Monday, August 17, 2026, written at 9:15am Eastern, forty-five minutes after the Empire State survey and fifteen minutes before the bell, with the NAHB housing index still to come at 10am. Two things happened over the weekend that the tape is treating as one bullish item and that we think are two different items pointing the same direction. Anthropic showed prospective investors a quarter that makes the AI revenue line real. Nvidia cut in half the amount of its own balance sheet standing behind somebody else’s data centre. Reconciling those is the note.
Top Call: The Revenue Got Proven and the Leverage Got Trimmed
Anthropic circulated preliminary figures showing more than $11.5 billion of second-quarter revenue against $787 million in the same quarter of 2025, a fourteen-fold increase, with positive adjusted operating income for the first time. The company is pitching a 2028 revenue forecast of $190 billion to $200 billion into an IPO conversation where bankers are reportedly applying revenue multiples to numbers two years forward. Separately, the Wall Street Journal reported that Nvidia has reduced its guarantee on OpenAI’s ten-gigawatt Pike County campus in Ohio from a planned $250 billion to under $120 billion, narrowing the commitment to the first phase after shareholders raised concerns about circular financing exposure.
Our take: Friday’s note argued that equity had priced the buildout while credit had only just started pricing its debt, and that the long end of the Treasury curve was becoming the binding input for AI infrastructure equities. Today is the first evidence of that pressure producing a behavioural change rather than just a repricing. Nvidia did not cut the backstop because demand fell. It cut the backstop because carrying $250 billion of contingent obligation against a single customer’s spending is what a credit analyst charges you for, and Broadcom’s 6.6% fall on Friday over a $370 billion senior debt estimate was the demonstration. Meanwhile the demand side delivered exactly the receipt the sceptics have been demanding, which is a large private AI company generating more than $11 billion in a quarter and turning an operating profit while doing it. That is a healthier configuration than the one we had a week ago, because a cycle funded by customer revenue is durable and a cycle funded by vendor guarantees is not. The equity market is right to buy this. What it should not do is read the Ohio cut as a negative for Nvidia, since removing $130 billion of contingent risk from a supplier’s balance sheet is a de-risking of the supplier, not a signal about the buyer.
The Tape
| This Morning | Prior Close (Aug 14) | |
|---|---|---|
| S&P 500 fut / cash | 7,804.25 (-0.01%) | 7,785.76 (-0.17%) |
| Nasdaq 100 fut / Comp | 30,208 (+0.22%) | 26,729.16 (-0.28%) |
| Dow fut / cash | 53,605 (-0.38%) | 53,732.41 (-0.20%) |
| Brent / WTI | $89.28 / $81.74, both easing | Brent +5% on the week |
| 10Y / 2Y | 4.69% / 4.17%, both a touch richer | September hold near 70% |
| Gold / VIX | $4,440 / 15.08 | volatility bid off the 2026 low |
The composition repeats a pattern this note has flagged for two weeks. Nasdaq futures are the only major up, Dow futures are the worst, and the Russell has given back some of Friday’s record. Storage and memory lead in the pre-market, with SanDisk up around 4% and Micron up roughly 3.5%. Broadcom is up about 1.2% after Friday’s fall. JPMorgan raised its S&P 500 target to 8,000 from 7,800 on cloud monetisation. What none of this is, is broad. The bid is one theme wearing several tickers.
Anthropic: The Most Consequential Print of the Weekend Was Not a Public Company
| Anthropic (Q2 2026, preliminary) | Prior Year | Reported | Read |
|---|---|---|---|
| Quarterly revenue | $787M | >$11.5B | roughly 14x |
| Sequential | ~$4.8B in Q1 | >$11.5B | more than doubled |
| Adjusted operating income | negative | positive, ~$559M projected | first ever |
| Disclosed run rate (May) | n/a | $47B | now well past it |
| 2028 revenue forecast | n/a | $190-200B | the IPO case |
There is no consensus line to beat here because the company is private, which is precisely why the number matters. For eighteen months the central objection to the AI capital cycle has been that the spending is visible and the revenue is not, and that the revenue which does exist circulates between a small number of counterparties who are also each other’s investors. A quarter of more than $11.5 billion, up fourteen times year over year, with positive adjusted operating income and roughly 80% of the mix coming from API and enterprise rather than consumer subscriptions, is a direct answer to that objection. Enterprises are paying, at scale, and the unit economics are no longer negative.
Where we would be careful is the 2028 forecast. Getting from an $11.5 billion quarter to $190 billion of annual revenue in two and a half years requires roughly four times the current run rate, and bankers marking valuation off a two-year-forward revenue multiple are doing something that is common in high-growth software and unusual at this distance. The honest framing is that the near-term print is verified and the terminal number is a pitch. Both are information. They are not the same quality of information.
