← Mike Taylor hub  ·  Research hub  ·  Research library

Mike Taylor — The AI Bubble and the Market Risks Ahead

"Where's this trillion dollars going to come from to pay for the depreciation alone?"
2026-SEP-15 · Hedgeye — Real Conversations (host Keith McCullough) · guest Mike Taylor (hedge-fund PM, ex-Millennium) · 58:09 · ▶ Watch · transcript · actionable insights
One-line take: Taylor's case that the AI build-out is a bubble rests on a top-down sanity model, not on quarterly numbers: roughly $3T of AI capex in the ground by end-2027 needs ~$400B/yr of depreciation and ~$1T/yr all-in to carry, against ~$100B+ of OpenAI + Anthropic revenue (70% of the market), which at an 80% margin leaves ~$40B to pay for compute. The bonds said no first: Oracle's credit balked at the OpenAI backlog in March–April, and CoreWeave's unsecured debt yields ~12% ("meaningfully distressed") while street models triple its debt, as in the 2006 housing top. He reads the weekend's AI-regulation press as a "psyop" to entrench an oligopoly with pricing power; expects Anthropic's raise to get done and OpenAI to be the one squeezed out (possibly through bankruptcy). He sees select software (ServiceNow, Palantir) becoming compute aggregators that push pricing down. On macro: consumer, housing, auto and student loans "stink"; sovereign long ends are breaking (JGB 10-year at 3%), and he's shorting Swiss stocks against a 0.58% ten-year. His thesis is that the US "outruns" a Europe heading toward protest-party governments and a return to national currencies. He expects short euro / long gold to be "a gigantic trade" next year, negative pricing power and inflation "materially down" as deficits lock in corporate capacity, and "preserve and protect" positioning now. Host McCullough adds Hedgeye's nowcast (inflation roughly halving by Q2 → Fed panic cuts, as in 2007). Timestamps link into the video.

