“The most important aspect of the Fed’s communication on Wednesday is going to be whether Powell characterizes policy as ‘in a good place’, as he did for the first several months of 2025 when the Fed was on hold, or if he repeats his description of policy being ‘modestly restrictive’ or ‘somewhat above neutral’. In the case of the latter, the door will remain open to further cuts in early 2026,” they told clients in a note seen by Fortune. “We do not expect that he will say policy rates are ‘in a good place’, but that will be the phrase to watch out for.” 

The context, of course, is that Powell is famously guided by the data. No matter what he says tomorrow, his decision in January will be based on incoming macroeconomic information between then and now.

And it’s not just Powell’s decision. He presides over an FOMC that is almost evenly divided against itself. Roughly half its members are wary of creating further new rounds of cheaper money that may be inflating a bubble in the stock market. The other half sees an economy on the verge of faltering, with rising unemployment, that needs easier money to avoid recession.

At the last Fed meeting, “there was a sharp division beneath the surface” of the FOMC, according to Macquarie’s David Doyle and Chinara Azizova. “Eight of 19 participants saw the policy rate in the 3.5 to 3.75% range [below where it is now at 3.75%]. This division is likely to remain apparent in the December update.”

“Given the likelihood for dissents, the growing differences in forward-looking policy projections are likely to be addressed. The chair is likely to emphasize that this is to be expected when the dual mandate is in tension due to rising unemployment and still elevated inflation,” they said.

Unemployment is trending upward, as shown in this chart from Macquarie:

At Goldman Sachs, chief U.S. economist David Mericle is also looking for signs of dissent. “There will most likely be two hawkish dissents in the statement, and we expect five participants to register soft dissents,” he told clients. “But we are not sure that all of this would add up to meaningful new information for the market.” 

Those dissents will hinge on how Fed members feel about the employment market, which seems to be weakening by the day. 

“It is not realistic to expect the FOMC to box itself in too much by signaling a very strong bias toward a pause in January because if the labor market is still actively softening at that point, a cut might be appropriate. In fact, participants will be even more uncertain than usual about what will be appropriate at the next meeting because we are now two employment reports behind schedule,” Mericle told clients.

Goldman estimates that U.S. job growth is below the “breakeven” rate vs job cuts:

Those missing employment reports—cancelled by the U.S. government shutdown—will leave the Fed more dependent than usual on anecdotal or imperfect private employment data. The Fed’s “beige book,” a periodic summary of quotes from American businesses, shows that employers are increasingly not creating new jobs. 

“Last week’s Beige Book suggested that labor demand is weakening via less hiring rather than layoffs – a fragile equilibrium in the labor market that will keep the Fed in a risk management mindset,” Oxford Economics analyst Michael Pearce.

Private employer data is equally gloomy, according to Bill Adams, chief economist for Comerica Bank in Dallas. ADP, Revelio Labs, and Challenger, Gray, & Christmas—three companies that compile private market jobs data—all saw payrolls falling in the last few months, he told Fortune. “Challenger, Gray, & Christmas reported employers announced plans for 71,000 job cuts in November, up 24% from the same month last year. They cited restructuring, market and economic conditions, and artificial intelligence as key reasons for layoff announcements,” he said. 

If the labor market continues to deteriorate, then it becomes less likely that Powell will say interest rates are “in a good place” and more likely that the Fed will deliver future cuts in 2026.

Here’s a snapshot of the markets ahead of the opening bell in New York this morning:

• S&P 500 futures were flat this morning. The last session closed down 0.35%. 
• STOXX Europe 600 were flat in early trading. 
• The U.K.’s FTSE 100 was flat in early trading. 
• Japan’s Nikkei 225 was up 0.14%. 
• China’s CSI 300 was down 0.51%.
• The South Korea KOSPI was down 0.27%.
• India’s NIFTY 50 was down 0.47%. 
• Bitcoin slid to $90K.

Answer: Look at the stack. In a true technology wave, the whole stack changes, not just one layer, says Kindred Ventures’ Steve Jang. And if you look at AI, he says, that’s exactly what’s going on: “Right now you’re seeing it all the way from chips all the way up through the application layer.”

Jang, Kindred’s founder and managing partner, was speaking at Fortune’s Brainstorm AI conference in San Francisco on Monday, on a panel to discuss how VCs are thinking about these bubbly times in the AI market. His point is that the angst over the AI bubble is kind of besides the point. What really matters is whether the underlying tech transformation is real or not. 

Sapphire Ventures partner Cathy Gao, who was also on the panel, said that valuations for some companies have clearly climbed far beyond any sort of fundamentals. But she also noted that the growth curves of certain companies right now “far outstrip the growth curves of companies we’ve ever seen before.”

To help sift through the hype in this environment and find the startups with real staying power, Gao said she uses a three-question test.

First, is the startup’s product a “feature” or a “workflow?” The stand-outs, she says, are “companies that are able to actually embed and fully overtake an existing workflow, while building significant switching costs.”

Second, is distribution built-in? People don’t want to learn how to use another tool, says Gao. The key qualities here are whether the tool is “integrated in usability and in all the other solutions that it needs to be in order for the workflow to be fully functional?”

