The federal government is still digging out from weeks of shutdown, and the usual flow of economic data remains disrupted. Some private data has certainly filled the void, but, overall, when data fails us, we can rely on well-sourced anecdotes. The stories told by businesses in the Federal Reserve’s most recent Beige Book (formally known as “Summary of Commentary on Current Economic Conditions by Federal Reserve District”) are the closest thing we have to a real-time, on-the-ground survey of how the economy is doing.
The latest edition of the Beige Book describes an economy that is still expanding, but not comfortably so. Absent the lift from the artificial intelligence boom, the U.S. economy would be dangerously close to falling flat, or worse. Do we need to worry about the AI lift being taken away? Or should we rejoice that it’s here?
Big picture: Slow grind, worse tone
Of the dozen Federal Reserve districts, two report modest declines, one modest growth, and the rest somewhere in between. Overall, consumer spending continues to slip, while higher-end retail remains solid.
Across districts, low-income consumers have been pulling back all year; now, contacts say the same behavior is spreading to the middle class. One Philadelphia retailer noted that middle-income households are starting to look a lot more like low-income households at the checkout line: trading down, skipping extras, and leaning harder on discounts.
Meanwhile, the high-end consumer is still living in a different world. New York reports that luxury retail and the Hamptons housing market remain strong, with wealthy buyers paying cash or using alternative financing while more leveraged buyers sit on the sidelines. Chicago notes that retail corridors catering to higher-earning consumers are thriving, while value retailers also see solid traffic from everyone else trading down.
Residential construction is off in some districts and flat in others. Agriculture and energy are described as “largely stable,” though low prices are weighing on margins.
Employment has declined, with many firms using hiring freezes and attrition rather than outright layoffs. A few contacts explicitly said that AI deployments allowed them to skip hiring entry-level workers or to forgo refilling vacancies.
Prices are rising at a moderate pace, but not in a way that feels comfortable to businesses. Tariffs, insurance, healthcare, utilities, and technology all show up as cost pressures. Margin compression is evident across multiple districts, albeit from high levels.
Manufacturing and capex: AI keeps the lights on
Manufacturing has ticked up “somewhat,” helped by AI-related investment and data center construction, but nonfinancial services are mostly flat to down. Several districts report modest increases in factory output and new orders, with strength in capital goods and anything tied to automation. Chicago notes a pickup in machinery sales and says firms are increasing capex, especially on automation that does not require corresponding increases in headcount.
Cleveland highlights a very specific growth driver: AI data centers. Producers of components and equipment for data centers say demand is “rapidly expanding.” At the same time, one contact described a “collective holding of breath” because AI construction is now the main thing keeping their order books full while other segments remain soft. San Francisco echoes that pattern on the construction side: New commercial buildings are “concentrated” in data centers, with much less activity elsewhere. Without that AI capex, the factory side of the economy would look a lot weaker.
Bottom line: Not a recession, but more fragile
The November Beige Book does not scream recession. It does, however, describe an economy that is more fragile than the GDP print suggests:
- Consumers outside the top income brackets are trading down, pulling back on travel and “extras,” and leaning harder on safety net programs.
- Manufacturers are increasingly dependent on AI-related demand, even as other end markets weaken.
- Credit quality is still OK, but clearly deteriorating at the margin.
If the AI boom weren’t happening, I suspect we would be heading toward a growth scare, or something worse.
So how much has AI “juiced” the economy since ChatGPT introduced generative AI to the mainstream in late 2022? Researchers at the St. Louis Fed estimate that generative AI tools may have raised U.S. labor productivity by up to about 1.3 percentage points since late 2022. Over time, GDP tracks the growth rate of prime-age employment plus the productivity growth rate.
In the first half of 2025, AI-related capex contributed 1.1 percent to GDP growth, outpacing the U.S. consumer as an engine of expansion, according to JP Morgan. And that is not even including the 1.3 percent from the productivity boom.
AI and the stock market: This cycle vs. Netscape’s
ChatGPT exploded into the zeitgeist three years ago, on November 30, 2022. Investors naturally want to know whether the last three years of AI-driven market gains are a déjà vu of the late 1990s dot-com bubble or something more sustainable.
Let’s use the Nasdaq 100 as a stand-in for the “AI winners” today, just as it captured the core of the 1990s tech boom.
- The last three years (post-ChatGPT era)
From the start of 2023 through late 2025, the Nasdaq 100 has posted calendar year total returns of about plus 54 percent (2023), plus 25 percent (2024), and plus 21 percent year to date in 2025. It was up 111.5 percent over the three years following November 30, 2022, roughly 28 percent compounded annually. For a three-year stretch, that is fantastic. But it is not unprecedented, and it is still below the truly manic phase of the dot-com era. - The three years after Netscape’s IPO (1995–1997)
Netscape went public in August 1995, after being released on December 19, 1994. The Nasdaq 100 returned roughly 122.4 percent after its launch, an approximately 30.5 percent annualized rate.
Bespoke compiled the three-year returns following the previous major technology releases over the last 50 years.
By comparison, today’s AI era run-up is similar but still milder than the three-year rocket ride that preceded the 1999–2000 blow-off, when it soared 104 percent in less than 15 months. The stock market in the 1990s became a bubble not because it doubled, but because it doubled twice, and the pace of stock price growth outpaced earnings growth.
- What happened next in the 1990s?
The following five years after that (1999–2003) are a useful cautionary tale:
-
- 1999 alone saw the Nasdaq 100 more than double to its March 10, 2022, peak, with a 104 percent gain.
- Then the bubble burst. From the March 2000 peak to the 2002 low, the index lost roughly 80 percent of its value.
- Even after a 49 percent rebound in 2003, an investor who bought the Nasdaq 100 at the start of 1999 was still down about 20 percent five years later.
The lesson is not that this cycle must end the same way. It is that three great years in growth stocks do not automatically mean “bubble.” Right now, the AI era looks more like the 1996–1998 track (strong but still grounded in earnings, capex, and real productivity) than the 1999–2000 blow-off that defined the 1990s bubble. The Beige Book’s anecdotes about real companies spending real money on data centers and automation support that view, even if the concentration leaves the economy vulnerable.
Allen Harris is an owner of Berkshire Money Management in Great Barrington and Dalton, managing more than $1 billion of investments. Unless specifically identified as original research or data gathering, some or all of the data cited is attributable to third-party sources. Unless stated otherwise, any mention of specific securities or investments is for illustrative purposes only. Advisor’s clients may or may not hold the securities discussed in their portfolios. Advisor makes no representation that any of the securities discussed have been or will be profitable. Full disclosures here. Direct inquiries to Allen at AHarris@BerkshireMM.com.