Our take: this is the strongest fundamental datapoint of the month for the buildout and it changes the burden of proof on the demand question. The reason it lifted Amazon and Alphabet this morning rather than only the chip names is that the revenue is being booked through cloud resellers, which means the hyperscaler capex line and the model developer revenue line are visibly the same flow of money. That is good for the durability of orders. It also means the concentration risk everybody worries about is real, because a handful of model companies now underwrite a very large share of what the semiconductor complex is building against.
Nvidia and Ohio: Vendor Financing Gets a Haircut
The reduced guarantee is the second retreat from balance-sheet-heavy AI financing in three sessions, and it is worth reading alongside the rest of the structure. Nvidia is still reportedly in talks for a $3 billion investment in SB Energy, the SoftBank subsidiary developing the site, along with up to $100 billion of credit support, and OpenAI is still negotiating a binding lease for the full ten gigawatts. The project is not shrinking. What is shrinking is the share of it that sits on the chip supplier’s balance sheet as contingent obligation.
That distinction is the whole story. On August 10 Nvidia announced six Wall Street firms mobilising more than $500 billion of third-party capital against GPU collateral, which was the market’s clever solution to funding compute without using the manufacturer’s equity. The Ohio cut is that solution being applied. Third-party capital takes the exposure, the vendor keeps the revenue, and the credit market prices the risk explicitly instead of hiding it in a footnote. Nvidia edged higher on the news, which is the right response.
Our take: the vendor-financing critique has been the single most effective bear argument against this cycle and the companies are now visibly dismantling it. That is bullish for the equity and it raises the stakes on the debt. Somebody still funds ten gigawatts in Pike County. If it is not Nvidia, it is a securitisation market that has never priced compute collateral through a downturn, and the 30-year Treasury at roughly 5.25% is the hurdle every one of those structures clears or does not.
Micron Through $1,000: The Shortage Became a Price
Micron traded above $1,000 for the first time, up roughly 3.5% in the pre-market and about 240% year to date, with HBM capacity booked through 2027 and DRAM in structural deficit. New Street moved to Buy with a $1,250 target and UBS reiterated Buy at $1,625. Management continues to describe conditions as tight beyond 2027. SanDisk added another 4% on top of a 35% five-session rebound, and in Tokyo, Kioxia closed up 15.07% at 61,840 yen.
This is the same phenomenon this note has tracked since the August 6 SanDisk print and the August 14 Applied Materials quarter, viewed from the third angle. Applied expanded semiconductor systems margin nearly five points because it sells into a shortage where the buyer’s constraint is capacity rather than price. Micron above $1,000 and Kioxia up 15% in a session are the manufacturers capturing the same rent. The rule that keeps working is that margin in this cycle accrues to whoever is capacity-constrained, and it drains from whoever sells into a buyer with alternatives.
Korean markets were shut today for the Liberation Day holiday observance, which removed the highest-beta expression of this trade from the tape and probably muted the move. The Kospi closed Friday at a record 6,977.34 after an 11.5% week. It reopens tomorrow into a memory complex that has repriced again without it.
Empire State at 20.6: The Manufacturing Rebound Arrives With a Catch
| Empire State (August) | Expected | Actual | Read |
|---|---|---|---|
| Headline | 10.2 | 20.6 | highest since early 2022 |
| New orders | n/a | 17.3 | expanding |
| Shipments | n/a | 11.7 | expanding |
| Employment | n/a | 9.3 | hiring, modestly |
| Prices paid | n/a | 58.6 | the problem |
| Selling prices | n/a | 22.7 | 36-point gap |
| Six-month outlook | n/a | 32.1 | confident |
The headline nearly doubled expectations and set a four-year high, five points above July’s 15.6. Supply availability at minus 13.4 says inputs are getting harder to source, which is consistent with a capital goods cycle running hot. On the surface this is the first genuinely strong American growth print in a month of soft ones, and it lands two sessions after retail sales fell 0.6% and Michigan sentiment collapsed to 51.0.
The internals are where the interesting part is. Prices paid at 58.6 against selling prices of 22.7 is a thirty-six point gap, and a gap that wide means manufacturers are absorbing input costs rather than passing them through. That is disinflationary for the CPI and margin-negative for the companies, which is a genuinely awkward combination for anyone hoping the manufacturing recovery shows up in earnings. It also complicates the Fed’s read, because a committee watching tariff and energy pass-through gets a survey saying costs are rising sharply and output prices are not.
