Three-quarters of the S&P 500's earnings growth next year will come from the AI ecosystem, with about two-thirds from semiconductors alone, Barclays strategist Alex Altmann said on Bloomberg's Odd Lots podcast in an episode recorded on October 28, 2026 - while warning that investors are in a "denial phase" about financing strains that have already reached the most indebted builders of AI computing capacity.

Altmann, global head of equities tactical strategies at Barclays, spoke to hosts Tracy Alloway and Joe Weisenthal at the end of a dense week for the AI trade. Anthropic had filed its registration statement for an initial public offering two days earlier. The Federal Reserve's rate-setting committee announced its decision that afternoon. Oracle put out long-term guidance from its investor day while the Fed chair was still speaking. And while the three were still recording, Alphabet and Microsoft published quarterly results that sent their shares in opposite directions.

The episode is a guide to how institutional money is reading those events, not a set of new disclosures. Most of its numbers are Altmann's estimates or Barclays' own research, offered in conversation. A few deserve caution, and the automatic transcript YouTube attaches to the video garbles several company names and figures. Both points are addressed below.

In Short

A strategist at the bank Barclays went on a Bloomberg podcast and said most big investors still want to own AI stocks, even though some smaller companies that rent out AI computers are now finding it hard to borrow money. He thinks those smaller firms are an early warning sign, and that many investors are ignoring it because they are so excited about AI. For anyone who buys ads from Google, Meta or Microsoft, it matters because those companies pay for much of the AI build-out with money their ad businesses earn, and the stock market now depends heavily on that spending continuing.

The episode, and a Barclays conflict

The video, titled "What Wall Street Really Thinks About the AI Trade Now," sits on the Bloomberg Podcasts YouTube channel, which has 604,000 subscribers. The video page dates it October 29, 2026; the recording date is stated on air. "So we're recording this on October 28th," one of the hosts said. It runs for nearly 40 minutes across 11 chapters, from "The Laws of the Economy" at the start to "Breaking Earnings News: Alphabet and Microsoft" at 32:00.

The description frames the conversation around "an increasing sense that the laws of the economy barely apply to AI." It introduces the guest as Alexander 'Alty' Altmann and promises a discussion of "the KOSPI-fication of the S&P 500," historical precedents for AI capital spending, and "if AI is really immune to interest rates." Altmann had appeared on the show before; his previous appearance, the hosts noted, was about leveraged exchange-traded funds.

One disclosure shapes how parts of the episode should be read. Midway through, Weisenthal noted that Barclays is involved in Anthropic's listing. "Barclays is I know managing the anthropic IPO so or as, you know, involved in this," he said, adding that the hosts would not put questions about Anthropic to Altmann. "You know, we understand the rules here." Barclays is one of the 22 underwriters named on Anthropic's S-1, which Goldman Sachs and Morgan Stanley lead. Altmann did not discuss Anthropic. Everything said about the company in the episode came from the hosts.

"We're not stopping this train"

Asked whether a consensus view on AI exists, Altmann said it does, and that it is bullish. Sample investors across the US, Europe and Asia, he said, and ask them "gun to their head, do you want to still be bullish or bearish the AI story, you're going to get an overwhelming positive answer." He put it plainly: "We're not stopping this train. That's still the consensus like to be very clear."

That consensus is exactly what worries him. Altmann referred to the familiar chart of a speculative bubble's life cycle, in which the first decline after the peak is labelled denial. He said he would not claim the bubble had burst. "I know there's commentary out there that actually has said that," he said, "but I if you were to sort of juxtapose those two narratives, we're in the denial phase right now because everyone is still extremely bullish."

The hosts tested the bullishness against valuations. Korean memory-chip stocks such as SK Hynix trade at around three times earnings, Altmann said - a figure, he noted, that requires no research to find. A price-earnings ratio that low signals that investors think memory profits are at a cyclical peak and will not last. The old rule, voiced during the exchange, is not to buy cyclical companies at peak earnings. Low multiples on the companies supplying the boom sit oddly beside a consensus that the boom will continue.

Neoclouds at the periphery

The core of Altmann's argument concerns neoclouds, the specialist companies that build and rent out AI computing capacity. He described them as businesses "effectively in some way, shape or form mandated to build compute, which they will then sell to predetermined customers," some with contracted buyers and some building on speculation.

