The Market Stopped Paying for Capex. Microsoft Spent Less and Won.

Thursday, October 29, 2026, written at 9:40 a.m. Eastern, before the opening bell. Nasdaq futures are down about 1.1% and the S&P contract about a third of a percent, with Alphabet off roughly 10% after last night’s report and Microsoft up around 3%. Third quarter gross domestic product came in at 3.4% and September core personal consumption expenditures landed in line at 8:30. The European Central Bank decides at 9:15 and Amazon reports after the close. Marks below are pre-open and will move.

Top Call: Capex Stopped Being a Demand Signal and Became a Bill

For most of the past two years the trade in large cap technology was simple. A hyperscaler raised its capital spending plan, the market read that as a statement about how much artificial intelligence demand existed, and the stock went up. Last night that relationship inverted inside four hours, and it inverted in the cleanest possible way, because the two companies that reported ran the experiment against each other.

Microsoft grew Azure 48% to $33.17 billion, beating its own 44% to 45% guide for at least the fifth consecutive quarter, and spent $45.2 billion of capital against a plan that had been telegraphed above $50 billion. It then trimmed calendar 2026 capital spending to roughly $170 billion from about $175 billion. The shares rose about 3%. Alphabet grew Google Cloud 103% to $30.7 billion, which is the best number any cloud business has ever printed at that scale, carried a $561 billion segment backlog, and spent $51.4 billion of capital. Free cash flow came in at minus $9.07 billion against roughly minus $2.70 billion expected, the second negative quarter since the 2004 listing. The shares fell about 10%.

Read those side by side and the market’s new rule is visible. Spending is no longer evidence that demand exists. Demand has been proven, repeatedly, by the backlogs and the sold out memory books we spent yesterday on. What is now unproven is whether the spending can be financed out of operating cash flow, and the market has started charging for the gap. Alphabet produced $42.3 billion of operating cash and spent $51.4 billion, so the difference has to come from the balance sheet, and it told the call that 2027 capital spending will increase significantly from a 2026 range of $195 billion to $205 billion. Microsoft produced a similar growth rate while shrinking the gap. One of those is a self funding build and one is not, and that is the entire difference between plus 3% and minus 10%.

This is the same argument we made on Tuesday, pointed at a different part of the chain. Then the claim was that the roughly 23% drawdown in data centre names since the October 6 record repriced the cost of money rather than the strength of demand. Today the hyperscalers get the same treatment, and the honest consequence is uncomfortable for us, because Alphabet’s demand was excellent and the stock still lost a tenth of its value. Being right that demand is real is no longer sufficient to be right about a price. The discriminating variable going forward is cash conversion, and we would rather own 48% growth funded internally than 103% growth funded by a widening deficit.

The Search line deserves one note of caution before anyone treats this as purely a financing story. Advertising revenue came in below forecasts, and the bear case that artificial intelligence answers cannibalise the query funnel that pays for everything else now has a data point rather than only a thesis. One quarter is not a trend. Two would be.

Alphabet: The Quarter Where Cash Flow Beat the Narrative

The earnings miss is partly optical. A large share of the shortfall against the $3.01 figure is unrealised losses in the investment portfolio rather than operating deterioration, and anyone trading the headline number is trading a mark to market line. The cash flow miss is not optical at all, and that is what moved the stock.

What makes this print genuinely interesting is that capital spending came in almost $4 billion below what the Street expected and the stock was still punished for spending. That tells you the complaint is not about the dollar amount. It is about the shape of the funding, because guidance was held at $195 billion to $205 billion for this year with a significant increase flagged for next, which means the under-spend in the quarter is a timing artefact rather than restraint. The company also told the call to expect near term margin pressure in Cloud as it begins leasing third party compute, which is an admission that it cannot build fast enough to serve its own backlog and will rent the difference at someone else’s margin.

Our take is that the cloud business is worth more this morning than it was yesterday and the equity is worth less, and both of those can be true. A $561 billion backlog growing at triple digits is a franchise. A build that consumes more than it generates for another two years is a financing exercise, and financing exercises get priced off the rate curve, which is at 5.01% on the ten year and not falling.

Samsung: A Record Nobody Paid For

Samsung posted its largest quarter ever and the stock fell 4%. Operating profit of 107.44 trillion won was up roughly 783% and still a hair under the 107.66 trillion won expected, revenue reached about 195.3 trillion won, memory revenue rose 427% to 140.7 trillion won, and management said 2027 high bandwidth memory supply is already sold out at agreed prices with DRAM staying tight through 2028. Net profit of about 75.2 trillion won fell well short of consensus near 89 trillion on financial losses and a mobile division that lost roughly 800 billion won as its own component costs rose.

Set this against Tuesday. SK hynix beat, claimed sold out capacity through 2029 and an 80% operating margin, and took the entire Kospi up 6.6% to its first close above 6,000. Samsung printed a bigger absolute number two days later, confirmed the same supply narrative, missed by two tenths of a percent and got sold while the index fell 1.24%. That asymmetry is the tell. When a record is already in the price, the beat has to be large to be paid, and the memory trade has moved from discovery into expectation in the space of a week.

