The AI boom is over - here's your bubble survival guide

By Jurica Dujmovic

Everyone's asking the wrong question about an AI bubble. Here are the stocks to buy - and when.

I've been watching the AI bubble inflate for months now - the unsustainable burn rates, the pilot projects that went nowhere, the startups that couldn't explain their unit economics. So when industry insiders finally started ringing alarm bells, I couldn't resist diving deeper into what's really happening.

What I found surprised me. Yes, there's an AI bubble. But the viral claim that it's 17-times larger than the dot-com crash is misleading. MacroStrategy Partnership's headline-grabbing analysis measured total capital misallocation across all asset classes, not AI specifically. It seems to me that the math behind the analysis was designed to terrify, not inform.

More importantly, everyone's asking the wrong question.

The debate shouldn't be whether the AI bubble will pop - that's already happening. Not in some dramatic, market-crushing way that will dominate headlines, but through a slow-motion deflation that's quietly reshaping the entire landscape. The casualties are already piling up: startup shutdowns surged in 2024; in 2025, 95% of enterprise AI pilots failed to deliver measurable P&L impact within six months of launch; and down rounds this year hit a decade high at 15.9% of all venture deals.

Many companies that scream the loudest about the bubble are either exaggerating the threat or fundamentally misunderstanding what's actually unfolding. This isn't the 2000 dot-com collapse all over again. It's far more interesting and far more profitable if you know where to look. I've identified three distinct market tiers facing radically different fates:

Tier 1: The hyperscalers (Microsoft (MSFT), Alphabet (GOOGL) (GOOG), Amazon.com (AMZN), Meta Platforms (META), Apple (AAPL)) are essentially unassailable. Their estimated $320 billion to $340 billion in 2025 capex spending - largely for AI and cloud infrastructure - comes from operating cash flow generated by their core businesses. They can weather extended periods of disappointing AI returns because their core businesses print money.

Tier 2: The unicorns in limbo (OpenAI, Anthropic, Scale AI) command massive valuations but face existential questions. Can they achieve returns that justify stratospheric valuations? Can they compete against both hyperscalers with infinite resources and substantially cheaper models coming in from overseas? The next 18 to 24 months will separate the winners and losers.

Tier 3: Mass casualties are showing up in the numbers. We're seeing more shutdowns (up 26% year over year in 2024), decade-high down rounds in 2025, thinner deal counts and smaller checks, plus enterprises scrapping a big chunk of AI work and failing to get a return on investment (ROI) from pilots.

Expectations are being managed to engineer an orderly deflation rather than a violent crash that could seriously harm AI's credibility for years to come.

When Sam Altman openly admits that investors as a whole are overexcited about AI at the same time his OpenAI reaches a valuation of $500 billion, he's not confessing failure - he's managing expectations to engineer an orderly deflation, rather than a violent crash that could seriously harm AI's credibility for years to come.

This creates a rare market asymmetry: The public sees "bubble" and flees everything AI-related, while sophisticated capital is making precise bets on deflation survivors. Here are three investment strategies to put you on the winning side of any correction.

1. Buy Tier 1 hyperscalers after the 15% to 20% correction

When the next correction hits the hyperscalers, treat it as an entry window, not a reason to flee.

Microsoft: Microsoft's AI business is already at a $13 billion annual run rate, up 175% year over year, but most of its money still comes from core products including Azure, Office and Windows. With about $72 billion in annual free cash flow, Microsoft can easily fund its AI buildout without stretching its balance sheet. A pullback in the stock to $400 to $420 would bring the valuation back to around 30-times earnings - right at historical norms and a strong long-term entry point.

Amazon.com: Amazon's cloud unit Amazon Web Services (AWS) is growing about 17.5% year over year, hitting a $123 billion annual run rate. The bulk of Amazon's revenue still comes from retail and e-commerce, giving the company the cash flow and scale to invest heavily in AI infrastructure without jeopardizing its core business. If the shares pulled back 15%, the valuation would drop into the 27- to 30-times-earnings range, offering a potentially compelling entry point.

Alphabet: Internet-search advertising still accounts for roughly 56% of revenue, and for the second quarter of 2025, the company posted an operating margin of 32.4%. Estimated capital expenditures for 2025 are roughly $85?billion, much of which is going into AI and data-center infrastructure, yet Alphabet maintains a strong balance sheet to absorb variable returns. A correction of 15% to 20% would lower the valuation to about 22- to 23-times earnings, potentially offering the best entry within the hyperscaler group.