"These are tip of the spear," he said. "These businesses don't have operating cash flow today or minimal operating cash flow today in many cases." That makes them, in his words, "the riskier part of the financing aspect of the AI spectrum." He acknowledged the counter-argument that neoclouds are really property companies in disguise and less risky than they look, and set it aside: compared with hyperscalers such as Google and Amazon, whose established businesses he called "cash volcano" operations, they belong to a different category.

He then made a comparison he was careful to qualify. Treat neoclouds as the periphery and hyperscalers as the core, he said, and draw a parallel to the global financial crisis - "not directly, because I'm not trying to make some sensationalist statement here but draw parallel to the GFC." The 2008 crisis did not begin with the largest banks. "It started with subprime vehicles 12, 18, 24 months beforehand."

The evidence he offered was recent. "And so what I find most interesting is that there is evidently a financing problem within neo clouds. Right? That is evident." He pointed to Cipher, a bitcoin miner that has moved into AI data centres. Cipher Digital announced an $850m offering of 0% convertible notes due 2032 on October 13, and its shares closed down 12.77% that day, according to The Motley Fool. Altmann said that deal, not anything CoreWeave itself said, set off the sell-off in CoreWeave, the best-known neocloud. "It was because cipher, another neo cloud try to get try to tap financing and in doing so materially hurt their own share price," he said. Investors "put two and two together," and the selling spread.

One of the hosts added a second trigger: an interview Brannin McBee, a CoreWeave co-founder, gave to the Financial Times, in which he talked about "unlimited amounts of borrowing." Benzinga's report on the interview says McBee was referring to capital from insurers. CNBC reported on October 14 that yields on CoreWeave's bonds were nearing 15%.

Altmann's point was that the stress had been visible in bond markets well before it reached equities. Bonds backing data centres, including those tied to Meta - most likely the Beignet Investor bonds that financed Meta's Hyperion campus in Louisiana, a $27.3bn, 23.6-year deal according to IFR - "had already been widening for months prior to that," he said. "So I guess the point I'm trying to make is, is that there's ripple effects in financing that start in in the periphery. And we're seeing that."

A survey that should say 100%

Barclays has been putting the question to clients directly. "And we literally just conducted a survey with, you know, group of investors," Altmann said. Of the responses received so far, about 60% answered yes when asked whether financing will be a problem for AI for the foreseeable future.

"Only 60% said yes," he said. "By definition, what's happened over the past couple of weeks mean that answer should be 100% right?" To him the result was "evidence of like almost a denial of a reality that's already here."

The survey numbers need care. Altmann described the response pool as incomplete and did not say how many investors took part. He also gave a second, inconsistent reading in the same passage, saying "let's just call it just over half think or just under half think" there is no financing issue - which would put the yes share at roughly 50%, not 60%. Barclays has not published the survey.

One of the hosts summarised the shift as a move from a period of apparently "unlimited money," in which AI borrowers could tap investment-grade and loan markets for whatever they needed, to a more selective one. Altmann agreed. His team travels to meet investors, he said, and had heard an almost universal view that AI was immune to high interest rates. "Yeah, evidently is not true." He cited recent results - the transcript renders the company as "metal," most likely Meta - as showing that financing has limits. Meta's third-quarter figures, published on October 21, showed $34.91bn of capital spending and negative free cash flow of $2.12bn.

2026 breaks a 30-year pattern

Over the summer, Barclays looked at net issuance across corporate investment-grade and high-yield bonds, Treasuries and equities, going back to about 1995. For each market and each year, the team measured net issuance as a share of that market's size and plotted the result against interest rates.

"And what we found was there was, unsurprisingly, a very strong relationship. When interest rates went down, overall issuance went up. And vice versa," Altmann said. Then came the exception. "2026 is completely shattered," he said - the year falls far outside the historical relationship. The hosts, who had seen Barclays' chart plotting the two-year Treasury yield against net issuance, put it in plain terms: borrowers are selling far more debt relative to benchmark rates than at any point in the data.

Altmann called it "kind of biblical." The dot-com era is the obvious comparison, but he said it does not fit: equity issuance was heavy then, interest rates were high and the US government was running a surplus. "So today is very unique."

That uniqueness matters because AI is not the only borrower. In Altmann's words, "there is this competition for capital. It is real." "There are other verticals that need to tap financing to and there is a finite amount." Barclays forecasts net investment-grade and high-yield issuance this year at $1.1 trillion. "And we just throw around trillions as though it's just this casual number that everyone's got," he said. "But like there is limitations, right."