1. Stocks & names mentioned

TickerNameResearchViewWhat he saidAt
NOWServiceNowQT · SA · STK · FAPositiveLong software alongside McCullough; "select software" such as ServiceNow will be "the middleman," already embedded in tens of thousands of companies and able to aggregate their compute buying for a better price. That is "free business for them" and deepens integration, though it lowers the price of compute.22:28
PLTRPalantir TechnologiesQT · SA · STK · FAPositiveNamed with ServiceNow as a potential compute "middleman": route thousands of clients' AI compute through one platform, negotiate the price down, and get more embedded, positioning software to win as compute pricing falls.22:53
GoldGold (commodity)Positive"You can short euro long gold. That is going to be a gigantic trade" — one of his big ones for next year, but timing-dependent: "I don't know if it's now, but when I see it, I'll know it." Rising bond yields hold gold back for now; "gold likes money printing a lot," and he sees printing as the only lever left.48:53
AnthropicAnthropic (private)NeutralWith OpenAI, ~70% of the AI market and ~$100B+ of combined revenue by year-end, "out of private financing." Still: "I think Anthropic is going to get their deal done… and I think Musk wants them to get their deal done." Anthropic survives the shake-out, but inside a revenue pool too small for the capex.17:27
SpaceXSpaceX (private)Neutral"SpaceX went at $2 trillion": private investors are already 10x+ on marks that are "marked to whatever the hell they want," and they want liquidity "to get the hell out." His evidence that private money for AI-era valuations is tapped out.7:35
MSFTMicrosoftQT · SA · STK · FANeutralThe example of a hyperscaler that "has shown an appetite to cut" AI spend, and "the market has told them that their stock can go back up again." Corporates won't hand 10% of free cash flow to AI immediately — more like 3–4%.19:50
GOOGLAlphabet (Google)QT · SA · STK · FANeutralMcCullough calls Google a stock that "has been terrible" and asks what happens if it cuts capex; Taylor flags "they have a lot of OpenAI exposure," the name he expects to be squeezed out. A risk flag, not a call on the stock.20:36
BMYBristol-Myers SquibbQT · SA · STK · FANeutralCited from his pharma background: "Bristol Myers and Lilly, others… want to bring these sort of things in-house and have their own AI in-house" rather than share sensitive data with a data center. Internalization is another drag on third-party compute pricing.11:15
LLYEli LillyQT · SA · STK · FANeutralNamed with Bristol Myers as a pharma company moving AI in-house to protect proprietary data. An example for his compute-pricing argument, no view on the stock.11:15
GSGoldman SachsQT · SA · STK · FANeutralPart of a ~$500B AI financing raised by "like six groups," firms that "wouldn't ordinarily work together." The hosts mock David Solomon's "I believe deeply in these numbers" as narrative in place of math. A financing reference.14:49
APOApollo Global ManagementQT · SA · STK · FANeutralNamed with Goldman as one of the unusual co-financiers of the ~$500B AI raise. Separately, Taylor wonders who will absorb CoreWeave's planned debt: "maybe insurance companies and annuities." A financing reference.15:12
KOCoca-ColaQT · SA · STK · FANeutralHis illustration of deficit lock-in: "Coca-Cola is building capacity to service a growing deficit." If spending stopped, it "would now have excess capacity and pricing implodes," so "we can't stop spending." An example of the macro argument, not a stock call.35:44
WMTWalmartQT · SA · STK · FANeutralA shopping aside on the squeezed consumer: people "can't make the ends meet," and blueberries cost $4.10 or $9.10 "unless you go to Walmart, which I recommend." A trade-down reference, not a view on the stock.36:11
SOXLDirexion Daily Semiconductor Bull 3x ETFSA · STKNeutral"Yesterday alone, the SOXL traded down 17%," as three-times-levered semis. Cited as retail and levered-ETF flow whipsawing the semis-minus-software spread that has driven 10% moves every 48 hours. A market-structure reference.23:57
CRWVCoreWeaveQT · SA · STK · FANegative"The unsecured CoreWeaves are trading around 12% yield… That is meaningfully distressed," yet street models have it tripling its debt over two years: "You are not going to raise that debt because the cost of capital is too high." Its 12% cost of capital sits inside his ~$1T/yr carrying cost for AI compute.1:56
ORCLOracleQT · SA · STK · FANegativeThe trigger for his work: in March–April the huge OpenAI commitment made Oracle's backlog "explode for multiple years," "and then I went and looked at the bonds. And the bonds said, no." Stocks and bonds disagreeing is "exactly what happened in 06" during the housing bubble.1:13
NVDANVIDIAQT · SA · STK · FANegativeHe couldn't model "what an NVIDIA is going to guarantee and what they're not going to guarantee and what their cost of capital really is," so he went top-down. Chip prices up "6x for NVIDIA" and memory ~7x have pushed the cost of building compute to "astronomical, ludicrous" levels against a revenue pool that can't carry it. The supplier to a build-out he calls a bubble.3:28
OpenAIOpenAI (private)NegativeNeeds ~$600B "to get through the next two years" (low end) with $400B committed to neoclouds, after a ~$960B last round, and "the private money is out." The regulation push is Altman's "panic button." With too many eating a small pie, "they got to get rid of one. Guess who that is? … it's OpenAI," possibly via bankruptcy that wipes out investors while the tech survives.17:53
SFTBYSoftBank GroupSA · STKNegativeHeavy OpenAI exposure; "whipping around like a jackrabbit, up and down 10%, 12% a day" on whether OpenAI gets its money: "this redefines levered gambling."20:36
EuroEuro (EUR — currency)NegativeProtest parties polling "giant" ahead of general elections from next year (UK, France) will take power in coalitions and say "the problem with the euro is that we can't control it and we're going back to the franc." France's GDP is ~59% government, overspending by 5%. The expression is short euro / long gold, "a gigantic trade," with the US meanwhile "outrunning" Europe.32:56
Swiss equitiesSwiss stocks (market)Negative"I started shorting Swiss stocks." Sovereign long ends are breaking everywhere (the JGB ten-year "finally got a three handle"), and Switzerland's is the low yield at 0.58%: "Why not 1.5%? Why can't it double twice?"27:20