DeepSeek's January 2025 reveal triggered a brief panic across AI stocks; hyperscalers slipped 2% to 5%, while AI chip and infrastructure names plunged 15% to 30% before rebounding within days. The takeaway for investors is clear: When the next 15% to 20% correction hits the hyperscalers, treat it as an entry window, not a reason to flee - the long-term infrastructure buildout remains intact.

Read: Believe the hype: AI is coming for your job. Plan now.

2. Invest in data centers before power constraints are a given

Data-center REITs provide diversified exposure without taking single-name AI risk.

Gartner's November 2024 forecast warns that 40% of AI data centers could face power-availability constraints by 2027, underscoring the physical limits of AI expansion. The firm projects that AI computing workloads alone could consume around 500 terawatt-hours of electricity annually by 2027 - more than two-and-a-half times the 195 TWh used in 2023.

Dominion Energy (D) is ground zero for the AI power crunch. In Virginia - the world's most populous area for data centers - data-center capacity demand has nearly doubled in just five months, jumping from 21 gigawatts under contract to more than 40. These servers already consume a quarter of the state's electricity, and Dominion expects demand to double again by 2039.

To keep up, Dominion is spending $50 billion over the next five years on new electric-transmission projects and infrastructure upgrades -including an offshore wind project, a solar and battery storage project, and five small modular reactors planned for the 2030s. The stock throws off a 4% to 5% dividend yield, but what really matters is positioning: Dominion isn't just a utility anymore - it's becoming the backbone of the AI era's grid.

Vistra (VST): Vistra signed a 20-year deal to supply 1,200 megawatts (MW) of nuclear power from its Comanche Peak facility to an investment-grade buyer. It's also adding 860 MW of new gas capacity at its Permian Basin site (boosting it to 1,185 MW) and has outlined up to 2,000 MW of Texas gas additions by 2028 to meet surging data-center and industrial load.

In addition, data-center REITs provide diversified exposure without taking single-name AI risk. These include:

Digital Realty Trust (DLR): Operates more than 300 data centers globally and has a long record of annual dividend increases. The stock's current dividend yield runs close to 3%.

Equinix (EQIX) operates more than 270 data centers serving in excess of 10,000 customers. Last year, the company marked 85 consecutive quarters of top-line growth and said announced projects were more than 85% leased and preleased as of the first quarter of 2025, underscoring demand from hyperscalers and large enterprises.

Vertiv (VRT): Nvidia (NVDA) last year partnered with Vertiv for AI data-center builds, and Vertiv's stock has soared. The company's valuation is high, but the stock is riding a market that Goldman Sachs sees jumping to $10.6 billion next year, from $4.1 billion in 2023, as liquid cooling reaches 57% of AI servers.

Schneider Electric (FR:SU) (SBGSY): This French multinational is a steadier way to play the data-center theme, with a 24% exposure to data-center end markets and 25% fourth-quarter growth in North American energy management on surging AI demand.

Conservative investors should build positions in companies poised to profit from the coming power crunch. As AI data centers push grid capacity to its limits, these firms stand to benefit from the multiyear investment cycle required to expand power generation, cooling and infrastructure. With the 2027 constraint window approaching, the next year offers a favorable entry period.

3. Buy what the market has mispriced

The stock market is committing a cardinal error: indiscriminately punishing profitable companies with documented ROI along with speculative AI startups with no path to profitability. This creates two distinct opportunities.

First, look at companies that solve unglamorous back-office problems through automation. These include:

UiPath (PATH): UiPath boasts 83% gross margins and its solid client base includes giants such as Orange Spain (FR:ORA) (ORANY) and Travis Perkins (UK:TPK) (TPRKY), yet the stock trades at about four-times sales.

BlackLine (BL): Third-party studies show triple-digit ROI from BlackLine's close automation - 234% ROI at Creditsafe (12.4-month payback) and 379% ROI at Red Wing (Nucleus Research). BlackLine returned to GAAP profitability in 2024 and authorized a $200 million share repurchase (later increased by another $200 million).

(MORE TO FOLLOW) Dow Jones Newswires

11-01-25 1119ET

Copyright (c) 2025 Dow Jones & Company, Inc.