The $30 trillion question

The hosts put the bull case to him directly: someone will pay for the GPUs, demand is rising, and the hardware will not be worthless. Altmann did not dispute it in principle. "There's nothing wrong with it in terms of build it and they will come," he said. His objections were the canary in the coal mine - "when you start to see cracks or canaries in coal mines or however you want to characterize it, you should probably pay attention" - and revenue.

On revenue, he pointed to a Barclays note published at the end of August under the title "Placing your AI chips more carefully." He set it against a claimed total addressable market of about $30 trillion - "of like the entire universe, not just Earth." The transcript renders the company as "space"; the reference is most likely SpaceX, whose IPO filing put its total addressable market at $28.5 trillion, according to Pulse 2.0. Altmann set that against a statistic he credited to Bloomberg's Cameron Crise: the cumulative revenue of Microsoft and Apple since 1990, adjusted for inflation, is "not even $10 trillion."

Then the arithmetic of the wider market. The S&P 500's companies generate about $17 trillion of revenue, Altmann said, with $11 trillion to $12 trillion of expenses, leaving net profit of roughly $5 trillion to $5.5 trillion. Claims about how much of that will flow to AI need to fit inside those totals. And the spending data he watches is not accelerating. "Like we all look at the Rap data or the open router data," he said - meaning Ramp's corporate spending data and the OpenRouter model marketplace - "and we can clearly see that that token expenditure has kind of begun to level off a bit, right, at least near term."

Weisenthal questioned how much OpenRouter's figures on tokens processed through its marketplace reveal, arguing that they capture "a specific slice of people who are usually looking for cheap Chinese models." He pointed instead to Anthropic, which has filed with the Securities and Exchange Commission. "But the top line numbers for anthropic look unbelievably good," he said. "That number is going straight up." Anthropic's filing shows $16.3bn of revenue in the first half of 2026 and no plan to sell advertising, as well as $269bn of compute commitments to Amazon, Google and Microsoft. The counterpoint, voiced on the show, was the scale gap: "billions in earnings versus valuation at trillions."

PPC Land has tracked both sides of that gap. Ramp data cited by Apollo's Torsten Slok showed that the top 10% of firms account for 99.5% of model serving and inference spending, with Slok finding no financial payoff outside the technology sector. OpenRouter's own data, meanwhile, showed AI agents using nearly five times more tokens per task than human users, with agent usage on the platform up fourteenfold over six months.

Open weights and the phone in the pocket

The hosts then asked about the debate between open-source models and the most advanced proprietary ones: open models lag on benchmarks, but by less than before, and cost far less to run. Altmann said the data here cut against the bullish story. Open-weight models, by his account, have gone from about 20% market share "to anything up to about 80% market share." He did not say which measure or source that figure comes from. He added that token prices have fallen sharply every year, calling tokens possibly among the fastest-deflating assets in history. When the hosts pointed out that GPU rental rates have been rising, he agreed: "They have yes, absolutely. And so there is clearly a two way debate."

The question he finds most important is not open weights alone but open-weight models run locally, on the user's own device. He calls it "being long or short human creativity." His analogy was the iPhone. In 2007, he said, a person told that a phone 20 years later would carry orders of magnitude more computing power with a battery of the same size would have answered that it was impossible. "You'd need an iPhone the size of this table, right?"

By the same logic, he questioned whether everyday tasks will keep requiring Nvidia's Hopper or Vera Rubin chips "in an Ohio data center" - tasks like an agent handling email or writing a complaint to a hotel. He mentioned Nvidia's Nemotron models and a multi-billion-parameter model from Meta that runs on a laptop. The direction of travel, he said, is along an efficiency curve, and that brings the revenue question back. Frontier models will matter for problems such as drug discovery. Most users will not need them for that. "We're probably not going to be discovering the new biotechnological drugs," he said.

Weisenthal described a term he had heard from someone in San Francisco: the "frontier sink," meaning the industries that need intelligence at "PhD level and beyond technology." His candidates were biotechnology, semiconductor design itself and quantitative trading. Beyond those, he said, "there might just be a lot of industry for which there is a certain level of intelligence. That is enough."

PPC Land has documented one such switch. In October, Ecosia dropped Mistral for open-weight models, with its chief executive saying costs fell by about half. The opposite pressure also exists: Google said it remained capacity constrained after Gemini 4 Argon processed almost 25 trillion tokens in its first 72 hours.

