DreamWorld

👀 Competition is taking its toll

OpenAI cut token prices in half when the wave of Chinese open models hit. Some even link this to the July chip sell-off. Then it postponed its IPO.

Now Anthropic’s token prices have also been cut by 50% over the past month (though they’re still higher than OpenAI’s).

Users should be happy, but the market has more and more questions and fewer and fewer answers. Expectations are clearly too high.

🙄 They say it’s not impressive

Anthropic showed investors its numbers ahead of an IPO

According to the FT, Anthropic’s revenue grew 46% in the third quarter to $16.8 billion. The company has been adjusted-profitable for the second consecutive quarter, has positive cash flow, and generates $1.70 in revenue for every $1 spent on computing.

But there’s a problem with the pace. Revenue more than doubled from the first to the second quarter, but growth has now slowed. SemiAnalysis estimates an annual run rate of $74 billion, although the company’s investors were talking about $100+ billion by year-end this summer.

Competition is taking its toll. Both Meta and Microsoft are switching to their own models, OpenAI is once again a strong competitor, and prices are falling because of low-cost Chinese players. Anthropic also restricted competitors that trained their models on its responses. That accounted for 5–7% of revenue.

💬 Investors in our LifeHUGS community are skeptical. Meeting the forecast won’t drive growth, because it’s already priced in. If Nvidia had grown at this pace in 2023, its shares would still be at the same levels. The AI sector is expected to deliver an earnings surprise.

That’s a weak spot for an IPO with a $2 trillion target valuation, especially with interest rates rising and more than $500 billion in computing commitments ahead.

The broader impact is already being felt, because Anthropic’s numbers are affecting sentiment around chips.

📖 Zuckerberg says their AI is already ahead of everyone else.

In an interview with the FT, he used the big words “frontier”, and that the leadership is based on “some benchmarks” that Meta … chose itself. The launch was also spoiled by bugs with limits.

The previous Muse Spark in April was clearly lagging behind. So it’s too early to take his word for it.

More importantly, Meta is abandoning open models and starting to sell the best versions of the business, from $0.20 per million tokens to $200 per month for premium. This is the first attempt to make money on AI directly, and not just through better advertising targeting. However, in the corporate segment, Anthropic and OpenAI already have both contracts and a reputation, so it will take a long time to win back customers.

The real strength of $META is in distribution. Muse is still the most popular app in the US app stores, and the stock has gained almost 10% since the beginning of September.

Zuckerberg sees his future in “personal superintelligence” for ordinary people, while OpenAI and Anthropic are betting on office automation. For a company with billions of users, this is a logical choice, although how it will make money from it is still unclear.

📊 Bank earnings season got off to a mixed start.

▪️Dow futures +0.9%, Nasdaq 100 futures -0.7%.
▪️Goldman Sachs (GS) +4% in premarket trading. The bank posted its second-highest revenue ever, at $18.7 billion (+23%). Earnings were $17.01 per share versus $12.86 expected. Equity trading revenue rose 72%.

▪️JPMorgan (JPM) missed expectations. Earnings were $5.57 per share versus $5.93, due to a $658 million loss on securities and a decline in client activity. Shares fell 4%, but recovered their losses after the bank forecast fourth-quarter net interest income of $28 billion, above consensus.

▪️Citigroup (C) increased profit by 40% to $5.25 billion. Wells Fargo (WFC) reported a 6% increase in profit, but revenue came in below forecasts. Both banks reported stable consumer spending.

▪️UnitedHealth (UNH) +3% after a strong quarter and an increase to its annual forecast.

▪️Chipmakers' shares are under pressure after China's CXMT announced a new type of DRAM. Samsung -7%, SK Hynix -10%, KOSPI -5.5%. Micron (MU) -3%, with Nvidia (NVDA), Marvell (MRVL), and Sandisk (SNDK) also falling.

▪️In a Bank of America survey, 45% of fund managers named long chip positions as the most crowded trade.

👀Bank of America (BAC), Morgan Stanley (MS), and BlackRock (BLK) report tomorrow.

📊 The Economist on how households in developed countries are increasingly making money from stocks rather than real estate.