The detail we would actually act on is the mobile loss. Samsung’s own handset business is now being squeezed by the chip prices Samsung is charging, which is the first concrete evidence that the memory shortage has started taxing downstream electronics rather than merely enriching upstream suppliers. Goldman estimates artificial intelligence is adding about 23 basis points to American personal consumption inflation, mostly through memory. That loop is worth sitting with, because the build is now an input into the inflation that raises the rate that discounts the build.

Macro: Strong Growth, Sticky Inflation, and Two Dissents That Look Right

The data landed at 8:30 and it validated the hawks. Gross domestic product grew at a 3.4% annualised pace in the third quarter against 3.2% expected and 2.2% prior, driven by investment, consumer spending and government outlays. Headline personal consumption expenditure prices rose 0.5% on the month to 3.6% on the year. Core was 0.3% and 3.1%, both in line. Personal income rose 0.2%.

Yesterday the Federal Open Market Committee held at 3.75% to 4.00% in a 10 to 2 vote, with Governor Cook and Cleveland’s Hammack dissenting for a quarter point hike. Chair Warsh called financial conditions tighter but not tight, said politics played no role six days before the midterms, and refused to guide, telling the room he does not believe in forward guidance absent a crisis. He also announced that the committee will consider rate changes at seven of eight meetings next year and six of eight in 2028, which reduces the number of scheduled opportunities to respond and is a more consequential institutional change than it sounds.

The combination this morning is 3.4% growth with 3.1% core inflation and a labour market nobody is worried about. There is no case in that mix for the easing that long duration equity needs. December sits near 85% on futures and 82% on prediction markets, the two year has slipped to about 4.68% on Warsh’s refusal to commit while the thirty year is three basis points higher at 5.33%, and that steepening is the market saying the front end is capped by a chair who will not pre-commit while the long end absorbs the inflation. Core at 0.3% rather than 0.4% is the only thing standing between this and a December hike being fully priced.

The Optics Deal: Marvell and the Laser Nobody Can Design Out

Bloomberg reported that Marvell is in advanced talks to acquire Lumentum, with an announcement possible as early as next week. Marvell fell about 11% and Lumentum rose about 5% to 9%. Lumentum carries a market value near $83 billion against Marvell’s $221 billion, and Marvell held roughly $3.92 billion of cash at its last report, so Bernstein’s observation that this cannot be funded with cash is an understatement rather than an insight.

The strategic logic is sound and the market is right to be nervous anyway. Marvell already buys lasers from Lumentum and pairs them with its own digital signal processors, and it is already acquiring Celestial AI for $3.25 billion. Optics is the part of the rack that co-packaged designs cannot remove, because the light has to come from somewhere even when the switch moves onto the chip. Owning the laser is owning the one component with no roadmap to zero.

What the 11% decline is pricing is the currency, not the asset. A deal of this size paid in stock at a moment when the market has just finished punishing Alphabet for a funding gap is going to be read as dilution first and strategy second, and the Nvidia complication is real, since Nvidia put $2 billion into Lumentum earlier this year and now has a position in a supplier about to be absorbed by a merchant competitor. We would treat the Lumentum move as the cleaner expression of the thesis and wait for terms before touching the acquirer.

Elsewhere in Earnings

Caterpillar is the one that matters for the build. Power and Energy sales rose 18% to $8.78 billion with segment profit up 31% to $2.14 billion, driven by turbines and large reciprocating engines going into data centres, and backlog hit a record $78.1 billion. The stock is down 9% because management held the annual sales forecast unchanged, roughly $128 million of operating profit came from tariff recoveries rather than operations, and backlog growth has cooled from earlier in the year. This is the Alphabet lesson again in a different sector. A record backlog is not a catalyst when the market has decided to price the second derivative.

ServiceNow and Fortinet together are the more disturbing pair. Both beat on the current quarter. ServiceNow raised its full year subscription outlook and still lost close to a fifth of its value on a fourth quarter revenue guide roughly $100 million light and a current remaining performance obligation guide of 18.5% against about 19.8% expected. Fortinet raised guidance for the third time this year and fell 27%. Software entered this earnings season priced for acceleration, and anything short of acceleration is being treated as a thesis break. That matters for every name whose story is that it can charge for agent deployment rather than seats, and it is the second consecutive day of evidence that the enterprise software multiple is the most fragile thing in this market.

Eli Lilly is the clean print of the morning. Revenue up 39%, the full year revenue guide lifted by roughly $5 billion to $91 billion to $92 billion, and only a trim at the earnings line. Sales of the obesity pill came in under the roughly $229 million the Street wanted, which is the single soft spot and the one to watch next quarter.

Energy and Geopolitics

Crude is giving back yesterday’s near 3% gain, with WTI down 1.3% to $101.44 and Brent off 0.6% near $114.60. The move is diplomatic rather than physical. Qatar continues to mediate, Iran keeps passing messages through intermediaries, and the United Nations has called publicly for talks to reopen the Strait of Hormuz, so the market has shaved some of the premium it added on Wednesday’s reporting that Washington warned Gulf states to expect a renewed campaign after November 3.

