Wall Street's AI Bull-Bear Debate: Bubble or New Cycle?
N.R. Finch
Three years into the AI trade, BCA Research staged a rare internal bull-bear showdown — bears say U.S. equity valuations have surpassed the 2000 dot-com peak; bulls counter that compute remains undersupplied. The real wager: whether capex returns can materialize before multiples compress.
Why do the bears say today's market is pricier than the 2000 dot-com peak?
BCA chief economist Peter Berezin and chief strategist Arthur Budaghyan argue U.S. stocks are in an "earnings bubble" — profit-margin durability is overstated, and valuations have priced in the best-case outcome.
Key number: adjust S&P 500 margins back to 2019 levels and the index trades at a forward P/E of roughly 27×, above the March 2000 historical peak of about 26.5×. This means → even ignoring tech, current valuations have no cushion by historical standards.
Strip out tech and financials, and the rest of the market trades at 26× trailing P/E — yet profit growth over the past two to three years has been only about 3%. In plain terms = prices have run far ahead; earnings barely moved.
Is AI capex inflating profits — or eating cash flow?
The bears flag an accounting illusion: hyperscalers — Amazon AWS, Microsoft Azure, Google Cloud and peers — are collectively pouring hundreds of billions into data centers, chips, and grid upgrades. Book profits are rising, but cash flow is not keeping pace.
The mechanism: every chip sold shows up as revenue and profit on the seller's books (e.g., Nvidia). The buyer (e.g., Microsoft) books it as capex, not operating cost. This means → the chip seller's profit is real cash, but the buyer's reported earnings are flattered by capitalization — actual cash is flowing out.
Berezin's analogy: AI may follow the path of electricity — it boosts efficiency, but if everyone has access, it won't necessarily make companies more profitable.
Why do the bulls say the market is underpricing AI returns?
BCA portfolio-construction head Juan Correa and equity-research head Noah Weisberger take the opposite view: compute remains in supply scarcity, not glut, and the market is underestimating capex returns.
Demand evidence: hyperscaler cloud backlogs — orders received but not yet fulfilled — grew by roughly $750 billion over the past two quarters. This means → customers are queuing for compute; supply is nowhere close to catching up.
Return evidence: latest-generation GPU lease prices have kept rising through the year; cloud-contract value per megawatt has roughly doubled year-on-year; EBIT per employee has surged. Meta's AI-driven ad-revenue lift and Google Search's strong rebound confirm that earnings improvement is broader than the market recognizes.
The bulls' wild card: are hyperscaler valuations actually at a decade low?
Despite widespread concern about stretched multiples, bull-side data shows hyperscalers are trading at their lowest valuations in nearly a decade — even as returns keep improving.
This reflects a mismatch: the market defaults to "AI narrative = expensive," but the companies actually spending on compute are being valued cheaply.
The bulls' call: "The market will be stunned over the next two quarters by the sheer scale of earnings these capital expenditures actually generate."
What are the two sides really betting on?
The debate's premise is not in dispute — both sides agree AI is fundamentally reshaping productivity, capital allocation, and labor markets. The disagreement is not "Is AI real?" but the time window.
The core wager: how much longer can AI spending enthusiasm outrun skepticism? And when the reckoning arrives, how severe will the shock be?
Strikingly, even the bulls concede this trade ends badly. Correa and Weisberger in their own words: "The capex trade will eventually end, and it will likely end in a destructive fashion."
So how should investors read this moment?
The bulls' framework: economic and fundamental tailwinds still dominate for now. The biggest risk is multiple compression, not recession or an earnings collapse.
In plain terms = the real bull-bear divide is not whether AI is good — it is whether "the speed at which capex returns show up" can outrun "the speed at which the market loses patience."
The market's real-time verdict the week of the debate: S&P 500 fell 0.2%, Dow slipped 0.6%, Nasdaq was roughly flat — no panic, but no conviction either.
Content is for reference only, not financial advice.