And that, for a moment, is a reversal of a twenty-year trend🫡.

According to the McKinsey Global Institute, of the $40 trillion increase in global household wealth in 2025, 57% came from stocks and just 15% from housing. Over the previous 25 years, real estate accounted for an average of 52%, while stocks made up 26%. In the US, according to Goldman Sachs estimates, stocks overtook real estate in the composition of household wealth for the first time since World War II this July.

There are several reasons.
▪️Housing is still very expensive, so young people can’t afford to buy it—and are turning to the stock market instead. JPMorgan says the share of 25-year-olds with investment accounts increased sixfold from 2015 to 2024.
▪️High interest rates are holding back home prices, and Moody’s Analytics expects annual growth of just 2.1% in the US, compared with 5% over the past decade.
▪️Meanwhile, governments are scaling back incentives for buyers and strengthening protections for renters.

At the same time, stocks are now valued at record highs, so wealth is simply shifting into a more volatile asset.

💰 On crypto’s mass adoption

Back in 2023, stablecoin cards were barely used. But in July 2026, spending on them exceeded $1 billion in a single month for the first time, according to Paymentscan data.
More than two-thirds of transactions were paid for with USDC and USDT, mostly for groceries, fuel, and taxis. In other words, they’re now being used for everyday purchases, not just speculation and transfers.

The mechanism still relies on the “usual suspects”—Visa and Mastercard—which convert digital dollars at the point of sale. Adoption is growing fastest in Latin America and Africa, where these cards can serve as a substitute for a dollar bank account.

Of course, the scale is still modest. The annualized rate is about $12.9 billion—very little compared with global payment networks—and around 38% of the volume is based on data provided by a single company...

Still, who’s making money from this besides those on the chart?
▪️Visa (V) and Mastercard (MA) collect a fee on every payment, though it’s not yet noticeable in their revenue.
▪️Coinbase (COIN) benefits because its blockchain, Base, has become the main platform for these payments.
▪️Circle (CRCL) earns interest on USDC reserves, so what matters more to it is how many stablecoins exist in the first place, not how much people spend with them.

🕯 They say that in a situation like this, sharp moves are unavoidable

Over two months, the Dow fell 5.7%, while the Nasdaq rose 4.2%. A gap of 9.9 percentage points occurs in fewer than 5% of cases since 1971, according to Mark Hulbert (Hulbert Ratings).

Historically, after similar gaps, the S&P 500 gained an average of 3% over three months, nearly 7% over six months, and about 9% over a year. When the Nasdaq lagged the Dow by just as much, however, the annual return was only around 2%.

Still, the biggest gap—over 40 percentage points—came in March 2000, at the dot-com peak. After that, the S&P 500 lost nearly 50% by October 2002. So Hulbert sees this as a signal, but not of direction: rather, it points to a heightened chance of a sharp move either way.

Right now, the gap between the Nasdaq and the Dow is normalizing. And Korea’s “party” is coming to an end.

💡Goldman: AI models don’t matter

They compete every quarter for the title of best.
But a lasting advantage comes from owning computing capacity. There isn’t enough of it, and it’s not just a matter of chips. Data centers lack electricity. Of the roughly 16 GW of capacity planned in the U.S. for 2026, only about 5 GW is under construction. According to Wood Mackenzie, more than two-thirds of the capacity data centers need will never come online.

This creates a paradox: the factors holding back the construction of new data centers make those already built even more valuable. Large cloud companies have invested the most in them. That’s why, in Goldman’s view, the market is right to value them more highly than others.

Last week, hedge funds sold tech stocks en masse following news about OpenAI’s revenue. Goldman advises investors to focus not on whose model is currently ahead, but on who owns the computing capacity.

🇪🇺 European investors are increasingly lending money to American tech giants for data centers.

Hyperscalers (Microsoft, Amazon, Alphabet, Meta) have raised $48 billion in bonds denominated in European currencies this year. That’s three times more than they raised in all of 2025. They account for 22% of investment-grade bond issuance in francs, 10% in pounds, and 3% in euros.

