Is AI a Bubble? Six Macro Experts Weigh In (2026) | Wealthion

Wealthion Editorial | August 2026

Is AI a bubble? The most common answer from six macro experts interviewed on Wealthion over the past two months: yes, but not where most investors are looking. The bubble sits in the financing of the buildout, not the technology, and the bond market, not the tech sector, will decide when it ends. That is the through-line connecting Jesse Felder, Peter Boockvar, Michael Howell, Jeff Currie, Michael Strain, and Steven Feldman, organised here around the questions investors are asking as of August 2026.

What is the AI bubble, exactly?

When experts talk about an AI bubble in 2026, they rarely mean the technology itself. They mean the capital spending behind it.

Peter Boockvar, chief investment officer at Bleakley Financial Group, drew the line precisely in his July conversation with Wealthion: "the bubble is clearly in the capex side, not necessarily the technology." In his framing, the users of AI ultimately benefit; the risk sits with the spenders, the hyperscalers committing more than $700 billion to data centers, chips, and power.

Boockvar pointed to Oracle as the clearest example: roughly half of its remaining performance obligations trace back to a single customer, OpenAI, while its capital spending climbed from half its revenue to roughly all of it. Markets now punish the spenders and reward the recipients: semiconductor, memory, and server companies.

The AI bubble, then: a capex boom of historic scale, financed increasingly by debt and circular arrangements, with a return on investment still unproven.

Is AI actually a bubble, or is that just noise?

Most of these experts say yes about the spending; the dissent is worth hearing. Michael Strain of the American Enterprise Institute offered the counterpoint in his July interview: "I am not worried that current AI companies are overvalued," at least in aggregate, because corporate earnings have been genuinely strong. His bigger worry is that government involvement in AI under a national security banner could make the sector too big to fail, preventing the market from rinsing out excess the way it normally would.

Jesse Felder, publisher of The Felder Report, sits at the other end. In his August conversation with Maggie Lake, he argues the quality of hyperscaler earnings has deteriorated even as headlines stay strong: "these companies have literally never reported lower quality earnings than they are today." Reported profits look healthy, but free cash flow has turned negative under the weight of capex, and rising depreciation charges have yet to fully hit income statements.

Boockvar added a telling detail: of one quarter's roughly 28 percent hyperscaler earnings growth, about 12 percentage points came from other income, largely mark-to-market gains on stakes in OpenAI, Anthropic, and SpaceX. Underlying growth was closer to 16 percent.

Is the AI bubble bursting right now?

Felder believes the unwind has begun: "I think it's the beginning of the end." His reasoning is market mechanics, not sentiment. The momentum trade that carried AI hardware stocks higher, including leveraged long-hardware, short-software positioning by high-profile funds, blew up in a single month. In his words, the last leveraged buyers were the greatest of the fools in a greater-fool game, and now "there are no more fools" to carry the trade higher.

Michael Howell of CrossBorder Capital frames the same moment through liquidity. "Liquidity is inflecting and slowing" even as the economy runs hot, commodity prices climb, and the yield curve bear-flattens, he told Wealthion in June. In his reading, everything about this market looks late-cycle except the tech surge itself, and AI spending is acting as a giant drain on financial market liquidity.

The signal both men watch is the bond market. Felder calls the 10-year Treasury "the most important chart in the world," because the entire buildout is financed at rates that keep drifting higher. Boockvar noted the unusual driver: real rates are rising even while TIPS inflation expectations stay muted, which he reads as the bond market finally caring about deficits, including the flood of hyperscaler debt issuance.

How does the AI bubble compare to the dot-com bubble?

The dot-com comparison came up in nearly every conversation, used in two distinct ways.

Howell uses it as a warning about where the deflation lands. He recalls the fiber-optic buildout of the late 1990s: enormous investment that eventually crushed the price of the product itself. He expects the same here. AI prices will tumble under all this investment, but in the short term the capex boom is inflationary for the broader economy. Deflation for the product, inflation for everything feeding it, from electricity to construction to commodities.