The consumer layer

What would make Altmann more constructive? The answer, he said, is a "killer consumer application layer, because at the end of the day, the consumers, everything, it's the economy." Business spending will not be enough on its own: "enterprise revenues, I just don't think is going to plug that gap, really," he said. AI revenue may reach the required level eventually, "but it can't get there in the time frame that the market wants to see the ROI."

And consumers may not need the most capable models to get there. "The consumer probably doesn't need frontier for the time being to do all of these things," he said. His "big mantra" is that many of the largest technological shifts come not from new inventions but from "changing the application of existing tech." He expects an agentic workflow of roughly average human capability - "100 IQ" in his shorthand - to become close to ubiquitous within 12 to 18 months. "And I think that's probably where you're really going to get the next upswing in AI."

Weisenthal supplied a test case from his own week. With his wife travelling and his children's schedules scattered across many emails, he used ChatGPT to ask where he needed to be in the next hour and when to collect his daughter from rehearsal. Altmann's response went straight to monetisation. "So what what's that worth to you a month," he asked. "It matters for the revenue profile, right?" A better calendar interface, it was suggested in reply, might have solved the same problem.

That question - what an AI assistant is worth to its user, and who pays - remains open across the industry. OpenAI reported in August that ChatGPT advertising had passed a $1bn annualised run rate. Anthropic's S-1 contains no advertising model. On the business side, KPMG found that 49% of organisations had narrowed, delayed or paused AI agent rollouts when costs outran value.

What could break the spell

Asked what would shake investor conviction, Altmann was specific. "Off the top of my head, I would say it would be some level of CapEx being reined in by one of the hyperscalers." A client had asked him that very morning whether this week's results might bring such a signal. "And the results this week and famous last words, it's probably still premature," he said. His suspicion was that "we have to wait till the new year" before enthusiasm cools.

Proximity to the technology, he said, shapes belief: "the closer you are to the epicenter, the more I would say that both the companies and investors are all bought in on that narrative," Altmann said, describing something "almost like an echo chamber" on the West Coast.

He credited Google with starting the latest wave. "Google kind of set this all off again for the next wave," he said, because at its previous results it said capital spending would be significantly higher. In July, Alphabet raised its 2026 capital expenditure guidance to $195bn-$205bn from $180bn-$190bn, after second-quarter capex of $44.9bn.

Oracle offered the opposite signal on the day of the recording. One of the hosts noted that Oracle had said at its investor day it would limit additional fundraising to about $20bn. Oracle set that cap on new funding from stock and bond investors for fiscal 2027 through 2030, excluding refinancing, according to StockTitan. The stock, the hosts said earlier, was up 12%. "But again like the reaction so far is this is a good thing," the host said, and the build-out can continue "as long as people are raising a reasonable amount of capital." In July, S&P Global Ratings had cut Oracle to BBB-, forecasting a fiscal 2027 free operating cash flow deficit of nearly $42bn and noting that OpenAI accounted for roughly half of Oracle's $638bn of remaining performance obligations.

For the market as a whole, Altmann expects churn rather than direction: a broad index that goes nowhere while money rotates between AI winners and losers, since what helps the hyperscalers can hurt memory makers and the reverse. "I call it like a knife fight in a phone booth," he said. For long-short funds, "it's a nightmare."

Why volatility gauges look calm

The hosts raised a paradox: anxiety about CoreWeave and its peers is everywhere, yet realised volatility is low and the VIX is not high. "We're seeing volatility everywhere except our measures of volatility." Altmann said it is the question he is asked most often, and gave a two-part answer.

The structural part is the growth of multi-manager hedge funds. "We've got 2020 the AUM deployed in multi manager funds was about $300 billion," he said. "And today it's in excess of $1 trillion." He explained why higher rates suit the model. A fund posts only margin on its long and short positions, using leverage from prime brokers, and can hold the bulk of its capital in instruments paying a healthy coupon while trying to generate returns on the gross positions. "That's the business model," he said.

The trouble is that the money has more than tripled, plus leverage, while the number of securities available to trade "has stayed roughly the same or actually shrunk over time. So that creates crowding out." Funds try to neutralise common risk factors - value, quality, size, growth - using commercial risk models; the transcript's "Baron" and "Acoma" most likely refer to the Barra and Axioma models. But because everyone does the same, Altmann said, "all you're creating is different kinds of crowding." Since much of that money is indifferent to the index, the turbulence shows up as violent rotations between factors rather than as index-level volatility.

