What Wall Street Really Thinks About the AI Trade Now

Alexander 'Alty' Altmann · Barclays

Neocloud financing is the fragile periphery of AI, showing the early stress that preceded the GFC.

Altman says neoclouds are contractually bound to build compute for customers, some of it speculative, and most have little or no operating cash flow. That makes them the riskiest part of AI financing, unlike hyperscalers such as Google and Amazon, which have large cash flows. He compares this to the GFC, which began with subprime vehicles 12 to 24 months before the big banks were hit. He cites Cipher's convertible deal, which hurt its own share price because it could not tap debt markets directly, and says that triggered the CoreWeave bond sell-off. Data center bonds had already been widening for months, so he sees ripple effects starting in the periphery.

“these are tip of the spear. These businesses don't have operating cash flow today or minimal operating cash flow today in many cases.”at 7:32

Alexander 'Alty' Altmann · Barclays

Efficiency gains let locally run models handle everyday tasks without massive data center compute.

Altman argues the leading question is not just open-weight models but locally run ones. He compares today's phones to the iPhone of 2007, saying the compute gain would have seemed impossible then with the same battery size. He says there is no need to run Hopper or Vera Rubin chips in an Ohio data center just to manage email. Multi-billion parameter models can already run on laptops, Meta has released one, and the market is moving along an efficiency curve. That shift affects the revenue question for frontier models, which he doubts will discover new drugs.

“the idea that we necessarily need to ping a bunch of Hoppers or Vera Rubins in an Ohio data center just so that we can sort of get our, you know, agent... Manage my email for me.”at 24:07

Alexander 'Alty' Altmann · Barclays

A hyperscaler capex cut would be the key spooking signal, but it is probably too early for this week's results.

Asked what would spook investors, Altman names some level of capex being reined in by one of the hyperscalers as the most likely trigger. A client had asked him whether this could appear in this week's results. He says it is probably still premature for that to show up now, and he suspects investors may have to wait until the new year for a real change in enthusiasm. He notes that Google set off the last wave on the capex side.

“I would say it would be some level of CapEx being reined in by one of the hyperscalers.”at 35:11

Alexander 'Alty' Altmann · Barclays

Hyperscaler AI capex is roughly 12% of GDP, the largest capital experiment in US history.

Taking the forward capex consensus at face value, Altmann puts AI spending at about 12% of GDP, roughly two Netherlands. He says that exceeds the Louisiana Purchase, the railroads, and the pyramids and the Great Wall in scale and speed. He hedges that the consensus numbers will change. His point is that companies with strong legacy cash-generating businesses are embarking on the largest capex experiment in American history.

“If you believe the CapEx numbers in the terms of the forward, and I'm sure that they're gonna change from what the consensus looks like now, but on face value, it's about 12% of GDP.”at 43:12

Alexander 'Alty' Altmann · Barclays

AI is in the denial phase of a bubble; investors remain overwhelmingly bullish despite financing stress.

Altman describes a bubble chart where the first decline is a denial phase. He says he would not claim the bubble has popped, but argues the market is in denial because nearly everyone is still extremely bullish. He backs this with a survey where only about 60% of investors said AI financing would be a problem, even though recent events should have pushed that to 100%. He reads the gap between the bullish consensus and the financing evidence as denial of a reality already here.

“I'm not gonna say we're at kind of this bubble's popped... we're in the denial phase right now because everyone is still extremely bullish.”at 5:43

Alexander 'Alty' Altmann · Barclays

Debt issuance has decoupled from rates in 2026; AI competes for finite capital, with $1.1T net IG and HY issuance forecast.

Barclays' data since 1995 shows net issuance across IG, high yield, Treasuries and equities normally rises when rates fall. In 2026 that link has broken down, with issuance far above what rates would suggest, which he calls almost biblical. He says this matters for AI because AI is not issuing debt in a vacuum and other sectors also need financing, so capital is finite. He also says the idea that AI was immune to higher rates is not true; the Meta results showed financing does have limits. He forecasts net IG and high yield issuance of $1.1 trillion this year and argues the casual use of trillions in headlines ignores those limits.

