Why AI Is NOT the Next Dot-Com Bubble with Andreas Steno and Mikkel Rosenvold | Macro Mondays

Anthropic

Across the shows: 51 speakers in other episodes, 37 bullish and 38 bearish calls, past 30 days

• Andreas Steno · Steno ResearchBearish

  Anthropic is openly short of compute and has lost enterprise momentum since June, mostly a capacity issue.

  Steno says Anthropic is very explicit that it needs more compute and knows it does not have enough. He has argued for a quarter that Anthropic lost momentum after June, which he attributes mostly to capacity rather than demand. Anthropic has forward compute deals with Alphabet and Amazon, so those hyperscalers must deliver the capacity. He doubts the market will finance that build-out without convincing returns, which makes the constraint the central question for the coming earnings calls.

  “Anthropic is very explicit when it comes to its compute capacity going forward. They need more and they are very well aware that they don't have enough.”at 0:46

• Andreas Steno · Steno ResearchBullish

  Anthropic's margin improved from -255% to about -5%, so it is near cash-flow breakeven and may turn positive this year.

  From the filing, Steno sees Anthropic's margin move from about minus 255% a year earlier to about minus 5% at the end of the first half. Extrapolating that into the third quarter puts the company very close to cash-flow neutral or a cash-flow-positive reading. He says he would not be overly surprised if it turned cash-flow positive this year.

  “They've basically gone from a margin of minus 255% a year ago to a margin of minus five, something like that. So if you extrapolate that into the third quarter, we're very, very close to cash flow neutral... I wouldn't be overly surprised if they turned cash flow positive this year.”at 2:09

• Andreas Steno · Steno ResearchBullish

  Anthropic's revenue growth outpaces compute costs, showing it can pass costs on to customers.

  In the filing, Steno sees Anthropic's revenue up about 15x in a year while compute costs rose about 4x. He reads this as proof that the company passes its costs on, and more, to the average client. He notes it has struggled to pass costs to all API clients, but overall it manages to pass them on sufficiently. Its enterprise momentum was instead hurt by running out of compute.

  “I think revenue has 15x in a year, while compute costs have 4x. It actually tells you that they're able to pass on the costs and more to the average client.”at 12:57

Meta Platforms

Across the shows: 44 speakers in other episodes, 43 bullish and 20 bearish calls, past 30 days

• Andreas Steno · Steno ResearchBearish

  Meta's new equity category to fund capex is light dilution of existing holders, explaining the stock's fall.

  Steno says Meta created a new equity category to fund its capex, which amounts to light dilution of existing shareholders. He says this explains why the stock took a beating. He would be surprised if Alphabet, Amazon or Microsoft made a similar move, and suggests they are more likely to use third-party neocloud providers.

  “Meta basically decided to come up with a new equity category to fund the CapEx. So it was some sort of light dilution of existing equity holders.”at 8:38

AI chips & semiconductors

• Andreas Steno · Steno Research

  GPU rental prices are rising and beating depreciation schedules, so the AI build-out returns look fine.

  Steno shows that the rental price of H100 GPUs, mostly installed in 2022 and 2023, is up about 30% this year, and that the first Blackwell series, the B200, has more than doubled. Consensus spreadsheets assume falling GPU prices, so he calls this bizarre. He says capex is depreciated over about five years, and rising rental prices mean every GPU installed since 2020 has beaten its depreciation profile by miles. He takes this as a sign the market is short of compute, which makes the return on investment fine if the trend continues.

  “They're up 30% in rental prices this year. Look at the [B200], the first Blackwell series. It's more than doubled this year.”at 10:36

• Andreas Steno · Steno Research

  Market prices the AI build-out as over; forecasts of chip growth falling to zero next year are debatable.

  Steno argues the market has priced in the end of the AI build-out. Either the build-out ends, in which case AI-related stocks should not rally, or it does not end, in which case AI-related stocks should rally hard. The assumption now is that chip growth will slow to zero next year and turn negative into Q2, which he finds highly debatable.

  “I'd argue that the market has basically priced now that the AI build out is over... the assumption now is that chips growth will basically slow to zero next year... which I find highly, highly debatable.”at 18:55

AI pricing & unit economics

• Mikkel Rosenvold · Steno Research (Real Vision)

  Falling compute costs could accelerate AI adoption through the Jevons paradox.

  Rosenvold argues that if compute prices drop, it is an issue for the providers, but it would trigger the Jevons paradox discussed last year. Cheaper compute could then speed up AI adoption, so falling costs would be net positive for the broader AI theme.

  “If compute prices drop, that's obviously an issue for their providers, but that would only trigger the Jevons paradox that we talked a lot about last year. Cost of compute falls, that could even speed up the AI adoption.”at 14:10

• Andreas Steno · Steno Research

  Quality-adjusted token prices have fallen about 95% over two years, so AI deflation is real.

  Steno argues that official inflation measures adjust prices for quality. A token is far more valuable now than a year ago because models are much better, so people pay roughly the same nominal price for much more output. On that basis he estimates the price of a token is down about 95% over the past two years.