The technical part is that nobody has managed to model an "AI factor." Investors ask him how Barclays has mapped it. "And the answer is no, because the effect is changing all of the time." At the start of 2026, betting against software stocks was part of the AI trade; now, he said, software is trading close to 12-month highs. As the narrative moves, so does the factor, which complicates hedging. Over the past three years, he said, being short AI has probably not been a lucrative trade.

Earnings arrive mid-recording

While Altmann was explaining factor crowding, Alphabet and Microsoft reported. Weisenthal read the headlines from the Bloomberg terminal: "Alphabet down 7.9%. Revenue misses estimates. Microsoft jumped 6.4% after Q1. Cloud revenue tops estimate." The moves were in extended trading, minutes after the releases. "So you get two gigantic stocks. They both have very violent reactions," Weisenthal said. "But if you're the index investor on net, looks like it all kind of washes out."

He then asked Altmann, who had no terminal in front of him, to guess Microsoft's Azure growth. "The bogey was I think 47% growth," Altmann said. The Bloomberg consensus estimate was 44.8%, Weisenthal said; Microsoft reported 48%. Google Cloud revenue came to about $30bn and grew 103%. "All these numbers are insane," Weisenthal said. Alphabet's results release put Google Cloud revenue at $30.7bn, up 103% from a year earlier.

Altmann's response returned to his core-and-periphery split. A company with a strong legacy business - "a cash volcano" - can afford to embark on "the largest, you know, CapEx experiment, really in the history of America." Taking current forward capex estimates at face value, he said, the spending comes to about 12% of GDP, "like two Netherlands." He did not say over what period or which economy's output the figure refers to, and it should be treated as a rough comparison. He searched for precedents - the Louisiana Purchase, the railroads, the pyramids, the Great Wall of China - and found none that matched the scale in so short a time.

That concentration is what the episode's description calls KOSPI-fication, a reference to South Korea's benchmark index, where memory-chip makers carry the heaviest weights. "It's actually it's about two thirds from semiconductors. And then the broader AI ecosystem gets you to around about 75%," Altmann said of next year's S&P 500 earnings growth.

The hosts' verdict

After Altmann left, the hosts drew their own conclusions. "I do think the key thing that's emerging is this differentiation idea, and the idea that we're moving from this sort of capital, abundant environment into an area where people are perhaps a little bit more selective," Alloway said.

Weisenthal went after the notion that rates no longer matter for an industry growing this fast. "But that is a lie that investors tell themselves, right? At least at the margin, at least at the tip of the sphere spear," he said, singling out leveraged companies such as the neoclouds. Alloway questioned the foundation of hyperscaler earnings power: the premise that other companies will buy AI services to raise their own productivity and profits. "But like, we're not actually seeing that materialize in their earnings numbers just yet," she said. Meanwhile, the whole index becomes more dependent on AI.

Two further observations closed the episode. Any large fund that made money in the past year, Weisenthal argued, did so in some way connected to AI. And one of the hosts recounted a recent conversation with a Stanford computer science student, for whom even discussing the possibility of an overabundance of GPUs meant "you just don't get it."

Why this matters for marketers

Altmann's split between core and periphery maps closely onto the advertising market. The "cash volcano" companies he described as the core are, to a large extent, advertising businesses. Meta's third-quarter ad revenue was $63.1bn of $64.6bn total, and the company has guided 2026 capital spending to $135bn-$145bn. Meta's head of infrastructure has put 2026 AI spending at well over $100bn, with the Hyperion campus in Louisiana targeted at 2GW by 2030. Alphabet has guided to as much as $205bn this year. Microsoft reported $41.0bn of capital spending in a single quarter in July, when Azure grew 43%; this week Azure grew 48%.

In Altmann's framework, the advertising budgets marketers spend on search, social and video are part of what lets the core keep spending while the periphery struggles to borrow. The warning sign he named - a hyperscaler reining in capital spending - would come from the same companies whose ad platforms now run on AI ranking, bidding and creative systems built on that infrastructure.

The dependence runs in both directions. Barclays itself estimated earlier this year that OpenAI and Anthropic account for 13% of Microsoft's cloud revenue this year and 18% next year, while UBS put the two labs at more than 48% of Google Cloud revenue in 2027. Cory Doctorow has set about $50bn of annual AI revenue against roughly $1tn of spending, and Ray Dalio has said his bubble indicator puts AI level with 1929 and 2000.