Jeff Currie, who coined the phrase "the revenge of the old economy" when the dot-com bubble was crashing in 2002, uses the comparison to point at what got starved. In his July appearance, he laid out the arithmetic. Commodities and energy make up roughly 3 percent of the S&P 500 today. The core of his supercycle argument is that the share belongs closer to 10 to 15 percent, which implies AI-linked market caps are overvalued relative to the hard assets that power them. Currie's first principle is blunt: "if you can't turn the lights on nothing happens and if you don't innovate you never progress."

Put Howell and Currie together and you get the paradox that separates this cycle from 1999: AI deflates its own product while inflating everything that feeds it. That is why these experts can be bearish on AI equities yet bullish on the real economy inputs behind them. Felder adds the market-structure rhyme: when the dot-com momentum trade broke in March 2000, value stocks outperformed through the rest of that year. He sees the same rotation beginning now.

What happens if the AI bubble bursts?

The spillover map runs through three channels: the private AI giants, Treasury yields, and equity multiples.

Felder's key tripwire is the private AI giants, OpenAI and Anthropic, which account for the majority of hyperscaler compute backlogs yet remain heavily loss-making. He notes that Amazon recently advanced OpenAI $35 billion ahead of schedule, money originally tied to milestones that had not been met. In his view, these companies must come public to fund their commitments, and any doubt about those IPOs would be the loudest warning sign yet for the buildout's sustainability.

Howell's concern is the transmission into rates and valuations. With nominal GDP growth running near 7 to 8 percent by his estimate, long-term Treasury yields face structural pressure to rise toward that level, and equity multiples get crushed when inflation picks up or yields spike. Felder goes further: he believes the Federal Reserve under Kevin Warsh will ultimately have to tighten financial conditions far beyond 2022 levels, enough to resemble a typical recession, to end five years of above-target inflation.

How are these experts positioning for it?

Each expert has a distinct playbook, worth stating precisely because they differ. Note the common thread: every one of them is positioned around the financing squeeze, not against the technology.

Felder has rotated toward value stocks, including beaten-down software names on the losing side of the momentum trade, and points to Warren Buffett holding roughly $400 billion in short-term Treasuries as the reference case for patience. Howell favors short to mid-duration bonds, where he argues the carry protects against further yield increases. Currie's answer is exposure to the old economy itself: energy, metals, and the physical inputs of the buildout, with a caution that futures-based commodity products carry roll-yield complications. Steven Feldman's focus, from his June interview, is the bottleneck: "Energy infrastructure has got to be the single biggest theme around AI." A trillion-dollar buildout still has to be plugged in.

None of this is a directive. It is a map of how experienced investors are reading the same data differently, which is what makes the moment worth studying.

FAQ: The AI Bubble in Brief

Is AI a bubble? Most experts Wealthion interviewed say the bubble is in AI capital spending, not the technology. The question is whether more than $700 billion in buildout spending can earn a return.

Is the AI bubble bursting now? Jesse Felder argues the unwind began when the leveraged momentum trade in AI hardware broke. Michael Strain disagrees, saying aggregate earnings still justify valuations. The answer is contested.

When will the AI bubble burst? No expert offered a date. The tripwires they watch instead: whether OpenAI and Anthropic can complete IPOs, whether the 10-year Treasury yield keeps making higher highs, and whether hyperscaler free cash flow keeps deteriorating.

How is the AI bubble different from the dot-com bubble? The dot-com bust deflated the product and the market together. This time, experts like Michael Howell expect AI product prices to fall while the buildout keeps inflation and commodity demand elevated in the broader economy.

What assets could benefit if the AI trade unwinds? The experts named value stocks (Felder), short to mid-duration bonds and gold as a monetary hedge (Howell), and energy, metals, and infrastructure tied to the physical buildout (Currie, Feldman). Each view is that expert's own position, not a recommendation.

Which experts and interviews does this article reference? Six Wealthion interviews from June to August 2026: Jesse Felder on Treasury yields and the AI boom, Peter Boockvar on the AI spending bubble, Michael Strain on the government's AI bet, Jeff Currie and Steven Feldman on the AI boom's fatal flaw, Steven Feldman on the AI resource rush, and Michael Howell on AI and inflation.

Wealthion editorial content is for informational purposes only and is not investment advice. The views quoted belong to the named guests.

The article is a map, not a directive. For a professional read on how these risks apply to your own holdings, request a free, no-obligation portfolio review at https://www.wealthion.com/advisors/