The Daily Spark

A new study from Ramp, linking observed AI spending to workforce records across 21,559 US firms, finds that companies adopting AI grow their headcount 10% over the two years following adoption.

These gains are entirely driven by high-intensity adopters with entry-level jobs rising by about 12%, suggesting that heavy AI investment is complementing workforce growth rather than replacing workers, see chart below and here.

For more discussion see also here and here.

Hyperscaler debt issuance continues to grow, reshaping dynamics across financial markets. This chart book tracks the AI trade across hyperscaler equity prices, credit spreads, CDS and earnings expectations, bringing together the signals that matter most as markets weigh the pace and payoff of AI investments. For more, see here.

Since congestion pricing launched in January 2025, about 500,000 vehicles enter Manhattan's Congestion Relief Zone on a typical weekday. The MTA estimates daily entries have fallen roughly 13% from their pre-toll baseline of about 640,000. Entries peak near 31,000 at 8 am, then hold a broad plateau through the afternoon rather than spiking again at evening rush, see chart below.

Consensus expects free cash flow for the hyperscalers to more than double over the coming years, see the first chart below.

But what if the payoff takes longer than consensus assumes? That question is particularly pressing given that token prices continue to decline and Chinese models are gaining ground, both in their share of the world's most-used models and in token usage, where they now lead their US counterparts among the top 20 models, see the second and third charts.

If Chinese models keep gaining and token prices keep falling, the hyperscaler cash flows expected may prove too optimistic.

What are the consequences if the AI payoff comes slower than expected in the first chart?

1) Cash flows and earnings disappoint: the projected free cash flow surge slips later while committed capex and heavy depreciation hit on schedule, squeezing margins and marking down the forecast in the first chart.

2) A Mag 7 sell-off that takes the market with it: equity prices built on a fast payoff re-rate, and because the Magnificent 7 now account for so much of the indices, the pain can't stay contained, it spreads to chips, power, data centers and the S&P 500 as a whole.

3) Balance sheets stretch and credit risk rises: with internal cash unable to cover spending, hyperscalers lean further on debt, raising leverage and inviting possible ratings downgrades if profits lag.

The bottom line is that AI has been the one thing holding up both the economy and markets, and with so much riding on so few names, a slower payoff wouldn't just be a sector problem, it would risk tipping the economy into recession and the S&P 500 into a correction.

When oil and yields stop moving together, it signals the Fed's problem has shifted from headline to core, see chart below.

With tariffs, a tight labor market and firm services prices still in play, the market expects core inflation to stay sticky even as energy costs fall.

The bottom line is that cheaper oil alone won't open the door to cuts, and the Fed is likely to stay on hold at its next meeting until it sees core inflation convincingly cooling.

Everyone talks about AI exposure in the labor market, but the reality is that there is no consensus about what this means.

This is illustrated by the five measures currently used in studies quantifying the impact of AI on the labor market:

1. One measure counts what people in each job actually do with Claude, using real chat logs to see which work tasks are being handed to AI.
2. Another does the same with Microsoft Copilot to see what those users lean on AI for.
3. A third skips usage entirely and has human experts judge which job skills AI is theoretically capable of doing, whether or not anyone is using it that way.
4. A fourth asks ChatGPT to grade its own usefulness on each task in a job, essentially letting AI predict where it could help.
5. A fifth scans millions of employer job postings for mentions of AI, treating demand for AI skills as a sign of how exposed a role is.

"What could AI do to this job?" and "What are workers actually using AI for?" are not the same question, and the theoretical measures (3, 4 and 5) run systematically higher than the usage-based ones (1 and 2) because they ignore whether adoption is even happening or worth the cost.

As a result, many occupations look highly exposed under one measure and barely touched under another.

What is most striking is that the five measures disagree most exactly where the stakes are highest, among the very jobs everyone wants to flag as at-risk, such as telemarketers, tax preparers and writers.

The bottom line is that when someone says a job is "highly exposed to AI," the honest first question is: Exposed by which measure, and measuring what? Until that is pinned down, the label "AI exposure" carries far less meaning than it appears to.

My colleague Sania Edlich and I will dig further into this in upcoming Sparks.

Japan's shrinking population has left 14% of its housing stock vacant, a share that has risen for decades and shows no sign of reversing, see chart below.

China's home prices have now been falling for four straight years, with both new and used home prices stuck in negative territory since 2022, see chart below.

Current ratings for software loans cluster heavily in the B2 and B3 buckets, but market-implied ratings spread loans toward both ends, a sign the market sees more credit dispersion in software leveraged loans than agency ratings currently reflect.

With more than 40% of Russell 2000 companies unprofitable, a higher-for-longer rate environment drives up debt servicing costs, threatening middle-market firms as interest costs take up a growing share of earnings, see chart below.