Two briefs project about $10.3 trillion in AI and data-center infrastructure spending between 2025 and 2032.
At the same time, model prices are falling. Use is growing, but the amount customers will pay for it is still unsettled. That leaves two questions over the buildout: how much capacity gets built, and what will people pay to use it?
The AI Buildout Carries Large Commitments
One brief projects annual AI and data-center investment averaging 3.63% of US GDP from 2025 to 2032.
A @value-investing blog says Alphabet, Amazon, Meta, Microsoft and Oracle planned $800 billion of spending this year. It says those companies have more than $3 trillion in commitments on and off their balance sheets.
The same blog says 70 to 75% of their AI revenue comes from OpenAI and Anthropic. Akamai ($AKAM) landed an $11.6 billion, seven-year cloud deal with Anthropic. The spending plans are large, and those two customers account for a large share of the revenue the blog describes.
The funding needs are large too. The @pboyle blog says SB Energy sought a valuation around $50 billion without an operational data center. SB Energy’s prospectus says it needs $174 billion to build what it promised. Nvidia guaranteed up to $105 billion for SB Energy’s Ohio campus, which is set to use Nvidia hardware exclusively for 20 years.
The Financial Times reported that OpenAI expects to burn through nearly $280 billion by the end of 2030.
These commitments depend on demand that still has to pay for the buildout.
Model Use Is Rising as Prices Fall
Financial Times-published transaction data show OpenRouter’s weekly routing volume growing roughly 25,000% since early last year. People are using models much more.
But the price of that use is moving the other way. Anthropic and OpenAI introduced variants priced 40% and 50% below their prior tiers. An Epoch AI study estimates that, since 2023, the cost of reaching a fixed level of model performance has fallen roughly 13-fold each year.
Ramp’s co-CEO said routing tasks to cheaper models cut Ramp’s AI spending by 40%. That is a clear example of how a customer can use AI while paying less for it.
Growing use does not yet tell us how much revenue model sellers will collect. That matters when so much infrastructure spending is planned around future demand.
Valuations and Product News Tell Different Stories
Anthropic is expected to seek a public valuation around $2 trillion. The @pboyle blog puts that at approximately 31 times revenue, while warning that unofficial figures suggesting a roughly $65 billion annualized run rate in August need care.
The same blog says Nvidia trades at less than 17 times forward earnings, despite estimated revenue of $410 billion this year. Its gross margin was 75% last quarter, while consensus estimates put it below 72% by year-end.
There is no public Anthropic price to compare yet. Its expected filing has not appeared. OpenAI has pushed its listing to next year, and only three IPOs have launched since Labor Day.
Public stocks still moved on AI product news. Microsoft ($MSFT) rose about 3.7% after releasing new Copilot tools. Qualcomm ($QCOM) gained about 4%, and Dell ($DELL) about 5%. A separate post claimed Meta’s Muse had around 3m downloads and reported multiple $META price target upgrades.
Elsewhere, Michael Burry told his Substack readers he bought QXO preferred stock. The @equityempireresearch blog says QXO is tracking around $20 billion in revenue over 12 months after three major acquisitions, toward Brad Jacobs’ $50 billion target for 2030.
Rates, Housing and Trade Add Pressure
Stocks closed higher as oil pulled back on hopes of a US-Iran deal. The 10-year Treasury yield also backed off after reaching a multi-year high. The @pboyle blog says a purchasing managers survey was accompanied by a rise in that yield to 5.23%.
Matt Kennedy of Renaissance Capital explained why rates matter for AI: higher rates reduce the present value of future profits and make borrowing to build data centers more expensive.
The inflation picture is disputed. Tom Lee says a Sept 30 change in how PCE is calculated could take 20 to 40 basis points off its yearly rate. He also says oil would need to reach near $150 to push inflation back up. Another commentator says the current rise in headline CPI comes almost entirely from energy, while core CPI is 2.4%.
Cullen Roche points to CPI falling from about 9% to about 3% in 2022 and 2023 as the Fed raised rates above 5%. He uses that to argue against the claim that rate hikes themselves raise inflation. Other commentators question whether higher rates can slow demand for AI compute and energy, or argue that interest costs feed into prices. @fundstrat_direct says higher rates will not slow AI spending, but flash memory does not really appear in consumer CPI. Fed Chair Kevin Warsh said a week earlier that he was committed to returning inflation to the central bank’s 2% target.
Housing shows a similar split in the reports. August new home sales reached an eight-month high as builders cut prices and offered incentives despite 7% mortgages. A former real estate developer writing on X says high rates have stalled projects and left developers unwilling to build.
Trump renewed a threat to stop trading with countries that have trade surpluses with the US unless the Fed cuts rates. After Trump said Bombardier must build in America to sell there, the company told CNBC it directly employs people in more than 20 US states and has sites in 10. Canada was preparing retaliatory tariffs on about $20 billion of US goods.
A @geopoliticaleconomyreport speaker argues that Trump’s September meeting with Xi Jinping was aimed at helping US corporations gain access to China’s market. The speaker says that corporate interest explains Trump’s softer approach to China after his trade war backfired.