AI Investment Boom: Tech Revolution or Bubble?

_{Photo by Michael Guesev, The Principal Madrid, Madrid 2024}

Introduction: The AI Surge in a Liquidity-Driven Market

U.S. equites over the last years have been led by a boom in AI tech stocks. This rally has pushed market concentration and valuations to potentially concerning levels. But to understand if we should in fact be concerned it would be prudent to compare the current dynamic to the closest historic precedent we have in the late 90s dot-com boom. Tech now makes up 34% of S&P 500 market capitalization, surpassing the 33% tech weight at the peak of the 2000 dot-com bubble. Of the top 10 largest U.S. companies, 8 are tech/AI-centric “Magnificent 7” names (Apple, Microsoft, Alphabet, Amazon, Nvidia, Tesla, Broadcom, Meta, plus non-tech Berkshire/JPM) where together these 10 giants account for 40% of the index’s value (compared to 25% for the top 10 in 1999). Meanwhile, forward price-earnings (P/E) multiples for the leaders while stretched above historical norms, remain below late-90s levels. The S&P 500 as a whole trades near 22× forward earnings in October 2025, compared to 24.5× at the peak in 2000, while the NASDAQ 100 is around 28–29× compared to 70× in 2000. So while concentration is indeed high compared to the late 90s, equity valuations do not appear quite as frothy as the dot-com extremes.

But what about the fundamentals? Earnings have significantly improved between 2023 and 2025, but as mentioned above are highly concentrated. S&P 500 aggregate earnings per share (EPS) are on track to rise about 10–11% in 2025 after a flat 2023, partly thanks to post-pandemic economic resilience and corporate tax cuts. However, that growth is disproportionately driven by Big Tech. In Q2 2025, for example, the six mega-cap tech firms contributed nearly half of the S&P’s 11.7% year-on-year EPS growth. Sectors tied to digital saw earnings explode 20–45% year-on-year. In contrast, more than a third of the remaining sectors lagged with single-digit or negative profit growth. The market’s earnings engine is as top-heavy as its valuations; still, these outsized profit gains in tech provide a fundamental underpinning that differentiates the AI era from the late 90s to today. Market frontrunners are delivering significant earnings and growth, compared to many dot-com stars that had little or no profits during the boom.

The macro-liquidity backdrop has also been extraordinarily supportive, further enabling high valuations. A confluence of expansionary fiscal policy, falling interest rates, and abundant cash has created a self-reinforcing liquidity regime for stocks (see my previous post on equity outlook: https://therationalmargin.com/us-equity-outlook-into-q4-2025/). The Fed has pivoted from tightening to rate cuts in 2025, bringing the fed funds rate down to 4% by Q4 and signaling more cuts to come. Lower policy rates are directly boosting equity appeal by reducing the return on cash and compressing discount rates in valuation models. Every hint of Fed easing has sent stocks to new highs in 2025 as falling yields make future earnings more valuable. By late-2025, the yield on 3-month T-bills has declined to the mid-4% range, so that the S&P 500’s earnings yield (4.0%–4.5%) has finally matched risk-free rates, an essential normalization after an unusual period earlier in the year when cash yields exceeded the S&P’s earnings yield (~3.7%), implying a negative equity risk premium. This normalization is crucial and has further raised the attractiveness of stocks. On the fiscal side, massive front-loaded tax cuts in 2025 and Treasury liquidity maneuvers (issuing short-term T-bills that drew cash out of the Fed’s overnight facility) have flooded markets with liquidity. Much of the $1 trillion+ in new 2025 T-bill issuance was absorbed by money market funds rotating out of the Fed’s reverse repo facility, effectively “unsterilizing” idle cash into private hands. This, along with a weaker U.S. dollar boosting foreign investment and corporate earnings, has created increasing tailwinds for equities overall. The AI-fueled equity surge has coincided with a goldilocks macro environment: moderating inflation, accommodative monetary policy, and ample global liquidity. All of which have helped justify the high valuations and delay any reckoning with gravity.

Yet the very combination of lofty valuations, market concentration, and abundant liquidity still raises the question: To what extent are we in a bubble and are we living in a replay of the late-1990s tech boom and crash, or is this a fundamentally justified tech revolution investment boom? The remainder of this article examines that question, comparing the current AI investment cycle to the dot-com era, analyzing cross-market signals of froth vs. fundamentals, and considering the potential macro impact if today’s AI optimism were to deflate.