Altmann's revenue argument also bears on the consumer layer that marketers compete for. He expects the next upswing to come from consumer applications built on cheaper, good-enough models, possibly running on the device itself. Whether those applications are paid for by subscriptions, as his "what's that worth to you a month" question implied, or by advertising, as ChatGPT's ad business suggests, is not settled. And by his own survey, a large minority of investors do not yet see financing as a problem at all.

What the automatic transcript gets wrong

YouTube's automatic transcript of the episode does not identify speakers; attributions in this article follow names used in the conversation and the flow of questions and answers. It also mishears several names and numbers. The most likely readings:

  • "Nio cloud," "NIO clouds" and "Neo clauses" are neoclouds.
  • "core weaves," "core we have," "core we've," "Cool. We've" and "Kobe" are CoreWeave.
  • "Kevin Walsh," in a line about the Fed press conference, is most likely Federal Reserve chair Kevin Warsh.
  • "Brandon McBee" is Brannin McBee of CoreWeave, and "the Ft" is the Financial Times.
  • "the ironies" is most likely IREN, another former bitcoin miner turned AI data centre operator.
  • "the bennies of this wall, the, you know, the the metal one" most likely refers to the Beignet bonds tied to Meta's Hyperion campus.
  • "the metal results" most likely means Meta's results.
  • "GCB banks" is most likely G-SIB banks, the global systemically important banks.
  • "space time" and "space is addressable" most likely refer to SpaceX.
  • "Cameron Crease" is most likely Bloomberg strategist Cameron Crise.
  • "Rap data" is Ramp data; "open root" is OpenRouter.
  • "open way" and "open wait" are open-weight models; "a genetic workflow" is an agentic workflow.
  • "Nema, Tron" is Nvidia's Nemotron; "Mets" is Meta; "hoppers" are Nvidia Hopper chips.
  • "Baron" and "Acoma" are most likely the Barra and Axioma risk models; "star factors" are style factors.
  • "leverage chops," "pile of jobs" and "punch up" are most likely pod shops, the multi-manager hedge funds.
  • "Cosby" and "cost bifurcation" are KOSPI and KOSPI-fication, as the video description confirms.

Some passages remain unclear. In the main conversation, the exchange about the laws of the economy no longer holding reads "Hold on to 25 and 2020"; the cold open at the start of the video, which replays the same moment, reads "Not so much in 2025 and 2026." The transcript says Oracle would limit fundraising to $20bn "for three weeks or so"; Oracle's cap runs from fiscal 2027 to 2030. Altmann's comparison of token prices, "a fraction with a 1/100 versus credible token," cannot be reconstructed with confidence, nor can his description of hyperscalers having "huge IBM" in their businesses. The chapter title spells the guest's surname "Altman," while the description and the hosts' introduction use "Altmann."

Timeline

Summary

Who: Alex Altmann, global head of equities tactical strategies at Barclays, interviewed by Tracy Alloway and Joe Weisenthal on Bloomberg's Odd Lots podcast. Barclays is one of 22 underwriters on Anthropic's IPO, and the hosts did not ask Altmann about Anthropic.

What: Altmann said the investor consensus on AI remains strongly bullish but is in a "denial phase." He argued that financing stress at neoclouds such as CoreWeave and Cipher is an early warning, that only about 60% of investors in a preliminary Barclays survey see AI financing as a problem, that 2026 net issuance has broken a relationship with interest rates going back to 1995, and that about 75% of S&P 500 earnings growth next year will come from the AI ecosystem.

When: The episode was recorded on October 28, 2026, the day of a Federal Reserve decision, Oracle's investor day and Alphabet's and Microsoft's quarterly results; the YouTube page dates it October 29, 2026.

Where: On the Bloomberg Podcasts YouTube channel and podcast feeds, with discussion covering US credit and equity markets, Korean memory stocks and AI companies in the United States and China.

Why: The episode shows how a major bank's strategist reads the gap between AI spending and AI revenue, and where he expects the first cracks. For marketers, the companies Altmann places at the safe core of the AI build-out - Alphabet, Meta, Microsoft and Amazon - fund much of that build-out from advertising, and the consumer applications he says the trade now needs are the same surfaces where advertising, subscriptions or both will have to pay for AI.