  “One thing is the nominal price you pay for an iPhone or pay for a token, but it will always be adjusted by the power of that iPhone and the power of that token... I'd argue that the price of a token is down 95% or so over the past two years.”at 14:46

AI bubble & tech valuations

• Andreas Steno · Steno Research

  The AI build-out is a very weak dot-com analogy: capacity is full, prices rising, and builders earn revenue.

  Steno compares the AI build-out with the dot-com era on utilization, price of capacity and revenue of those building it. Today there is effectively full capacity or a shortage relative to demand, and capacity prices are rising, the opposite of the bandwidth price collapse in the dot-com era. Anthropic and OpenAI generate substantial revenue, which he says is unlike pets.com-style businesses. He calls the analogy very weak, while adding that the credit market is stuck in a dot-com-like mood.

  “The comparison between the AI build out and the .com is very, very weak on a range of parameters... on all three parameters... The price of that capacity is going up, which is in sharp contrast to what happened with the price of bandwidth back in .com.”at 2:50

Credit markets

• Andreas Steno · Steno Research

  Bond market distrust of AI build-out financing is a serious risk until hyperscalers prove returns on capex.

  Steno argues that the bond market's view is decisive: if it sees AI as a credit risk, AI is a credit risk until the market is convinced otherwise. The build-out needs capital, and hyperscalers such as Amazon and Alphabet can no longer fund it from free cash flow, so they have issued bonds and tapped markets. He expects them to deliver strong returns on already-invested capex. Credit stress is not a major issue while earnings deliver, but if earnings disappoint the skeptics in the bond market will grow more concerned.

  “If the bond market finds AI to be a credit risk, then AI is a credit risk until you convince the bond market that it isn't.”at 6:21

US rates & the Fed

• Andreas Steno · Steno Research

  The Fed is blinded by the AI build-out while the rest of the economy struggles with restrictive rates.

  Steno says the Fed will not admit rates are restrictive because the AI build-out masks weakness. Outside AI, construction activity is down, and construction is typically a gauge of how interest rates affect the economy. Outside AI, data centres and power are the only areas getting capital, while the rest suffers from scarce capital and high bond yields.

  “When the Federal Reserve is not willing to say that interest rates are restrictive, I think they're kind of blindfolded by the whole AI build out because everything else than AI is down in terms of construction activity.”at 17:31

US stock market outlook

• Andreas Steno · Steno Research

  Broad equities could rebound sharply if the Fed acknowledges tight financial conditions, a short squeeze.

  Steno argues that beneath the surface many equities are suffering, not only AI, because the economy cannot cope with current interest rate levels. Rotation has occurred, but broad equities have been hit. If the Fed acknowledges tighter financial conditions, or even hints the hiking cycle will not last long, the broad market could rebound sharply. He describes the market as one message away from a short squeeze.

  “If the Fed starts to acknowledge these tighter financial conditions... I actually think that the broad equity market will rebound like crazy because, you know, beneath the hood... a lot of equity suffer. It's not only AI because the economy cannot cope with interest rates at these levels.”at 17:57

Emerging markets

• Andreas Steno · Steno Research

  Brazil's selloff is not over: Bolsonaro may refuse the result and Trump could impose tariffs.

  After Lula's win, the real fell about 10% and Brazilian equities dropped sharply. Steno does not think this is over, because Bolsonaro is not willing to accept the result. He also notes the right-wing shift in South America has been cheered by the White House, so Trump could get involved with extra tariffs on Brazil. He says he would not rule that out.

  “I don't think this is over. No. Especially since, you know, Bolsonaro is not willing to accept the result, first of all.”at 21:14

Middle East & Iran

• Mikkel Rosenvold · Steno Research (Real Vision)

  Israel's election likely yields a hung parliament, and even a Eisenkot win would not shift foreign policy.

  Rosenvold expects a hung parliament because neither Netanyahu nor Eisenkot will easily build a majority. A coalition between them looks tricky after harsh words in the campaign, and Eisenkot may need Arab parties. Even if Eisenkot forms a government, regional policy should not change much: possible tweaks on the West Bank and Gaza but no major shift on Iran, since Eisenkot backed hitting civilian infrastructure in the war.

  “We are expecting a hung parliament... I don't expect, and that's one thing a lot of people get wrong. Even if Eisenkort is able to form a government, we shouldn't expect a flip flop or change of course in the regional policies.”at 22:29

Tariffs & trade policy

• Andreas Steno · Steno Research

  The EU's new import curbs on Chinese goods are long overdue after years of refereeing without enforcement.

  Steno says the EU let Chinese EVs flood its market, and then gave them tax-free treatment in many countries, which gave China an unfair advantage over locally produced EVs. Chinese goods also benefit from subsidised air freight in Denmark, since China is treated as an emerging market under Danish law. Everyone else is already in the arena, and the EU has spent three or four years acting as a referee that nobody listens to. He thinks the move signals a wave of change within the union.

  “I personally impatiently waited for a message like this from the European Commission... the EU... has basically spent the last three or four years trying to act as a referee telling everyone else, oh, you cannot do that.”at 24:30