AI 2023–2025 vs. Dot-Com 1998–2001: Parallels and Contrasts

Valuation Dynamics: Both the dot-com boom of before and AI-driven boom of today feature soaring equity valuations anchored in transformative technology narratives. During the 1998–2000 run-up, equity indices (especially the NASDAQ) climbed rapidly. The NASDAQ rose 250% over two years before finally crashing after valuations broke records with the S&P 500’s P/E above 30 and the NASDAQ 100’s forward P/E exceeding 70 by 2000. Many individual dot-com stocks traded at triple-digit multiples (or had no earnings at all), sustained purely on optimism and hype. By comparison, the 2023–2025 AI rally, while dramatic, is demonstrating more sober valuations on average. The S&P 500 forward P/E of 22× and the tech sector’s 29.5× in late-2025 are elevated but below the levels reached in 2000. Today’s market leaders are also mainly established firms (Apple, Google, Microsoft) with substantial earnings and cash flow, which creates a fundaments floor under valuations. Nvidia, for example, has genuine and substantial profits and grew earnings by 170% in 2024 alone. In 2000, by contrast, many highfliers (Pets.com, Webvan) had no profits or revenue, making their valuations purely speculative. Granted, some pockets of the current boom do resemble the late-90s mania with smaller AI software firms and chipmakers trading at 50–100× earnings, and some newly IPO’d tech firms in 2023–24 popping 100%+ out of the gate despite poor financials. Overall though, the AI boom’s valuation profile, while a bit stretched, is much less egregious than the dot-com bubble, and is led by gigantic incumbents with strong profitability and cash flows.

Capital Misallocation: A classic risk in investment mania is capital rushing into projects that won’t deliver commensurate returns in the near term (if ever). The dot-com era famously overbuilt infrastructure and startups: telecom companies laid enormous fiber-optic cable networks and expanded data centers on the assumption of infinite internet traffic growth, leading to a glut of capacity by 2001. Venture capital and IPO money funded hundreds of web startups (online pet food, grocery delivery, etc.) that quickly failed. This overinvestment cycle resulted in short-run overcapacity, e.g., vast stretches of dark fiber and idle network gear, which took years for demand to catch up. Similarly, the AI boom has triggered a massive surge in investment in computing infrastructure. Analysts estimate that realizing the potential of generative AI will require roughly $2.9 trillion in global data center spending through 2028, including $1.6T on hardware (chips/servers) and $1.3T on related infrastructure. By 2028, annual AI-related capex could approach $900B – an investment level on par with the entire S&P 500’s capital spending in 2022. Already, real outlays on information-processing equipment and software have spiked since late 2022, growing at a trajectory comparable to the late-90s boom in communications equipment. Data center construction is up sharply, as real investment in data center infrastructure has more than doubled since 2022. The risk is that this could lead to a near-term glut of AI capacity. If AI adoption or software progress lags behind the hardware buildout. In that case, companies may find themselves with idle GPU clusters and underutilized cloud infrastructure (i.e., telecom firms with excess fiber in 2001). So investors and firms are essentially betting that future AI demand will justify today’s significant capex. If that demand “takes longer to materialize or is overestimated, a scenario the dot-com bust exemplified, then 2025’s AI investment may look like 1999’s telecom bubble, with a crash and retrenchment.

On the other hand, some overbuilt capital can have long-run benefits: much like the dot-com bubble built fiber networks that eventually facilitated the broadband revolution, the current boom’s investments in cloud infrastructure and AI research could yield lasting productivity gains even if many individual investments fail to pay off. The challenge for investors is that the timing and distribution of returns may be highly uneven, as was the case after 2000, when Internet usage continued to rise while many early investors never recovered their capital.

Funding Structure Then vs. Now: The financing environment during the AI boom differs from that of the dot-com era in notable ways. Late-90s tech investment was often funded through public equity, with a flood of IPOs and secondary offerings, allowing pre-revenue companies to tap retail money. From 1998 to 2000, over 1,500 companies went public in the U.S., many in tech, raising capital on expectations. Venture capital was necessary in seeding companies, but the IPO market provided the big payoffs that fueled retail speculation. By contrast, the 2023–2025 AI wave has seen comparatively few IPOs, as the IPO market has been subdued since the 2022 downturn. Instead, funding has come from private capital and corporate balance sheets. Venture capital and private equity have poured significant funds into AI startups – in just Q1 2025, AI startups raised $73.1 billion globally, ~58% of all VC funding for that quarter. Mega-rounds are common: for example, OpenAI (creator of ChatGPT) reportedly secured $40 billion in a capital raising in 2025, drawing in sovereign wealth funds and tech giants. Rather than IPO, many AI startups are staying private longer or being acquired by larger tech firms. Large established companies (like Alphabet, Microsoft, Meta) are themselves investing heavily in AI R&D and data centers, primarily financed by internal cash flows or corporate bond issuance at relatively low yields. The result is a bubble (if it is one) being funded by big balance sheets and institutional money more than by retail frenzy. This structure could mean that when the music stops, the losses concentrate in VC portfolios, Softbank-style vehicles, and mega-cap market caps, rather than a cascade of penny-stock busts hitting retail investors. It also implies less immediate financial contagion: banks are not heavily exposed (no widespread bank lending to AI startups analogous to subprime lending). Leverage in the system is lower than in many past bubbles and is similar to that of the dot-com boom, which didn’t involve much bank leverage either. One reason the bust, while painful for investors, did not trigger a banking crisis. Similarly, an AI bust might be mostly an equity market event. However, one cautionary parallel is the role of “fear of missing out” (FOMO) in both eras. In 1999, professional investors felt compelled to buy into dot-coms to keep up with benchmarks, and today, many institutional investors feel pressure not to miss the AI trade. Psychology can induce herding and override prudent risk management, regardless of funding structures.