Our take: the two-speed economy this note has been describing for two weeks is now measurable in a single release. The industrial and capital goods side is accelerating, the household side is contracting, and the manufacturing side is eating the cost increase rather than sending it to the consumer. September hike odds sit near 30% against 47% a month ago and this print does not move them, because a regional survey with prices paid at 58.6 and selling prices at 22.7 is not an inflation signal. It is a margin signal.
China: The Same Split, Drawn Harder
| China (July) | Expected | Actual | Prior |
|---|---|---|---|
| Industrial output y/y | +5.0% | +4.5% | +5.3% |
| Retail sales y/y | +1.5% | +0.6% | +1.0% |
| Fixed asset investment YTD | -6.2% | -6.7% | deepening |
Every line missed, industrial output slowed for the first time in three months, retail sales grew six-tenths of a percent, and the investment slump steepened. New home prices extended their decline. This is a sluggish start to the second half and it puts pressure back on Beijing to do something.
Chinese equity rallied anyway. The Shanghai Composite was up 0.8% at 3,960 with the CSI 300 up the same, the ChiNext gained 1.3%, the STAR 50 added 3%, the CSI Semiconductor Index rallied 4% and the AI index rose 1.5%. The Hang Seng closed up 1.3%, or 336 points, at 25,435, snapping four consecutive declines, with SenseTime up about 8.9% after guiding to its first profit since listing.
Our take: this is the most important structural observation in today’s note. The American version of this split, where a weak consumer coexists with a strong chip complex, has been explicable as an artefact of index composition. Seeing the identical pattern in China, where retail sales growth of 0.6% and a 4% semiconductor rally print on the same screen, says the split is not local. Global equity has organised itself around a capital expenditure cycle that does not route through household demand in either of the world’s two largest economies. That is a coherent trade while it lasts and it has an obvious end condition, which is the moment the capex has to be serviced out of consumption that is not growing. Nobody has to answer that this year.
Japan: Growth Undershoots and the Consumer Turns Negative
Japan’s second quarter annualised at 1.1% against 2.0% expected, 0.3% quarter on quarter against 0.5%, down from a revised 1.9%. Private consumption was flat to slightly negative for the first time in eight quarters. The Nikkei rose 0.74% anyway to 69,220.25, within reach of 70,000, carried by Kioxia up 15.07% and SoftBank up 2.56%.
The bond market is the part to watch. The Japanese ten-year sits at 2.925%, near a thirty-year high, and a growth print this soft with consumption turning negative complicates the Bank of Japan’s September timeline considerably. The yen firmed 0.2% to 159.06 despite the miss, which tells you the move was dollar-driven rather than a vote on Japanese growth. A weak GDP number that does not weaken the currency is a market that has already decided the BOJ is hiking regardless, or one that thinks the Fed is the bigger variable. Both readings point the same way for now.
Europe: Quiet and Slightly Better
The Stoxx 600 added 0.2% to 659.30, holding the record shelf it has occupied all month, with basic resources up 1.4% on firmer copper and technology contributing. The FTSE 100 was on track to break a six-session losing streak. Argenx rose 7.7% on positive trial data and Rolls-Royce added 1.8%. Deutsche Telekom agreed to buy Polish fibre operators Fiberhost and Inea for around one billion euros.
The UK datapoint worth noting is that asking prices for homes fell 2.0%, the largest August decline since 2018, which is another entry in the global housing-and-consumer column that keeps getting weaker while the industrial column gets stronger.
Oil and Geopolitics: A Stalemate That Keeps Getting More Expensive
Brent rose as much as 1% to $89.40 before settling back 24 cents to $89.28, and WTI fell 67 cents to $81.74. Both contracts gained more than 5% last week after drone attacks on ADNOC tankers in the Strait and a strike on a Saudi Aramco refinery. Transit volumes remain far below the roughly 73 vessels a day of normal conditions, and some tankers are crossing with transponders switched off, which means the reported flow understates the actual one by an unknown margin.
The diplomatic position has not moved. American and Iranian talks are stalled, and Tehran’s stated line remains that there will be no negotiation until 2029. Treasury Secretary Bessent flagged measures last week that he described as economic isolation of an unprecedented kind, and those are expected this week. Elsewhere, the European Union is preparing what it calls its most far-reaching Russia sanctions package for the autumn, Ukraine has said it has zero Patriot interceptors remaining ahead of a winter in which Russia has historically targeted energy infrastructure, and Washington is preparing for a US and China summit on September 24 with the current trade truce expiring in November. Taiwan has proposed a 2027 defence budget of T$1.1225 trillion, a 16% increase taking it above 3% of GDP.