And finally, does the company get stronger over time? This is called “compounding durability,” says Gao. “With every new user, does the solution get better, does it get cheaper, does it get faster?”

We’ll have more questions, and answers, at Brainstorm AI today. Watch the livestream here.

See you tomorrow,

Alexei Oreskovic
X:@lexnfx
Email:alexei.oreskovic@fortune.com
Submit a deal for the Term Sheet newsletter here.

Joey Abrams curated the deals section of today’s newsletter.Subscribe here.

Venture Deals

– UnconventionalAI, a San Francisco-based developer of chips and computer systems designed for AI, raised $475 million in seed funding. AndreessenHorowitz and Lightspeed led the round and were joined by Sequoia, LuxCapital, DCVC, FutureVentures, JeffBezos, and others.

– Airwallex, a San Francisco and Singapore-based payments and banking platform for businesses, raised $330 million in Series G funding. Addition led the round and was joined by T. Rowe Price, Activant, Lingotto, and RobinhoodVentures.

– BlueCurrent, a Hayward, Calif.-based developer of silicon solid-state batteries, raised $80 million in a Series D extension from Amazon, KochDisruptiveTechnologies, PiedmontCapital, RusheenCapitalPartners, and Allen& Company.

– Crown, a São Paulo, Brazil-based stablecoin issuer, raised $13.5 million in Series A funding. Paradigm led the round.

– ResembleAI, a Mountain View, Calif.-based security platform for enterprise AI, raised $13 million in funding from SonyInnovationFund, BerkeleyFrontierFund, ComcastVentures, CraftVentures, and others.

– Scowtt, a Seattle Wash.-based AI-powered advertising optimization platform, raised $12 million in Series A funding. InspiredCapital led the round and was joined by LiveRampVentures, Angeles Investors, and AngelesVentures.

– Equixly, a Verona, Italy-based agentic AI platform designed for API security testing, raised €10 million ($11.6 million) in Series A funding. 33NVentures led the round and was joined by existing investors.

Private Equity

– ContextLogic agreed to acquire US Salt Parent Holdings, a Watkins Glen, N.Y.-based producer of evaporated salt products, from private equity funds managed by EmeraldLakeCapitalManagement for $907.5 million. 

Exits

– BerkshirePartners agreed to acquire UnitedFlowTechnologies, an Irving, Texas-based provider of process and equipment solutions for water and wastewater systems, from H.I.G.Capital. Financial terms were not disclosed.

Even after adjusting for inflation, big tech companies are issuing more bonds than during the late 1990s. And the companies aren’t just refinancing existing debt—they’re taking on additional debt.

“While the increasingly aggressive (and creative) borrowing by AI companies won’t be their downfall, if they do fall short of investors’ expectations and their stock prices suffer, their debts could quickly become a problem,” Zandi wrote. 

“Borrowing by AI companies should be on the radar screen as a mounting potential threat to the financial system and broader economy.”

The 10 largest AI companies, including Meta, Amazon, Nvidia and Alphabet, will issue more than $120 billion this year, Zandi said in a LinkedIn analysis last week.

And this time is different from dot-com era debt issuance, as internet companies back then didn’t have a lot of debt, he pointed out. Instead, they were funded by stocks and venture capital.

“That’s not the case with the AI boom,” Zandi added.

Even though hyperscalers like Amazon, Google, Meta, and Microsoft could pay for the AI buildout with their profits, bond issuance is the “cheapest and cleanest” way to finance an infrastructure buildout of this scale, which will likely last more than a decade and be worth trillions of dollars, Shay Boloor, chief market strategist at Futurum Equities, told Fortune.

“These companies are a lot more comfortable issuing 10- to 40-year papers, for example, at very low spreads, because the market now views them as quasi-utility names—because they’re building all this infrastructure—not just a pure tech company anymore,” Boloor said.

He added that in the previous six months, tech companies have shown “proof in the pudding” that future demand for AI is booming.

Despite AI bubble concerns, Nvidia delivered a strong earnings report for its third quarter last month, saying its AI data center revenue increased by 66% from last year. 

Still, critics warn that the buildout may not keep up with how rapidly AI is developing.

Computer hardware, which makes up most AI data centers’ cost, may be more susceptible to becoming obsolete and replaced by more advanced technology during the AI boom as opposed to wireless and internet buildouts, much of which still runs today, George Calhoun, professor and director of the Hanlon Financial Systems Center at Stevens Institute of Technology, told Fortune.

“The cycle of innovation in the chip industry is much faster than for wireless technology or fiber optics,” he said explained. “There is a real risk that much of that hardware may become competitively disadvantaged by newer technologies in a much shorter timeframe,” before being fully paid off.

At the same time, big players in the AI boom—namely OpenAI—do not have the profits currently to cushion their massive investments at the moment, increasing their risk, Calhoun said.

“If OpenAI fails, the snowball effect of that is gonna be substantial,” Futuruum Equities’ Boloor said. Though larger tech companies won’t likely be impacted much by a potential OpenAI bust, companies that largely rely on its business like Oracle could, he added.

Still, Boloor is optimistic about the AI buildout, saying the main bottleneck for its success is U.S. energy capacity.

“I think that the risk is that trillions of dollars of AI capacity gets built faster than the North American grid can support it, which could slow realization,” he warned.