Post-90s Bubble Economic Consequences: The aftermath of the dot-com bubble offers a guide and warning of what could happen if the AI boom turns to bust. The dot-com collapse in 2000–2001 erased around $5 trillion in market value, as the NASDAQ plunged 75% and the S&P 500 fell 49% from peak to trough. The real economy impact was a mild U.S. recession in 2001: GDP dipped as business investment contracted sharply (real private investment in tech equipment fell by double digits), and unemployment rose from 4% to 6% by 2003. Certain regions (Silicon Valley, etc.) and sectors (telecom, IT hardware) were hit disproportionately with massive layoffs and bankruptcies. However, the broader economy was cushioned by the Fed’s aggressive rate cuts (from 6.5% in 2000 to 1.75% by the end of 2001) and a housing market uptick. The dot-com bust, while dramatic in equity markets, was not a systemic financial crisis. Still, it took 14 years for the NASDAQ to regain its 2000 peak, and many once-dominant firms never recovered (e.g., Sun Microsystems, Lucent, and numerous dot-coms disappeared). A similar narrative could unfold if the AI boom deflates: we could expect a significant equity correction (potentially 30–50% drops in the AI-heavy NASDAQ or in key AI stocks). The hit to wealth would be substantial: today’s top 10 are worth $22 trillion, so a 30% drop could erase $6–7 trillion, denting consumer confidence and spending via wealth effects. More critically, a collapse in AI enthusiasm might lead to a pullback in corporate investment: companies could scale down AI projects, and the promised $900B/year global AI capex by 2028 might never fully materialize, at least not on the initially expected timeline. This would particularly affect the U.S. tech sector, potentially tipping the U.S. into a mild growth recession if not offset. One positive difference in a hypothetical AI bust scenario is that the core financial system would likely remain standing, as banks aren’t heavily exposed to equity risk, and household balance sheets (aside from 401(k)s) aren’t as levered to stocks as they were to housing in 2008.

Any deflation in the AI bubble would entail painful repricing and temporary economic losses in output. Still, the dot-com bust showed that an innovation-driven bubble can burst without long-term ruin: the groundwork laid by the internet boom eventually yielded the digital economy we enjoy today, even though investors in 2000 overpaid. Likewise, even if many AI investments misfire, the technology itself could continue advancing and contributing to growth, albeit on a more realistic timetable. The key to macroeconomic stability is whether the transition is relatively orderly (a correction and a mild recession) or whether it sparks a larger crisis of confidence.

Bubble vs. Fundamentals

Are current asset prices and investment flows “bubbling” beyond fundamental value, or do they reflect rational expectations of future earnings? We next examine key signals across public equities, private capital, and real investment, as well as funding structures and systemic risk considerations invoking classic bubble diagnostics from Kindleberger and Minsky, Blanchard and Watson, Brunnermeier and Oehmke, and others to interpret the evidence.

Public equity markets: Several indicators in equities are consistent with a late-stage, euphoric market (Minsky’s “mania” phase), driven by narratives rather than current fundamentals alone. Valuations and concentration were discussed above – e.g., the top-heavy market cap and thin risk premia. Historically, Kindleberger’s bubble anatomy starts with a displacement (a new technology) that, with ample liquidity, leads to a boom, then euphoria and FOMO. Prices have been rising faster than fundamental earnings. For example, Nvidia’s stock rose over 200% in 2023, far outpacing the upgrade in its earnings outlook (even though earnings jumped impressively, the valuation overshot those higher earnings). “A rational bubble is present whenever an asset price deviates progressively more quickly from the path dictated by its economic fundamentals.” By that definition, one could argue that a rational bubble component is at play, and investors may be buying AI-heavy stocks not just for their current cash flows, but also in anticipation of selling at an even higher price later. This aligns with the Blanchard–Watson model of rational expectations bubbles, wherein as long as enough investors believe the hype and liquidity is plentiful, prices can depart from intrinsic value and keep rising. Indeed, surveys and fund flow data in 2025 show investors crowding into tech despite stretched valuations, suggesting a sentiment-driven bid.