"View" is Mike Taylor's stance in this conversation (Positive / Neutral / Negative), not a price rating. The auto-transcript has no speaker labels, so attributions follow context. Where host Keith McCullough makes a point (Hedgeye's inflation nowcast, "Google has been terrible," the SOXL and high-beta momentum data, "we're long software") it is noted as his; Taylor agrees on software ("as am I"). NVDA is rated from his bubble thesis (the chip supplier to a build-out whose economics he says don't close), not from an explicit short. Not tabled: Millennium (his former fund, private), Seawolf Capital (Vinny Daniel & Porter Collins, private), Meta/Mark Zuckerberg and Google DeepMind (passing remarks), South Korea's KOSPI (McCullough: "down 27%… looks exactly like that chart"), Japanese 10-year JGBs and US Treasuries (macro), and the TV shows Lioness / Landman. Research: QT Qualtrim · SA Seeking Alpha · STK Stock Analysis.

2. Talking points

1:13 Oracle's backlog exploded — and the bonds said no

1:56 CoreWeave unsecured at ~12%

3:08 Erase the quarters: a top-down model

4:21 $3T needs ~$1T a year; revenue leaves ~$40B

5:54 Stress-testing it with the bulls

7:12 Private money is out

8:27 The AI-regulation "psyop"

10:40 Pricing is the next big problem

12:02 Narratives, not numbers

17:27 Anthropic gets its deal; OpenAI gets squeezed

19:01 The pie: S&P top-50 free cash flow

21:18 Momentum collapse and pod-shop knife fighting

22:28 Long select software: the compute middlemen

24:24 The end-June index-arb event broke books

25:21 The hard part: the economy outside AI

26:56 Sovereign long ends are breaking; short Swiss stocks

28:36 M2 must grow 2–3%, but units are shrinking

30:50 Outrun the slowest bear: Europe

34:26 Deficits have become corporate capacity

38:19 Next year: negative pricing power and falling inflation

42:42 Bessent dealt 2-7 off-suit

46:39 Preserve and protect the pile; passive over active

48:53 Short euro, long gold, when the moment comes

50:02 Gundlach's +7% import/export prices is a base effect

53:46 A generation that hasn't seen a crash

3. In plain English

A jargon-free summary of the thesis behind each pick: what it actually is and why he holds that view. (Plain-language companion to the table above; renders on each ticker's consolidated page.)

NOW — ServiceNow Positive

ServiceNow sells workflow software that big companies use to run IT, HR and customer-service processes, so it already sits inside thousands of corporate systems. Taylor thinks AI computing power is being overbuilt and will get cheaper. He sees a new role for software like this as a buying agent: instead of each company renting AI compute on its own, ServiceNow could pool the demand of 10,000 customers, negotiate a better price and route the work through its own platform.

For ServiceNow that is extra business at little cost that makes it harder to replace. His broader view is to own the software that uses AI rather than the companies spending trillions to build it. He and the host are both long software.

PLTR — Palantir Technologies Positive

Palantir builds data and AI platforms for governments and corporations. Taylor names it alongside ServiceNow as a likely "middleman" in AI: a company already embedded with customers that can aggregate their AI computing needs, drive down what they pay the data-center owners, and deepen its own role. If computing power becomes a surplus commodity, he expects companies that control the customer relationship to win and landlords of the chips to lose.

Gold — Gold (commodity) Positive

Taylor's macro story is that heavily indebted governments can no longer borrow cheaply for long periods, so the only way out is printing money, which is what gold likes. His specific trade pairs gold with a bet against the euro: borrow or sell euros and hold gold, so the position pays if European politics fractures the single currency and if central banks print.

He's explicit that timing matters. Right now bond yields are rising quickly, which hurts gold because gold pays no interest. He expects "a moment next year" when yields turn and printing resumes, and says he'll know it when he sees it.

Anthropic — Anthropic (private) Neutral

Anthropic makes the Claude AI models, which Taylor uses himself (he read a 280-page contract with it in an hour). Together with OpenAI it takes about 70% of AI spending, but he thinks their combined revenue is far too small to pay for the trillions of data centers built for them, and private investors have run out of appetite. Even so, he expects Anthropic's current fund-raise to close, helped by Elon Musk wanting it to. He sees it as a survivor of the shake-out, not a winner of the math.