Why is Europe so willing to buy? Local companies mostly have A or BBB ratings, while hyperscalers are rated AA- or higher. They also issue long-dated bonds of 10+ years, which are scarce in Europe apart from government bonds. Britain is also shortening the maturities of its borrowing. So pension funds and insurers get reliable long-term assets they had been lacking.

This has not yet crowded out other borrowers. European companies are issuing bonds just as actively as last year, and risk premiums remain low.

According to Hugh van Steenis (a British financier and banking-sector analyst), hyperscaler borrowing will rise by another 25% in 2027, and their share of European bond indices will only continue to grow.

👀 Nvidia earns 55,60$ in profit for every 100 $ in revenue, the highest rate among the 30 largest companies on the Fortune Global 500 list.

Big Tech dominates the top of the rankings, with Microsoft, Alphabet, and Meta keeping more than 30 $ in profit for every 100$ in revenue. It doesn’t get much better than that...

On the other hand, several of the world’s largest retail, healthcare, and energy companies keep less than 5$ for every $100.

*This chart ranks the world’s 30 largest companies by the amount of profit they earn for every $100 in revenue, based on data from the Fortune Global 500. Profit is measured after taxes, extraordinary credits or charges, changes in accounting, and noncontrolling interests, but before preferred dividends are paid.

📈 The market won’t wait

While the Senate once again stalls on the CLARITY Act, CFTC Chair Michael Selig has announced the first comprehensive set of rules for crypto exchanges. Instead of getting licensed separately in every state, an exchange will be able to voluntarily opt into a single federal oversight regime, with proof of reserves, customer fund protections, and legal margin trading.
This is just another piece of the puzzle.

In the spring, the SEC and CFTC sorted crypto assets into categories, recognizing BTC and ETH as commodities rather than securities.

In August, the SEC proposed a separate capital-raising regime for projects, and just days ago it issued rules on crypto custody for funds.

Whether to let crypto into the U.S. financial system is no longer the question. What matters more now is the rules it will operate under.

You understand that nobody cares about the crowd’s interests, so the next wave seems likely to be less about altcoins (although there are related projects there too) and more about licenses, reporting, and exchange infrastructure.

Boring? Maybe. But this is exactly the kind of thing that builds a market big capital is willing to enter.

👇 This is one of the slides Carson showed her advisers.

Stocks are at all-time highs, U.S. nominal GDP is growing at almost 8%, and investor sentiment feels like the calm before a storm.

According to BofA and EPFR, money market funds saw their largest weekly inflow since April 2020. Back then, money was fleeing the pandemic; now it’s fleeing a rising market.
▪️The Economist asks on its cover: could bonds explode?
▪️According to Axios, executives at Anthropic, OpenAI, and other AI companies are gaming out scenarios of a public revolt after a catastrophic AI event—and factoring one in for 2027.
▪️And Michael Burry writes that the market is at the first of five stages of accepting the inevitable—“denial,” which, based on the experience of 2000 and 2008, lasts 6–9 months.

For Carson, this is a contrarian signal. Fear amid a strong economy is more often fuel for further growth than a sign of a peak. And sooner or later, money in money market funds will return to stocks in search of yield.

💡 Bank earnings season kicks off tomorrow

We’re keeping an eye on lending activity, lending standards (whether banks are tightening their lending conditions), and reserves (whether they’re setting aside additional reserves).

👀 The market’s reaction will also be important: if stocks sell off despite strong earnings reports, that’s bad for the market. And vice versa…

$FTT ’s Old Bitcoin-Mining Assets Are Now Part of the AI Boom

Data-center company sees potential in what has been described as one of Sam Bankman-Fried’s ‘most reckless investments’

A bitcoin-mining business that appeared left for dead after the collapse of Sam Bankman-Fried’s crypto empire is being resuscitated to serve the artificial-intelligence build-out.

What everyone is afraid to say out loud…

Jefferies predicts the boom will end in massive capital destruction in the US

It sounds like a contrarian thesis, and in part it is. Chris Wood has been skeptical of the US market for years and favors Asia, so pessimism about the US is a natural position for him. There’s nothing surprising about it.