On the other hand, a fundamental element is not entirely absent: AI leaders are delivering earnings growth and market gains, and many expect AI to boost long-term growth and profits across the economy. Equity analysts have been raising 5-year growth forecasts for the tech sector from AI opportunities. If these forecasts come true, some of today’s multiples might prove reasonable, confronting what Gürkaynak (2008) highlighted: it is notoriously tricky to empirically distinguish a bubble from rationally changing fundamentals in real time. For every metric that looks overvalued, bulls have a story – e.g., “AI will lead to new products and efficiencies not yet in earnings, justifying the valuation.” Gürkaynak concludes that no econometric test can definitively prove a bubble’s existence ex ante, as one can always hypothesize some unobserved fundamental (future growth, options value of innovation, etc.) to justify the price. The caution applies today: valuations appear out of line with current earnings, but perhaps not with potential earnings a few years out if AI truly transforms the economy. The key market signal to watch is whether price increases start to decouple even from optimistic fundamental benchmarks, for example, if prices keep rising without further upgrades in earnings prospects, or if we see a proliferation of new stock issuance/SPACs trying to capitalize on AI hype (which we saw in 1999 and again briefly in the 2020 SPAC bubble). Thus far, equity issuance has been relatively contained (a point in favor of fundamentals). To summarize, public equities are sending mixed signals: in valuation and behavior, there are bubble-like aspects, but earnings and macro conditions provide a rationalizing narrative. As Kindleberger wrote, during booms, “there is always a plausible story,” and AI is indeed a plausible revolutionary force, which makes this boom particularly tricky to classify as a pure bubble.

Private Capital and Venture Markets: In private markets, the evidence of bubble-like excess is more convincing. Venture capital flows into AI have been astounding. By 2025, virtually every VC firm and growth investor has refocused on AI startups, driving mega funding rounds at lightning speed. PitchBook data show 57.9% of all global VC funding in Q1 2025 went into AI deals. This kind of single-theme concentration is historically unusual (even at the height of crypto hype or prior tech waves, no sector has ever accounted for more than half of VC funding). The valuation anecdotes coming out of VC panels are eyebrow-raising: Investors report that “any startup with an ‘AI’ label will be valued right up there at huge multiples of whatever small revenue it has”. Bryan Yeo, CIO at Singapore’s sovereign fund GIC, noted valuations of $400 million to $1.2 billion per employee for some early-stage AI ventures, calling that “breathtaking” and indicative of FOMO-driven excess. This mirrors the late-90s, when adding “.com” to a business model or prospectus could instantaneously inflate valuation. According to Minsky’s framework, this is the speculative and Ponzi finance stage, as new financing is secured not on the cash flow of projects (many AI startups burn cash with no clear path to profit) but on the expectation that someone else will provide more capital later. As long as venture investors believe they can exit to someone (greater fool theory), be it via a high-profile IPO or an acquisition by a tech giant flush with cash, the game continues. But if the expected exits don’t materialize (IPO window shuts, big firms tighten M&A), the funding chain collapses. Kindleberger noted that toward the end of manias, “the supply of greater fools may run out.” There are hints that this is beginning. By late 2025, some VC insiders caution that many AI startups are overvalued and that “market expectations could be way ahead of what the technology can deliver”. Still, until there is a specific trigger (a major flop, a shift in liquidity), the momentum in private funding can persist. One structural positive is that private capital bubbles don’t immediately hurt the public, as losses will be borne by VC and PE funds (and their limited partners) rather than retail investors. However, the scope of potential capital misallocation with tens of billions into possibly duplicative AI models and speculative use cases could mean a lot of deadweight investment if the bubble pops. The private market seems to be flashing bright red bubble signals: unprecedented funding share, overvaluation, and reliance on hype rather than financials. All of which resembles a “big market delusion” (a term used by researchers to describe when investors all bet on a vast potential market without thinking about competition, leading to overvaluation of an entire cohort). Unless the vast promised market for AI truly yields enormous revenues fast, many of these investments will likely be impaired.

