AI Lattice / Questions
AI money, in plain answers
Short answers, each linked to its sourced evidence on Does the AI money add up? Data as of 2026-11-01.
Is AI a bubble or a house of cards?
Not a house of cards: most AI data centers are paid for by Amazon, Alphabet, Microsoft and Meta, which can absorb poor returns. The riskiest part rests on one company. OpenAI needs about $60 billion to $105 billion of new money by 2028 just to stay on its own plan, and it cut its latest funding round from $30 billion to about $20 billion on October 30, 2026.
Who is paying for the AI build-out?
Mostly the four biggest US tech companies, from their own cash and, increasingly, bonds: about $700 billion of capital spending in 2026. Outside money fills the rest: equity rounds for OpenAI and Anthropic, SoftBank's borrowing, private credit and bank loans for data-center developers, and supplier investments from Nvidia, AMD and Broadcom.
Is AI spending broad-based or concentrated in a few companies?
The building is spread across about eight large companies, but the outside demand behind it is concentrated. OpenAI and Anthropic have signed about $1.2 trillion of firm compute purchases, and OpenAI alone was about 45% of Microsoft's contracted backlog at the end of 2025 and about half of Oracle's. Use is broad but spending per business is thin: the median paying business spends about $16 per employee per month.
How much money does OpenAI need to raise?
On our stress test of its reported projections, OpenAI needs about $60 billion to $105 billion of new outside money by the end of 2028 even on plan, about $320 billion by 2030 if growth is half the plan, and about $470 billion if revenue stalls. Its July 2026 projections show about $278 billion of negative free cash flow from 2026 to 2030.
Can Anthropic pay for its compute commitments?
Largely yes, on current evidence. Anthropic held $131.7 billion of cash and short-term investments at June 30, 2026 and was cash-positive from operations in the first half; in our harshest case, a revenue stall while paying contracts at about $100 billion a year, it would need its IPO money in 2029.
Who loses if OpenAI cannot pay its bills?
The largest exposures are Oracle (about half of its $664 billion backlog), the lenders behind Oracle's Stargate data centers, SoftBank (about $65 billion invested, much of it borrowed), CoreWeave and Cerebras, and from 2028 Nvidia through up to $105 billion of lease guarantees. These are gross amounts before any recovery. Microsoft and Amazon have about $70 billion invested or owed, small next to their cash flow.
What are the circular AI deals?
A loop is when a supplier invests in a customer that then spends the money back with that supplier: Amazon and Google investing in Anthropic, Microsoft and Amazon in OpenAI, Nvidia in OpenAI, CoreWeave and xAI, AMD giving warrants to OpenAI and Meta. AI Lattice counts about $93B of firm supplier money in such loops against about $822B promised back.
Are Big Tech companies borrowing to pay for AI?
Yes, increasingly. In July to September 2026, Amazon, Alphabet and Meta each spent more on capital spending than their businesses brought in, and Microsoft sold $25 billion of bonds on October 29, 2026. Hyperscaler bond issuance was $132 billion in 2026 through July, and Microsoft and Oracle have $617 billion of signed data-center leases not yet started.
What are the chokepoints in the AI supply chain?
AI Lattice tracks 12, including advanced chip packaging at TSMC, high-bandwidth memory from three makers, ASML's EUV machines, lasers for optical links, gas turbines, grid connections, transformers and government permission to sell models. Each lists who holds it and the sourced evidence.