OpenAI expects revenue to reach $350 billion in 2030.
But through 2030, it also projects $278 billion in negative free cash flow and $856 billion in compute and infrastructure spending.
The demand is clear. The cost of serving it is just as clear.
Revenue potential alone does not pay for the buildout.
Money Is Getting More Expensive
The Fed raised rates to 3.75% to 4.00%, while the 10-year Treasury yield returned to about 5%.
Oil stayed above $100, even after falling for a third session. Higher oil can add inflation pressure, making further rate increases harder to avoid.
Japan is tightening too. The Bank of Japan raised its rate to 1.25%, a 31-year high.
Borrowing is also getting more expensive for households. The U.S. 30-year mortgage rate jumped to 6.95%.
The economic signals are mixed.
Industrial production was flat in August. Factory output fell 0.3%, and the Conference Board’s Leading Economic Index slipped 0.1%.
Inflation-adjusted median household income rose $2,250, or 2.6%, in 2025 to a record $87,460. But those gains were uneven.
Income reached $261,300 for the top 10% of households. It fell $160, or 0.8%, to $20,010 for the bottom 10%.
Investors have more money in retirement accounts, but they are also using more borrowed money.
Fidelity says the average 401(k) balance rose 10.5% from March through June. U.S. margin debt increased $37 billion in August to $1.45 trillion and is up 19% this year.
At the same time, the Strategic Petroleum Reserve fell to 285 million barrels, its lowest level since November 1982.
AI companies need huge amounts of capital. That need is growing while money is becoming more expensive.
AI Needs More Chips, Data Centers and Debt
OpenAI expects revenue to rise from $36 billion this year to $350 billion in 2030.
That growth still does not cover the full bill.
Its $122 billion March cash raise is projected to run out in 2028. OpenAI is discussing another funding round, with investors approaching the company around a $1.2 trillion valuation while it seeks more.
The shortage of computing capacity adds to the pressure.
Dan Ives described chip orders as running at 13 for every chip that can be produced. He said supply and demand might not balance until early 2029.
Anthropic agreed to pay SpaceX roughly $1.25 billion per month through May 2029 for the full compute capacity of the Colossus 1 data center. Another account said Anthropic was targeting a doubling of compute capacity in 2027.
CoreWeave said short-duration Q3 contracts were priced at roughly $40 million per MW. It plans to issue $3 billion of convertible senior notes, with an option for another $500 million.
Costs are moving higher across the chip supply chain. AMD reportedly told partners to expect an approximately 10% price increase because of higher TSMC wafer costs.
Nvidia plans to increase supply. Jensen Huang said the company would double chip volume next year because of strong AI demand.
Companies are also looking for savings. Bank of America estimates that Meta could save roughly $8.5 billion in 2027 by using its own inference chips.
The spending reaches far beyond model companies. Generac signed an $8 billion agreement to supply generators to Amazon data centers. Krux AI secured $22 billion in bank loans to buy Google TPUs.
More demand means more revenue potential. It also means more chips, power, data centers, debt and outside capital.
Consumer AI Faces Payment and Trust Tests
Meta launched Muse, a personal agent that can book appointments, complete electronic forms and monitor home security feeds.
Muse has a free tier, plus plans costing $20 or $100 per month. A weekly stock recap and an X post said it reached number one on Apple’s App Store.
Downloads are only one test.
Meta says Muse runs in an isolated environment, cannot see passwords or payment details, and asks permission before sensitive actions. But users must opt out if they do not want eligible Muse interactions used for AI training.
Meta is also considering taking a share of shopping transactions completed through AI agents, though Alexandr Wang said there is no concrete plan.
Continued use, payment and trust will be harder tests.
Safety adds another challenge. A market recap said an unreleased Astra-family model showed 6 unexpected or concerning incidents during reinforcement learning over the past 6 months.
OpenAI announced a framework to investigate and publicly disclose model misalignment.
A separate report said SpaceXAI informally discussed buying customer and operational data from troubled or defunct startups to train Grok. Those talks may not produce acquisitions.
The AI Case Still Has Clear Risks
Tom Lee named a real bubble and monetary policy intervention as possible threats to the technology bull case.
Dan Ives called political opposition to data centers one of the largest risks to the AI buildout. He said restrictions on U.S. development would help China narrow the technology gap.
Market structure is changing too.
U.S. exchanges are preparing for 23-hour weekday trading starting December 6, 2026. The normal 9:30 a.m. to 4 p.m. session will remain.
Brokers will decide whether customers receive overnight access. Lower liquidity can mean wider spreads and larger price moves.
AI demand remains enormous. But the buildout still needs more chips, data centers, debt, revenue and outside capital while borrowing costs are rising.