The weakest point in the report is the OpenRouter data, since the developers who use this aggregator are mostly looking for the cheapest tokens. OpenAI and Anthropic sell most of their volume directly and through the clouds of Microsoft, Amazon, and Google. So the 76 trillion tokens from Chinese models reflect the price-sensitive segment, not the entire market.

Still, Wood’s arguments shouldn’t be dismissed. Data centers are increasingly being built with debt rather than hyperscalers’ own money, and repayment of that debt is expected to come from sustained demand from OpenAI and Anthropic. If token prices keep falling as fast as they are now, revenue may not be enough to service the loans.

Capital destruction, of course, doesn’t mean AI will fail. Fiber-optic cable didn’t disappear after 2000 either; it’s just that the people who laid it weren’t the ones who made money from it 🧐.

And the biggest risks are still with the lenders.

😬 It’s almost impossible to make money in the market from fears about dangerous AI.

Economist Tyler Cowen suggests alarmists back up their predictions with trades. In other words, if catastrophe is likely, markets should feel it in advance. To some extent, they already do. Fears of AI-powered cyberattacks have boosted cybersecurity stocks. CrowdStrike (CRWD) has more than doubled in value in six months and trades at nearly 200 times earnings. You can also bet against vulnerable companies—for example, by buying CDS on regional banks’ debt…

Insuring yourself against AI is much harder. Insurers exclude generative AI-related damage from policies because they don’t know how to assess it. This isn’t a risk with known probabilities, but uncertainty with no history or data.

The main point of The Economist article is that the worst-case scenario can’t be hedged in principle. If humanity disappears, there’ll be no one to pay out on the policy. So Cowen’s “prove it with a bet” test doesn’t work here.

Could a hypothetical Polymarket help? It could help assess moderate AI threats, such as cyberattacks, outages, or regulation. But it can’t price existential risk either.

So the way to make money is not by betting on prejudices, but in the market.

👀 Look at who’s not here…

Nvidia (NVDA) and other megacaps didn’t make the top 50. All the money flowed into bottlenecks in AI infrastructure.

▪️Memory and data storage rose the most. SanDisk (SNDK) +566%, Micron (MU) +261%, Seagate (STX) +184%, Western Digital (WDC) +131%.

▪️Server and optics manufacturers are close behind: Dell (DELL), HPE, Marvell (MRVL), Lumentum (LITE), Ciena (CIEN).

▪️Power for data centers is also growing, including Bloom Energy (BE) +223%, GE Vernova (GEV), Quanta (PWR).

Outside the AI theme, refiners stand out: Marathon (MPC), Valero (VLO), Phillips 66 (PSX), along with cybersecurity companies: CRWD, PANW, FTNT.

💪 The year’s leader is Moderna (MRNA), up 663%. This one is outside the trends

🇺🇸 The U.S. economy’s shock absorbers are worn out.
This could lead to instability if investors’ recent concerns intensify, the WSJ reports.

Usually, when the economy slows, Uncle Sam steps in to help. Investors rush to buy Treasury bonds, their yields fall, and the real cost of borrowing for businesses and people may even decline, despite widening credit spreads. That’s what happened during the dot-com crash and the 2001 recession. Investment-grade spreads rose by about 1.8 percentage points, while the yield on 10-year U.S. Treasuries fell by as much as 2.8 percentage points.

Now that shock absorber isn’t working. Treasury yields are near their highest level in a quarter-century, and economist David Rosenberg says it’s not because the economy is strong. Lenders are already beginning to price in the risk of defaults. GDP growth is almost entirely underpinned by investment in AI data centers, while transportation, housing, restaurants and retail are sending warning signals.

Rates are also under pressure from doubts about government debt and a tougher tone from the Fed, which is increasingly expected to raise rates (not). In 2022, a similar situation didn’t end in recession thanks to $2 trillion in excess household savings accumulated after the pandemic.

Trump bought up to $5 million worth of SpaceX bonds two days before announcing his space initiative. In addition, in August he bought $25 million worth of Meta (META) shares, $6 million worth of Microsoft (MSFT) shares, and Oracle (ORCL) shares — CNN.