Infrastructure and Real Economy Investment: As noted earlier, data confirms an exceptional boom in AI-related capex. A Richmond Fed analysis compared the current surge in information processing equipment/software investment with the 1990s telecom buildout: the growth rates are similar, but the absolute level of investment in AI-related categories is even higher now (in inflation-adjusted terms) than telecom’s peak in the late ’90s. Moreover, AI data center construction has taken off explosively, whereas ’90s telecom construction boomed late and peaked lower. Real private investment in data centers has roughly tripled over the past three years. Such rapid capacity expansion can be interpreted through two lenses:

• Fundamentals-driven: Perhaps AI requires rapid infrastructure scaling (for training large AI models, deploying cloud AI services, etc.). If so, the investment could be meeting a real need, and current spending will yield high returns (productivity gains, new services) in the coming years. Under this view, what appears to be overinvestment might not be “malinvestment” but rather an efficient response to a transformative general-purpose technology. Historically, electrification and railroads also saw investment booms; some ended in busts, but those that were truly paradigm-shifting ultimately justified huge capital outlays (though often not for the first movers).
• Bubble-driven: Alternatively, companies may be engaging in herding and overestimating short-term demand, a key ingredient in bubbles, per Kindleberger. For example, suppose every primary cloud provider (Amazon, Google, Microsoft, Oracle) is racing to build AI data centers to compete. In that case, they might collectively build more than the market really needs by 2026. The Richmond Fed piece explicitly warns that the late-90s telecom investment “led to tremendous, short-run overcapacity and a bust,” though the capacity was eventually put to use. The same could happen with AI: a few years of investment overshoot, a painful shakeout (data center projects canceled, GPU prices collapsing from oversupply, etc.), and then consolidation until demand catches up. The presence of rapid price inflation in AI hardware (GPUs) and shortages in 2023–24 might have lulled firms into extrapolating infinite demand, but if usage or AI adoption plateaus, that pricing power could evaporate.

Current signals within the real economy include extremely tight labor markets for AI talent and semiconductor engineers (wage and hiring wars) and rising prices for AI-related inputs (advanced chips). These are symptomatic of a boom phase. If we start to see those trends reverse (e.g., reports of cancelled chip orders or layoffs in AI teams), it could indicate the boom is tipping into distress. As of late 2025, though, capex plans remain ambitious. A sign that confidence in AI’s promise remains high for now. In Kindleberger-Minsky terms, the “displacement” (AI breakthroughs) has clearly led to a capital spending boom; whether it turns into a crisis depends on whether expectations were too far ahead of reality.

Funding structures and systemic risk considerations: According to Brunnermeier and Oehmke’s framework, one must assess whether a potential bubble is likely to trigger a financial crisis via leverage and maturity mismatches. In the AI boom, systemic financial risks appear relatively contained because, as noted, it is mainly equity-financed. We do not see banks making many AI loans, or consumers taking on debt to invest in AI (unlike the housing bubble, when mortgages were the transmission mechanism). The stock market wealth being created (and potentially destroyed) is mainly held by diversified investors and funds, not on bank balance sheets. That said, a second-order risk is that a sharp market downturn could indirectly cause tightening: e.g., if VC funds get crushed, they might pull back funding to other startups, some of which default or lay off employees, affecting local economies and specific loan portfolios (e.g. Silicon Valley Bank’s experience in 2022–2023 showed how concentrated tech exposure can bite a lender).

Additionally, market-based financing could amplify volatility. If AI-thematic exchange-traded funds (ETFs) or leveraged products face redemptions, they might need to dump shares, causing fire-sale dynamics quickly. But these are likely to be contained episodes, not broad banking failures. From a systemic viewpoint, the AI investment surge might be called a “speculative boom” but not yet a credit bubble, which means its unwinding, while painful for equity investors, should be less damaging to the core financial system.

Sampling on academic bubble tests or indicators, a classic test is looking for explosiveness in price time series beyond what fundamentals can explain. Some economists attempt econophysics-style bubble detection (e.g., looking for accelerating logarithmic price increases). Tech stocks have exhibited super-exponential price rises in early 2023 (Nvidia’s chart almost went vertical mid-2023), which may be a danger sign. Another indicator is IPO volume and first-day returns. In bubbles, those tend to spike (as in 1999). But in late 2023/2024, IPOs like ARM, Birkenstock, and a few others had mixed receptions, not mania, suggesting the public market bubble may not be as intense as the one in private markets. Kindleberger’s anecdotal signs of late-stage mania also include fraud and scandal (e.g., the 1720 South Sea Bubble or the 2008 subprime crisis). We have not yet seen major fraud in AI, although one could argue that some startups might be overhyping capabilities (Theranos-like outcomes are possible if scrutiny increases). The absence of fraud at scale suggests we are perhaps not yet at a final “panic” stage, cold comfort though, as frauds are often only revealed after a crash (as with Enron/WorldCom post-2000).

Bottom Line: Across public and private markets and real investment, the patterns are consistent with a classic boom fueled by a compelling narrative and easy money. The fundamentals of AI are real, but the market’s pricing of those fundamentals appears to be front-loaded and excessive, which is the essence of a bubble. To quote a Research Affiliates commentary, “today’s AI darlings must exceed already lofty expectations to beat the market… If cracks form in the narrative – if the fundamentals fail to keep pace with investors’ fanciful projections – the broader story may crumble”. We are in a period when the story (AI will change everything) is widely believed and partially true. Still, if reality underwhelms over the next couple of years, a lot of paper wealth could evaporate very quickly.