Nothing physical improved. The Revolutionary Guard said three oil tankers struck naval mines in the southern lane that American authorities have spent two months describing as safe, and a tanker was burning in the strait overnight. Two of the three claims are uncorroborated, which is the pattern we have flagged repeatedly, but the first one is a fire that can be seen. Secretary Rubio said Wednesday that Iran has lost control of the strait while an adviser to the Guard said Iran will shortly block what it calls illegal routes. Traffic continues at a fraction of normal volumes, supported by American escorts and by overland pipeline routing.

Our read is unchanged and this morning does not challenge it. The premium is attached to transit and to the option value of escalation, not to barrels lost, which is why a mediation headline can take a dollar out while a mined tanker cannot put one in. President Trump has said he will not resume strikes before the midterms. That is five days of calendar, and the risk is asymmetric on the other side of it.

Asia and Europe

Asia split on the two prints. The Nikkei rose 1.10% to 64,659.47 on artificial intelligence related strength ahead of tomorrow’s Bank of Japan decision, with the yen near 157.82. Korea went the other way, the Kospi down 1.24% as Samsung fell 4.1% and gave back part of Tuesday’s record. The Hang Seng closed down 0.59% at 22,958.65 with the Hang Seng Tech Index off 2.84%, and the CSI 300 fell 1.03% to 4,041.84. The Alphabet read-through travelled straight into Asian hardware, which is the cost of a region whose index weight is a derivative of American capital spending plans.

Europe is higher and ignoring it. The Stoxx 600 is up about 0.40% near 646.72, the DAX up 0.30% at 25,780.29, the CAC 40 up roughly 0.8% near 7,944 and the FTSE 100 up 0.2% at about 10,743, with banks the weakest sector and semiconductor names tracking the American futures lower. Earnings are carrying it. Shell beat and raised its buyback, Adidas lifted annual guidance, Stellantis jumped on results, and Volkswagen fell after nine month operating profit collapsed 98.7%.

The institutional news out of Europe is larger than the tape. Christine Lagarde resigned this morning, stepping down on February 7 rather than finishing a term that runs to October 2027, and she framed the early exit as giving European leaders time to choose a successor. She leaves with headline inflation at 3.8%, the highest since 2023, a deposit rate at 2.50% after hikes in June and September, roughly 80% odds on another quarter point in December, and French politics still generating street protests over austerity. Whoever takes that chair inherits a tightening cycle in progress and a fiscal argument in France that bond markets have been pricing for a month. The decision at 9:15 is expected to be a hold with a hawkish signal, and the succession question is now the more important variable for European duration than anything in today’s statement.

Also On The Tape

• DoorDash agreed to acquire Instacart for about $17 billion of equity value at $65.00 a share, a 26% premium to the October 22 close, at least 75% in cash with stock capped at a quarter of the consideration and unanimous board approval. DoorDash is up about 5% and Instacart slightly lower, which is a rare pairing and says the market thinks the buyer got the asset cheaply. Uber walked away from the same conversation
• CoreWeave’s five year credit default swap is still in the 1,038 to 1,088 basis point range after touching above 1,140, against more than $51 billion of debt and obligations. Our October 23 test resolves at tomorrow’s close, and above 1,000 it resolves yes, meaning the funding stress in the neoclouds is structural rather than a single illiquid quote
• SpaceX has a binding commitment with an Apollo managed entity for roughly $40 billion of graphics processor financing, with the vehicle buying the hardware and SpaceX leasing it for about five years from the first half of 2027. Fitch left the BBB+ rating unchanged on Wednesday, which is the clearest statement yet that non artificial intelligence cash flow is what the credit market is actually underwriting
• Chip and software complexes both weakened after hours, with Nvidia, AMD and Micron lower and the software index falling on the ServiceNow and Fortinet guides. SAP fell in Europe on the same read
• Amazon reports after the close with the Street near $1.96 on about $202 billion of revenue and Amazon Web Services growth around 37% to roughly $45 billion. Capital spending is already guided to approximately $220 billion for 2026 and has exceeded record operating cash flow for two consecutive quarters. On last night’s evidence the capex line will be read as a liability rather than a boast, and that is the asymmetry going into the print

On The Record

Tuesday’s note argued the October drawdown repriced the cost of capital rather than demand, and recommended discriminating by funding structure rather than selling the complex. Last night supports the mechanism and embarrasses the comfort we took from it. Alphabet delivered the strongest demand evidence of the cycle, 103% cloud growth and a $561 billion backlog, and lost a tenth of its market value on a cash flow line. We were right that the market is pricing funding structure. We were too slow to say that a company can have perfect demand and still be the wrong thing to own while it funds that demand externally at a 5% ten year. The distinction we drew between suppliers who convert orders into cash and customers who do not now applies inside the hyperscaler group as well, and Microsoft against Alphabet is the cleanest version of that test anyone is going to get.