AI Investment Meets the Price of Credit – October 15 Daily Briefing

The contrast I see today is striking: some chipmakers are already asking about equipment years from now, while the cost of borrowing to build AI infrastructure is being questioned today. Strong demand and affordable investment are not the same thing.

Information cutoff: October 15, 2026, 8:45 a.m. Korea time. I selected new results, product releases and market reactions from October 14. I am not repeating the previous day’s coverage of Gemini 4 Argon, frontier-AI containment proposals, CXMT and the Korean selloff, the IMF’s financial-stability warning, or U.S. crypto policy and the midterms.

1. ASML customers are asking about equipment for 2029 and 2030

ASML reported third-quarter sales of €11.7 billion and net income of €5.0 billion. It put its 2026 sales outlook at €44 billion–€45 billion. The longer-range signal was that some customers had expressed interest in placing equipment orders for 2029 and even 2030. ASML said its currently planned supply for 2027 and 2028 would not fully meet customer demand. It plans to raise 2027 capacity for low-NA EUV and DUV immersion systems by 30%, and is investigating a further 30% increase for 2028. ASML’s Q3 results and capacity outlook

I would not describe those expressions of interest as confirmed orders or booked revenue. Still, they suggest that customers are planning for advanced chipmaking capacity well beyond the next few quarters. Lithography equipment patterns extremely fine features onto silicon wafers, so its availability can limit how quickly leading-edge chip production expands. That is evidence against assuming that AI-related hardware demand is about to disappear. It does not, by itself, justify every semiconductor valuation or guarantee that all planned investment will earn a return. I would keep actual orders, deliveries, wafer output and packaging capacity as separate checks.

2. U.S. inflation reaches 3.7%, and rate expectations shift again

The U.S. Consumer Price Index rose 3.7% over the year through September, up from 3.4% in August. It increased 0.6% in September alone. Gasoline prices rose 9.1% during the month and accounted for more than half of that monthly increase. Core inflation, which excludes food and energy, also moved up: 2.6% over the year versus 2.4% in August. That means the report was not only about fuel, though it also does not mean every price was rising at the same pace. U.S. Bureau of Labor Statistics September CPI

CME FedWatch pricing for an October rate hike rose from roughly 22% the day before the report; one later snapshot put it near 35%, while other reports captured a higher immediate move. These are changing estimates inferred from futures prices, not a Federal Reserve decision. The Nasdaq fell 1.9% that day and the S&P 500 fell 1.1%. Market close and inflation reaction, CME FedWatch odds reported after CPI

The link I am watching is straightforward: if inflation proves persistent, borrowing costs and the present value investors assign to future profits can both come under pressure. That makes a long-term AI growth story less sufficient on its own. For teams paying for cloud and model services in dollars, U.S. rate moves can also reach their local costs through exchange rates. I would track that sensitivity rather than assume a dollar price is the full cost.

3. CoreWeave’s bond yield nears 15%, putting AI funding under scrutiny

On October 14, CoreWeave shares fell more than 15%, and the market yield on one of its bonds due in 2032 approached 15%. Other AI-infrastructure and cloud-related stocks also declined. The move showed investors demanding more compensation to hold debt tied to a capital-intensive data-center buildout as inflation and interest-rate concerns rose. Reuters market report

The 15% figure is not the interest rate on every dollar CoreWeave owes. A bond’s market yield is the return investors demand at its current price; it can rise when the bond price falls. Higher yields can make new borrowing or refinancing more expensive, but they do not automatically reset every existing loan to 15%. Even so, access to GPUs is only one part of the equation. The cost of building the site and the time it takes to turn that capacity into revenue matter just as much. Yesterday’s IMF warning described the broader financial risk; today’s bond market gave one company-specific example of how that concern can be priced.

4. Hesai’s Kosmo turns real spaces into 3D robot-training environments

On October 14, Hesai introduced Kosmo, a system combining the Kosmo One spatial camera with the Kosmo World cloud platform. The camera integrates lidar, an imaging sensor and an inertial measurement unit to capture a place’s geometry, color and position data. The platform turns those scans into 3D environments for robot training and simulation. Hesai also announced partnerships with embodied-AI companies Galbot and Dexmal. Hesai’s Kosmo announcement

I see the practical problem it is trying to solve as a data problem. Language models can learn from huge collections of text, but robots need examples of how objects and spaces behave in the physical world. Hesai says a Dairy Queen training environment that previously took about 300 hours of manual modeling could be scanned and reconstructed in about 15 minutes. In that one deployment, the company reported a 99% task-success rate for a Sharpa robot. That is a company-reported result from a specific site, not a general success rate for robots in other stores or tasks. I would treat it as a useful demonstration, not proof that the broader problem is solved.

5. NVIDIA pledges five years of compute for UK AI-safety research

NVIDIA said it would provide $250 million worth of computing access over five years for AI-agent safety and security research in the U.K. The plan is valued at $50 million in compute access each year; it is not a single cash grant. Most of the support is for the wider U.K. research community, with a portion intended for Oxford’s BOLD, UCL’s SOFAIR and Imperial College London. NVIDIA’s announcement

I see this as distinct from yesterday’s scientists’ call for international oversight and containment rules. That was a proposal about governance; this is a concrete commitment of computing resources for research. Testing agents that can use tools requires more than checking question-and-answer responses. Researchers need isolated environments in which to probe misuse, vulnerabilities and failure modes. The donation does not guarantee useful findings or safer products, so I will look for the tests and results that become public.

What I’m taking away today

These five developments point to different speeds in the AI economy. ASML describes customers asking about equipment years ahead, while CoreWeave’s bond illustrates how expensive it can be to finance new capacity. Inflation could widen that gap. Robots also need more than a capable model: they need usable data from real spaces. And evaluating powerful agents requires computing resources of its own.

I do not want to reduce the cycle to whether demand exists. I would separately ask whether demand becomes orders and revenue, whether the investment can be financed, and whether the data and execution environments are good enough to use these systems safely. TSMC’s earnings conference is scheduled for later today; it will be a next opportunity to compare ASML’s long-range signal with actual sales, margins and investment plans. At this 8:45 a.m. cutoff, those results are not yet available. TSMC Q3 results and conference schedule

I did not add a crypto or blockchain item because I found no new confirmed regulatory decision or structural event worth elevating above daily price movement. This is my record of technology and markets, not investment advice. Investment decisions and responsibility remain with each investor.