Impacts of a Potential AI Valuation Reset

Given signs of some overexuberance, it is prudent to consider a scenario in which AI-related valuations undergo a material reset, whether through a sharp correction or a gradual deflation, and to analyze the macroeconomic and financial market implications. What if the “AI bubble” pops in 2026? Here, I outline the expected effects on GDP, investment, labor markets, credit, and overall financial conditions, drawing parallels to the dot-com aftermath while accounting for today’s policy context.

Equity Market and Wealth Effects: A collapse in AI-driven stock valuations would directly hit the broader stock indices due to their heavy weight. If the top 10 market-cap companies (40% of the S&P) fell, say, 30–50% in value, the S&P 500 could swiftly drop on the order of 15–20% or more, even if the other 490 stocks were flat. In practice, a severe tech bear market would likely drag down the entire market via sentiment contagion, so a post-bubble bear market (decline >20%) is a reasonable baseline. This kind of equity correction, while significant, is not unprecedented and might resemble the 2000–2002 bear market. The loss of household wealth from a 20% broad-market decline (which could amount to a $10 trillion loss in market cap) would feed through to consumer spending with a lag. Standard estimates are that consumers cut spending by about 3–5 cents per dollar of wealth lost. So a $10T drop in equity wealth might trim consumer spending by $300–500 billion (1–2% of GDP) over a couple of years. In the year immediately following the crash, U.S. GDP growth could be reduced by perhaps 0.5–1 percentage points due to the wealth effect, other things equal. This is a rough magnitude based on past episodes (after 2000, consumption spending growth slowed but did not collapse, partly because housing wealth was rising at the time; in a 2026 scenario, if other sectors are neutral, the hit would be noticeable but not catastrophic).

Business Investment and Sectoral Impacts: The most direct macro hit would likely come from a pullback in corporate investment, especially in technology and infrastructure. If boards and investors turn more skeptical on AI returns, we could see a rapid downscaling of capex plans. The astronomical data center spending projections (nearly $3T globally to 2028) would be revised down. Companies like cloud providers might postpone or cancel server orders; startups would slash R&D spending to conserve cash. Equipment orders and industrial production related to tech would weaken. In the early 2000s, real investment in Information Processing Equipment contracted for multiple quarters, a repeat of which could knock out a key driver of late-2020s growth. Quantitatively, if tech capex was contributing +0.5% to U.S. GDP growth in 2025 (which is likely currently is), its sudden stop could swing that to a –0.5% drag in 2026. Regions tied to the tech supply chain (semiconductor hubs in East Asia, software hubs in the U.S.) would feel it too. However, other sectors might offset somewhat – e.g., if AI doesn’t require as much capital, resources might shift to other investments, or consumer spending could get a boost from lower tech prices. Net net, a pronounced AI investment retrenchment could be a material headwind to U.S. GDP for a couple of years, potentially meaning the difference between 2% growth and a mild recession.

Specific industries would be particularly hard-hit. Semiconductor manufacturers (especially those making high-end AI chips) could see a sharp downturn in orders reminiscent of how Cisco and other network gear makers saw orders evaporate in 2001, leaving excess inventory. Enterprise software and cloud services geared toward AI might also slow down, affecting firms such as enterprise IT providers. On the flip side, industries that were facing high input costs or competition for talent from the AI boom (like non-tech sectors losing workers to tech) might actually get relief if the boom deflates. E.g., a non-tech firm finds it easier to hire engineers once the Googles and OpenAIs are in hiring freeze mode.

Labor Market: While the tech sector accounts for a relatively small share of total employment, it punches above its weight in wage bills and economic dynamism. A severe correction in the AI/tech sector would likely lead to hiring freezes and layoffs at tech companies and startups. We’ve already seen a mini-version in late 2022 when many big tech firms cut 5–10% of staff to trim “pandemic excess,” a larger shock would yield deeper cuts. In a burst-bubble scenario, hundreds of AI startups could shut down, displacing their employees. Large firms might cancel ambitious “moonshot” AI projects, affecting the teams of engineers working on them. However, laid-off tech workers often eventually find jobs elsewhere (sometimes after a relocation or retraining delay). In 2001–03, unemployment in the information sector spiked, but overall U.S. unemployment rose by only 2 percentage points. We might expect a similar modest rise in joblessness, perhaps pushing the U.S. unemployment rate from 4% up toward 5–6% over a year or two. The impact would be concentrated in specific areas, such as Silicon Valley, Seattle, and Austin, and in high-skill occupational categories. Another aspect is labor reallocation: a bubble popping can free up scarce talent for other productive uses. For example, if fewer people chase speculative AI startup jobs, more might work on incremental improvements in existing industries (or even outside tech entirely). It’s conceivable that some sectors (like manufacturing or healthcare) could then more easily hire IT personnel who were previously locked into frothy ventures. In aggregate, though, the near-term effect is adverse for employment and income in the tech sector. Lower bonuses, lower equity compensation (many tech employees would see their stock options underwater), and thus less consumer spending in affluent metro areas. State and local governments in tech-heavy regions could also feel a revenue hit (as California did when capital gains tax receipts plunged after 2000).

