Editor’s Note
AI demand kept producing strong business numbers, but investors spent more of the period asking how much it costs to build the infrastructure behind that demand. By the end of the week, high inflation, expensive oil and financing worries had hit technology shares hard, even as parts of the market held up better.
AI Demand Is Growing, While Funding Gets More Scrutiny
The spending plans are large and still rising. Creative Planning’s Charlie Bilello put projected 2026 investment by Google, Amazon, Meta, Microsoft and Oracle at $789 billion, up 89% from 2025. His expectation for 2029 was $1.2 trillion a year. Broadcom was also seeking more than $50 billion for OpenAI’s custom chip program, though the talks were preliminary and the amount could change.
There are signs of demand and revenue. TSMC reported its sixth straight record quarterly profit, with net income up about 69% from a year earlier on strong AI demand. ASML reported a 57.5% gross margin and said its planned supply for 2027 and 2028 would not fully meet customer demand. Micron’s management said demand would exceed supply in 2027 and 2028, with more than 75% of its 2027 output already committed.
But these numbers do not settle whether every company building AI infrastructure can fund it profitably. CoreWeave’s contracted backlog grew from roughly $26 billion in Q1 2025 to over $104 billion in Q2 2026, according to Bank of America. Its bonds were reported to yield near 15%, while its CEO said the company had raised $35 billion in 18 months. Those figures describe both a large customer pipeline and the cost and scale of financing the buildout.
Revenue figures also need care. The Financial Times reported that OpenAI told investors its annualized revenue was approaching $50 billion at September’s end, below a roughly $70 billion figure widely reported the previous month. The digest attributed the gap to different accounting methods. Meta’s personal agent, offered with free, $20 and $100 monthly plans, is one attempt to put AI into a paid product. Meta is also exploring a cut of shopping transactions, without a settled plan.
The Tech Selloff Was Severe, but Not Market-Wide
The Nasdaq 100 fell 5.5% for the week, its largest weekly decline since April 2025. The index ended 7% below its high from 10 days earlier, and semiconductor stocks fell for 9 straight sessions through October 16. AI infrastructure shares were among the hardest hit as borrowing costs and debt came under scrutiny.
The weakness had been building while the S&P 500 stayed close to a record. The index was 0.2% from its all-time high, yet 84% of its stocks were at least 10% below their own highs. A basket of 42 AI-related stocks supplied 67% of the index’s total price return since the start of 2024. Those figures show how a strong index level can coexist with weak performance across many individual stocks.
There were exceptions to the selling. On one day, 378 S&P 500 stocks rose, the most since June 12, and healthcare reached new all-time highs, according to the digest. Investors also put $3.8 billion into the 3x leveraged semiconductor fund SOXL during the week, even as it fell 35% over the previous seven trading days. The inflows show buyers were still stepping in, but they do not show that the decline had ended.
Oil and Inflation Complicated the Rate Picture
September consumer inflation rose 3.7% from a year earlier, while core inflation was 2.6%. Wholesale prices rose 1.3% in September and 6.3% from a year earlier. The digests attributed the headline consumer increase largely to energy, and Brent crude approached $110 as tensions around the Strait of Hormuz and Iran continued.
That left markets weighing inflation against signs of a softer labor market. The September jobs report showed 29,000 new payroll jobs and 4.2% unemployment, according to the macro digest. Fed officials said more rate hikes were likely, though some saw no urgency to act at the October 27-28 meeting. Market estimates for a hike moved sharply: one digest put the odds at roughly 40-50%, while a later post reported a fall from 55% to 13%. Those are changing market expectations, not a Fed decision.
The 10-year Treasury yield fell to near 5.1% as investors sought safety amid AI financing fears. Oil, inflation, rate expectations and technology financing were therefore moving at once, but not in one direction. The period did not resolve how the Fed would weigh inflation against the labor market, or how long expensive oil would last.
Counter-Thesis and Risk Watch
Skeptics focused on whether AI infrastructure debt can be repaid on time. @wallstreetmillennial described large planned campuses, borrowing and leases with payments on different timelines, and said the Texas campus appeared behind schedule. On CoreWeave, a separate skeptic said credit default swaps implied a 58% chance of default within five years. These are attributed concerns and interpretations, not established outcomes.
There were also opposing views on the selloff. Dan Ives called the decline in CoreWeave, IREN and Nebius a buying opportunity, while estimating that 5-10% of planned data centers may never be built because of financing and political risks. Strong demand is visible in the backlog and memory commitments. Whether financing permits the planned buildout to be completed remains open.