“AI is not issuing debt in a vacuum. There are other verticals that need to tap financing too... And there is a finite amount... So IG issuance and a high yield this year on a net basis, we forecast to be $1,100,000,000,000.”at 13:41

Alexander 'Alty' Altmann · Barclays

AI revenue lags the spending story: tech TAM claims are overstated and token spending is levelling off near term.

Altman argues that the trillion-dollar AI TAM claims need checking. He cites Barclays research showing that the cumulative inflation-adjusted revenue of Microsoft and Apple since 1990 was under $10 trillion, while SpaceX's addressable market is pitched at about three times that. He then sets S&P 500 revenue of about $17 trillion against $11-12 trillion of expenses, leaving net profit of roughly $5-5.5 trillion as the pool that AI spending has to come out of. Ramp and OpenRouter data show token spending has begun to level off, at least near term.

“But last time I checked, we all look at the ramp data or the open router data, and we can clearly see that that token expenditure has kind of begun to level off a bit. Right? At least near term.”at 21:07

Alexander 'Alty' Altmann · Barclays

Open-weight models have grown from about 20% to up to 80% market share, with token costs deflating fast.

Altman says the data across several sources shows open-weight models rising from about 20% market share to as much as 80%. Token costs are a fraction of closed-model prices, which he calls one of the fastest deflating assets in history. He calls this a contradiction to the bullish frontier-model narrative. He also notes GPU rental rates have risen, so the debate is two-way rather than a one-way train.

“the data is the data you can evidently see on a whole bunch of different sources that open weight has kind of gone from a 20% market share to anything up to about an 80% market share.”at 23:15

Alexander 'Alty' Altmann · Barclays

A consumer killer app is needed to re-ignite AI; 100-IQ agentic workflows could be ubiquitous within 12 to 18 months.

Altman says he would get more constructive on AI once a killer consumer application emerges, because the consumer is the economy. He argues that the consumer does not need frontier models for everyday tasks such as managing email or handling disputes with a rental company. He expects the 'good enough' agentic workflow to become almost ubiquitous within 12 to 18 months. Creative application of existing technology, rather than new advanced technology, is where he thinks the next upswing in AI will come from.

“I said, when we figure out that killer consumer application layer. Because at the end of the day, the consumer is everything. It's the economy.”at 31:41

Alexander 'Alty' Altmann · Barclays

Enterprise AI revenue will not close the gap within the time frame the market wants to see ROI.

Altman says enterprise AI revenue will probably get there eventually, but it cannot close the revenue gap on the timeline the market wants to see returns. Because of this, he argues AI needs to break into the consumer layer. He sees this as the main revenue question for the sector and ties it to his view that consumer adoption is the missing piece.

“To the point about revenues, like, enterprise revenues, I just don't think is gonna plug that gap.”at 31:52

Alexander 'Alty' Altmann · Barclays

Crowding from multi-manager funds suppresses index volatility while AI-factor rotations stay violent.

Multi-manager fund AUM rose from about $300B in 2020 to over $1T today. Long-short funds deploy gross capital and leverage it through prime brokers, while the number of tradable securities has stayed flat or shrunk. The result is crowding. Funds try to neutralise style factors, but everyone does this, so the crowding just shows up as different kinds of crowding. That is why realised volatility and the VIX look benign at the index level while individual stocks rotate violently. The AI factor is the hardest to map because its betas change with the narrative (shorting software was an AI-factor trade at the start of the year, and software is now near 12-month highs). Altmann's bottom line is that you are either long AI or short AI.

“The problem is that you've effectively tripled... More than tripled the AUM plus, obviously, leverage, and the number of securities you can trade has stayed roughly the same or actually shrunk over time. So that creates crowding out.”at 39:16

Alexander 'Alty' Altmann · Barclays

AI ecosystem, led by semiconductors, will supply about 75% of S&P 500 earnings growth next year.

Altmann argues the S&P 500 is becoming concentrated in AI-linked earnings. He expects about two-thirds of next year's S&P earnings growth to come from semiconductors, and about 75% once the broader AI ecosystem is included. That concentration is why he says the S&P 500 is becoming a Kospi-style index dominated by one theme, and why the index has been hard to beat.

“It's about two thirds from semiconductors, and then the broader AI ecosystem gets you to around about 75%.”at 44:49