Credit Spreads and Financial Conditions: A bubble burst tends to trigger a risk-off repricing across financial markets. We would expect credit spreads to widen as investors demand higher compensation for risk once the illusion of ever-rising asset prices is shattered. In 2001–2002, investment-grade and high-yield spreads both widened (the latter spiked by hundreds of basis points), not because banks were failing, but because default risks rose during an economic slowdown and risk aversion increased. In 2026, if an AI bust triggers a broader equity downturn and recession fears, high-yield corporate bond spreads could widen sharply. Venture-funded firms that had been planning IPOs might run out of cash and default on obligations (leases, supplier contracts), causing some losses in the financial system, though probably not large enough to cause systemic issues. Banks with exposure to the tech industry might see an uptick in non-performing loans. But broadly, the U.S. banking system’s capital is substantial, and its direct exposure to equity market fluctuations is limited so that we wouldn’t expect a banking crisis.

Overall financial conditions, including equity prices, credit spreads, volatility, and the dollar and rates, would tighten initially. Stock declines and wider spreads make financing costlier; the wealth and confidence drop can strengthen the dollar (as a safe-haven) if global investors flock to U.S. Treasuries, or potentially weaken it if the U.S. is seen as hurt more than others (though given U.S. tech dominance, an AI bust is somewhat “made in America,” so a flight to quality into Treasuries is likely, implying a stronger dollar and lower Treasury yields). Indeed, typically when equity bubbles burst, Treasury bonds rally (yields fall) as investors seek safety and central banks cut rates. The dot-com bust saw the 10-year yield fall from 6.5% in early 2000 to 4% by late 2001. In a 2026 scenario, the Fed is already in easing mode; a market rout and growth downturn would spur them to cut rates more aggressively. By the end of 2026, we might imagine the Fed funds rate back near the zero lower bound (as it reached 1% in 2003). The Federal Reserve’s response function today is perhaps slightly constrained by higher inflation than in 2000, but if the bubble popping coincides with disinflation (which asset busts often do), the Fed will have leeway to ease.

Financial market plumbing could face some stress in the adjustment. For instance, if there’s a rush to unwind leveraged positions, markets could become illiquid for a time (spiking volatility, maybe a “flash crash” or two). But the post-2000 experience (and even 2020’s swift correction) shows that markets can adapt and the Fed can step in with liquidity facilities if needed to prevent a spiral. We might also see the corporate bond market bifurcate: top-rated companies will likely have no trouble issuing debt; in fact, they might issue more to buy back stock on the cheap or acquire distressed competitors. Lower-rated issuers, however, could be shut out of the market for a period.

Summary: The macro fallout of an AI valuation reset would likely be significant but manageable: a blow to wealth and investment, leading to a growth slowdown, partially offset by policy easing. It would not resemble 2008’s financial crisis, but more the aftermath of a typical asset bubble bursting, painful for investors, with some potentially losing decades of gains (as with NASDAQ post-2000), somewhat sobering for the economy, yet ultimately survivable and perhaps even instructive. As Brunnermeier and Oehmke note, bubbles can have silver linings by financing innovation. Still, the crash part tends to eliminate the weakest players and reset valuations to more reasonable levels, potentially setting the stage for more sustainable growth. If AI is truly transformative, the end of the bubble could separate hype from reality, with strong companies continuing to advance AI while excesses are stripped away.

Policymakers will, however, have to be vigilant. The Fed in 2025 has been cautious about not explicitly targeting asset prices. Still, if a collapse were to threaten employment or inflation significantly to the downside, more aggressive actions (even unorthodox ones like corporate bond purchases) could come back on the table. Financial regulators might also examine whether any non-bank financial institutions (hedge funds, etc.) were dangerously leveraged to AI stocks, as was done after the LTCM crisis in 1998. One tail risk is that the bubble extends further and intertwines with something like housing or credit. However, at the moment, there’s little sign of that (housing is actually cool due to high mortgage rates), but bubbles sometimes migrate (after the dot-com bubble, the bubble energy moved into housing by 2004–07).

