. The prospectus shows the cost, with compute and infrastructure expense rising to 58% of revenue in the third quarter from 48% in the second.

Pricing power has held because users pay a premium for the best model. But each leader has only a window of weeks or months to recoup Training costs before a successor arrives.

Developers are pursuing stickier revenue. Anthropic has formed a venture with Blackstone, Hellman & Friedman and Goldman Sachs to offer AI services to businesses. Other routes include applying models directly in areas such as Drug Development and robotics, or limiting access to a few high-paying customers, which would also curb the copying of model outputs.

Politics Enters the Model

The midterms put AI infrastructure on the ballot. Of 41 local data center ballot measures in 10 states on Nov. 3, at least 25 were approved, and a review of the results found them skewing almost unanimously against development. Two of the approved measures, in Illinois, were advisory rather than binding.

In Texas, Democrat Gina Hinojosa won the governorship after pledging a moratorium on all data center approvals until the legislature passes new protections. Governor Greg Abbott had already paused approvals for data centers seeking to connect to the state's power grid, citing reliability concerns.

What the Gap Asks of the Industry

The USD 3 trillion figure works less as a forecast than as a yardstick for what the buildout has to demonstrate. The reported run-rates of the two largest developers, about USD 81 billion and about USD 50 billion, together come to less than 5% of it, and run-rate is not revenue. The sources of future growth are still taking shape: enterprise spending per employee outside the technology sector, specialist products and applications built directly into areas such as drug development and robotics.

Each depends on customers paying for the best model rather than the cheapest one that is good enough. Local restrictions add a second constraint, since ballot results and state-level pauses affect where capacity can be built. Anthropic's quarterly disclosures Supply the first regular readings on how far the climb has progressed, including how its compute costs move against revenue and whether run-rate growth slows as the prospectus anticipates.