Our take: crude giving back its early gain on a session with no supply resolution and no diplomatic progress is the signature of a market that has settled into pricing a frozen conflict rather than an escalating one. Six months in, both sides have found a sustainable level of pain, and the world is drawing down stockpiles to cover the difference. We keep the $85 to $90 Brent band with risk skewed higher, and we would treat the Treasury measures due this week as a dollar-plumbing event before an energy one. The genuinely underpriced item is the calendar convergence in September, which puts the US and China summit, the Fed, and an expiring trade truce inside a five-week window.
Also Worth the Time
L3Harris parted ways with chairman and chief executive Christopher Kubasik after an independent investigation found conduct inconsistent with the company’s code of conduct, and named Sam Mehta chief executive effective immediately, with Lewis Hay III as independent chairman. Mehta ran Space and Mission Systems along with Communications and Spectrum Dominance, segments representing more than 80% of revenue, which makes this an unusually clean succession for an abrupt one. The company said the conduct has no bearing on financial reporting, controls, customer relationships or operations. Shares fell about 3%, and given that the incoming chief executive already runs the majority of the business, we read the move as governance noise rather than an operational event.
Rocket Lab extended a seven-session run of roughly 28% on a $397 million Space Force contract for threat-tracking satellites launched on Neutron, alongside its pending $8 billion acquisition of Iridium. SpaceX disclosure continues to trickle out, with Intesa Sanpaolo revealing a $966 million stake. Shein is targeting a Hong Kong listing at roughly $25 billion to $28 billion, down from the $100 billion once discussed, which is a useful reminder of where non-AI growth multiples actually sit. Alibaba is selling Lingxi Games to Trustar Capital for more than $2 billion. OceanaGold is acquiring Ausgold for A$776 million. Bristol Myers Squibb received accelerated approval for Zenbexus in relapsed or refractory multiple myeloma last week ahead of today’s decision date. Duolingo fell 7.8% and Heartflow rose 35.7%.
Key Events Today
- 8:30am: Empire State manufacturing, released at 20.6 against 10.2 expected, prior 15.6
- 10:00am: NAHB housing market index, 35.0 expected against 34.0 prior, against UK asking prices down 2.0%
- Overnight: China July activity data, all three lines missed; Japan Q2 GDP at 1.1% annualised
- Tomorrow: Home Depot, housing starts, and the Kospi reopening after the holiday
- Wednesday and Thursday: Target and Lowe’s, then Walmart and jobless claims, the cleanest read available on whether the low-income consumer is cracking or trading down
- This week: the Treasury measures on Iran that Bessent flagged
- August 26: Nvidia FQ2, consensus roughly $93 to $95 billion
- August 27-29: Jackson Hole, the first formal framing of September with the long end near 5.25%
- September 16-17: FOMC, roughly 30% priced for a hike against 47% a month ago
What We Are Actually Watching
Three threads.
The first is that the AI trade has quietly changed its funding model in the space of a week and the equity market has not fully absorbed what that means. Ten days ago the structure was that chip vendors guaranteed customer data centres, which converted supplier equity into customer capex and produced the circularity critique. Nvidia cutting Ohio from $250 billion to under $120 billion, immediately after Broadcom fell 6.6% on a debt estimate, is the vendors handing that exposure to third-party credit. The good version is that risk is now priced explicitly by people whose job is pricing it. The version to watch is that compute-backed securitisation has never been tested through a demand air pocket, and the collateral is equipment with a depreciation schedule nobody agrees on.
The second is that Anthropic’s quarter resolves the demand question and sharpens the concentration one. More than $11.5 billion of revenue against $787 million a year ago, with positive adjusted operating income, ends the argument that nobody is paying for this. It also means a very small number of model developers now generate the revenue that justifies the entire buildout, and the largest of them is pitching a 2028 number four times its current run rate to bankers who are valuing it off that forecast. Verified present, projected future. Do not confuse the two.
The third is that the two-speed economy is now a global structure rather than an American quirk. Today produced weak Chinese retail sales alongside a 4% Chinese semiconductor rally, a weak Japanese consumer alongside a Nikkei approaching 70,000, weak UK housing alongside a Stoxx 600 on its record shelf, and an American manufacturing survey at a four-year high two sessions after retail sales contracted. The consistency is the finding. Capital is compounding in the part of the economy that builds capacity and stalling in the part that consumes output, and the resolution of that requires either the household to recover or the capex to be serviced by something else. Neither has to happen this quarter, which is why nothing in this morning’s price action looks worried.
Marks above are pre-market and will move by the open. NAHB at 10am is the only remaining print today, and the week’s real information arrives Thursday with Walmart.