MSFT — Microsoft Neutral

Microsoft is one of the "hyperscalers" pouring money into AI data centers. Taylor points out that when it signaled willingness to cut that spending, its stock went back up. Shareholders are rewarding restraint rather than endless capex funded by negative free cash flow and more debt. He expects big companies to put only 3–4% of their free cash flow toward AI at first, not the 10% the build-out would need.

CRWV — CoreWeave Negative

CoreWeave is a "neocloud": it borrows heavily to buy NVIDIA chips and rents the computing power to AI labs. Taylor's key clue is its bonds. Its unsecured debt (debt not backed by specific assets) trades at about a 12% yield, a level lenders demand from companies they worry may not pay them back, even though the stock market treats CoreWeave as a secure growth story.

Wall Street models assume CoreWeave triples its debt in two years. Taylor says that can't happen at a 12% cost of borrowing, and that the high interest bill is part of why AI computing costs about a trillion dollars a year to carry. When bond investors and stock investors disagree this sharply, he trusts the bonds, as in 2006 before the housing crash.

ORCL — Oracle Negative

Oracle signed an enormous cloud-computing commitment with OpenAI that made its order backlog soar for years ahead, and the stock loved it. Taylor looked at Oracle's bonds instead, and they sold off from March–April. Lenders were saying the deal adds risk, because it depends on a customer (OpenAI) that must raise hundreds of billions of dollars it doesn't have. That stock-up, bonds-down split launched his whole AI-bubble analysis.

NVDA — NVIDIA Negative

NVIDIA sells the chips behind the AI build-out. Taylor didn't call for shorting it outright, but his numbers point against it: about $3 trillion of AI data centers will exist by end-2027, and wearing out that equipment alone costs around $400 billion a year. Revenue from the main AI customers leaves only about $40 billion to pay for it. Chip prices up roughly six-fold and memory seven-fold have made that gap worse. He also says it's unclear how much of its customers' financing NVIDIA is effectively guaranteeing. If the spending has to slow, the company selling the shovels feels it.

OpenAI — OpenAI (private) Negative

OpenAI (ChatGPT) last raised money at roughly a $960 billion valuation and, by Taylor's estimate, needs at least $600 billion more over two years, having promised $400 billion to data-center builders. He thinks private investors are tapped out. He reads the sudden wave of "AI must be regulated" headlines as an attempt to lock in a small club of US players who could keep prices high.

If that fails, AI pricing falls and not everyone survives. Taylor expects OpenAI to be the one squeezed out. That doesn't mean the technology disappears: companies can keep operating through bankruptcy and be bought. But the current investors would be wiped out.

SFTBY — SoftBank Group Negative

SoftBank is a Japanese investment conglomerate with a huge, borrowed-against stake in OpenAI. Its shares now swing 10–12% a day on each headline about whether OpenAI will get its funding. Taylor calls that "levered gambling": a leveraged bet on the one AI lab he expects to lose the shake-out.

Euro — Euro (EUR — currency) Negative

Taylor expects anti-establishment "protest" parties, now polling far ahead in countries like France and the UK, to win power in coalitions during general elections starting next year. France's economy is about 59% government spending and runs 5% overspending that EU rules can't stop. He thinks a populist government will blame the euro, since it can't print its own money, and push to bring back the franc.

His trade is to bet against the euro and own gold. He also argues this helps the US: like running from a bear, America only has to outrun the slowest, and if Europe blows up, US government debt looks safer by comparison.

Swiss equities — Swiss stocks Negative

Government bond yields are rising around the world as heavily indebted countries struggle to borrow long-term; Japan's 10-year just hit 3%. Switzerland still pays only about 0.58% on its 10-year bond. Taylor asks why that can't double twice, to 1.5% or more. If it does, Swiss stocks, many of them steady low-growth companies valued like bonds, get repriced lower. He has started shorting them.


Summary & timestamps derived from the public YouTube video (transcript in transcript.html) for personal study. Not investment advice. © Hedgeye for source material.