Conclusion: If we are indeed in an AI bubble and it were to pop, it would be a setback, but not the end of the world. The U.S. economy’s diversified structure means many sectors (finance, energy, consumer staples, etc.) could grind along even if tech slumps. The overarching growth narrative might shift from “tech-driven future” to something else for a time. Still, ultimately, the valuable parts of AI would continue to be adopted, just at a less rapid pace. If anything, the 2000–2003 experience suggests that innovation continues through the bust. The early 2000s saw the rise of truly viable internet business models (search, online retail) even as stock prices languished. Similarly, an AI bust might weed out frivolous use cases but accelerate focus on high-value (such as enterprise AI, healthcare, etc.). The macro adjustment would be a story of short-term pain, long-term realignment.

Sources

1. Blanchard, Olivier J., and Mark W. Watson. “Bubbles, Rational Expectations and Financial Markets.” In Crises in the Economic and Financial Structure, edited by Paul Wachtel, 295–315. Lexington, MA: Lexington Books, 1982.
2. Brunnermeier, Markus K., and Martin Oehmke. “Bubbles, Financial Crises, and Systemic Risk.” In Handbook of the Economics of Finance, Vol. 2, Part B, edited by George M. Constantinides, Milton Harris, and René M. Stulz, 1221–1288. Amsterdam: Elsevier, 2013.
3. Gürkaynak, Refet S. “Econometric Tests of Asset Price Bubbles: Taking Stock.” Journal of Economic Surveys 22, no. 1 (February 2008): 166–186.
4. Kindleberger, Charles P., and Robert Z. Aliber. Manias, Panics, and Crashes: A History of Financial Crises. 5th ed. Hoboken, NJ: John Wiley & Sons, 2005.
5. Bank for International Settlements. BIS Quarterly Review. Basel: BIS, September 2025.
6. International Monetary Fund. World Economic Outlook: October 2025. Washington, DC: IMF, October 2025.
7. O’Trakoun, John. “Seeing Double: An AI Bubble?” Macro Minute (blog). Federal Reserve Bank of Richmond, October 7, 2025.
8. Arnott, Rob, Trent Commins, and Xi Liu. “The AI Boom vs. the Dot-Com Bubble: Have We Seen This Movie Before?” Research Affiliates, March 2025.
9. Goldman Sachs Global Investment Research (Peter Oppenheimer). “Why Global Stocks Are Not Yet in a Bubble.” Goldman Sachs, October 21, 2025.
10. J.P. Morgan Asset Management. “Does Circularity in AI Deals Warn of a Bubble?” On the Minds of Investors (market update). J.P. Morgan, October 2025.
11. Metcalfe, Michael. “Institutional Investor Indicators: July 2025.” State Street Global Markets (Macro Research commentary), July 2025.
12. Gannatti, Christopher. “Understanding Bubble Narratives through Historical Context.” WisdomTree (Investment Insights blog), August 26, 2025.
13. Dudley, Bill. “AI Can Transform the Economy and Still Be a Bubble.” Bloomberg, October 9, 2025.
14. Financial Times (Ruchir Sharma). “The Markets Are Alive with the Sound of Echo Bubbles.” Financial Times, February 27, 2023.
15. Reuters. “IMF Says AI Investment Bubble Could Burst, Comparable to Dot-Com Bubble.” Reuters (via Al Jazeera), October 14, 2025.
16. Jen, Stephen. “AI Bubble Isn’t Near a Peak. It’s Only at ‘Base Camp’.” Reuters, October 22, 2025.
17. Stanton, Rich. “World’s Top Banker Says the AI Bubble Will Burst, and Shedloads of Money ‘Will Probably Be Lost’.” PC Gamer, October 9, 2025.
18. Carvão, Paulo. “Is AI a Boom or a Bubble?” Harvard Business Review, October 16, 2025.
19. FactSet. Financial market data on S&P 500 and technology sector valuations (dataset).
20. PitchBook Data, Inc. Quantitative Perspectives: A Fork in the Road (Q3 2025). Seattle: PitchBook, September 2025.
21. S&P Dow Jones Indices. S&P 500 North American Expanded Technology Sector Index – forward P/E and historical data. S&P Global Market
22. Azhar, Azeem, and Nathan Warren. “Is AI a Bubble? A Practical Framework to Answer the Biggest Question in Tech.” Exponential View (Substack newsletter), September 17, 2025.
23. Donthireddy, Sriharsha. “The AI Bubble: Why It Feels Like 2008 All Over Again (And How to Spot the Cracks).” Medium, October 15, 2025.

Disclaimer: The information contained in this publication is for informational and educational purposes only and does not constitute investment, legal, or tax advice. The views expressed reflect current market and macroeconomic analysis as of the publication date and are subject to change without notice. Past performance is not indicative of future results. Readers should not rely on this material as the sole basis for investment decisions and should consult their own financial advisors before making any investment choices.