NextFin

Why $SNDK Fell After a Double Beat: The Market Trades the Slope, Not the Level Revenue of $11.03B beat $10.64B. Adjusted EPS of $49.51 beat $46.22. Gross margin hit a record 85.1%. Revenue is up 378% from a year ago, net income went from $112 million to $8.1 billion, and Q2 guidance is above consensus too. Plus an 8-for-1 stock split. And the stock is red. The same thing happened in August. Here’s why. Growth is still huge, but the slope is flattening. Look at sequential revenue growth: Fiscal Q4 2026: +51% ($5.95B → $8.97B) Fiscal Q1 2027: +23% ($8.97B → $11.03B) Fiscal Q2 2027 guide: ~+12% (midpoint $12.4B) Growth is roughly halving every quarter. In a cyclical industry like memory, stocks don’t peak when earnings peak. They peak when the rate of improvement peaks. A decelerating slope is exactly what investors who lived through past memory cycles are trained to sell. Margins have hit a ceiling. Gross margin went from 78.4% to 84.6% to 85.1%, and guidance is 84.5–85.5%. There’s not much room above 85%. When margins stop expanding, earnings growth has to come entirely from volume and price, and price is what’s slowing. NAND price hikes are cooling. TrendForce estimated NAND contract prices rose 70–75% in the spring quarter, then projected 10–15% for the quarter after. Goldman’s view is that hikes continue through Q4 2026, then prices start falling in 2027 as new supply comes online. That’s the bear case in one sentence: the best pricing may be right now. The bull case is contracts. Unlike past memory cycles, this one is built on long-term agreements. Last quarter Sandisk reported $59.8B of remaining performance obligations, or $91.1B including deals signed after quarter end, with about half of fiscal 2027 bit shipments and two-thirds of fiscal 2028 already committed. Data center revenue jumped 61% sequentially to $4.80B, and that’s where the contracts sit. Contracted volume at locked-in terms is the reason this cycle shouldn’t collapse the way 2019 or 2023 did. The warning light in the report: consumer. Consumer revenue fell 39% year over year to $397M. High memory prices are crushing demand for consumer storage, the same dynamic that drove PC shipments down 20% and sank $INTC yesterday. At these prices, memory makers are winning data centers and losing everyone else. About the split. An 8-for-1 split changes nothing about the value of the company. It makes shares cheaper per unit and easier to buy for employees and retail investors. It’s a signal of confidence, not a fundamental catalyst. Bottom line: this was a great quarter, and the market sold it anyway because memory stocks trade on direction, not level. Growth is decelerating, margins have plateaued, and pricing momentum is fading. The counterweight is a contract book unlike anything in past cycles. The real test isn’t this quarter. It’s whether 2027 contract prices hold when new supply arrives. Record earnings. Flattening slope. In memory, the slope wins.

$AMZN Proved AI Demand Is Real. The Market Is Now Pricing the Bill. AWS revenue came in at $46.0B, up 39.3%, its fastest growth in nearly five years and ahead of the $44.9B estimate. Operating income of $28.1B beat both the $26.1B consensus and the top of Amazon’s own guidance. Andy Jassy said he expects AWS to grow more than 40% in both 2027 and 2028. And the stock fell as much as 12% after hours. The demand question is settled. This week, all three major clouds accelerated: AWS: $46.0B, +39% Azure: $33.2B, +48% Google Cloud: $30.7B, roughly doubled AWS has now accelerated for six straight quarters, from 17% to 39%, on a base that’s now running at roughly $184B a year. Jassy says the constraint is capacity, not customers. Nobody serious is arguing AI cloud demand is fading. So why the selloff? Look at the cost side. Capex: Amazon raised its 2026 capital spending plan to $220B last quarter, mostly because of higher memory chip costs. That’s the same memory squeeze that crushed $INTC’s PC business today. Cash: trailing free cash flow had already turned negative at −$7.6B, from +$18.2B a year earlier. Debt: long-term debt climbed to $128.9B from $65.6B at the start of the year. Amazon is funding the AI build-out with borrowed money and every dollar of operating cash it generates. The near-term misses gave the market an excuse. EPS of $1.89 missed $1.96. Revenue of $201.9B was a hair below $202.1B. Advertising, one of the highest-margin businesses at Amazon, fell short of estimates. And Q4 operating income guidance of $31B came in below the $33B expected. None of these is a big miss alone. Together, after the stock jumped more than 15% on its last report, they were enough. The pattern this week. Look at how each hyperscaler was judged: $MSFT: Azure accelerated and capex came in under plan. Stock up. $GOOGL: Cloud doubled, but the cash engine (Search) stalled. Stock down. $AMZN: AWS accelerated, but cash flow is negative, debt is up and the profit guide was light. Stock down. The market is no longer rewarding growth alone. It’s rewarding growth that doesn’t require borrowing to fund it. What would change the story. The 2027 capex outlook and the path back to positive free cash flow. Analysts were expecting negative free cash flow for 2026 and a recovery starting in 2027. If the call pushes that recovery out, the pressure stays. If AWS keeps compounding at 40%+ while capex growth flattens, today’s selloff ends up looking like the cost of being early. Bottom line: AWS just delivered its best growth in five years, and the CEO is guiding for more. The stock is falling because Amazon is paying for that growth with negative cash flow and nearly double the debt. This is no longer a demand debate. It’s a return-on-capital debate, and the market wants to see the returns before it pays for more spending. Demand is proven. The bill is due. Watch free cash flow.

$INTC Got Crushed by the Same Thing Making Memory Stocks Rich Intel beat on Q3: EPS of $0.52 vs $0.40 expected, gross margin of 44% vs 42%, revenue in line at $16.5B. Then it guided Q4 revenue to $15–16B against a $17.1B estimate, and EPS to $0.25 vs $0.43. The stock fell more than 20%. Intel blamed “softness in the PC market.” That’s true, but it undersells what’s happening. The PC market isn’t soft. It’s collapsing. Global PC shipments fell 20.1% year over year in Q3 to 62.7 million units, the worst quarter since 2023. Part of that is payback: PC makers stockpiled earlier this year to get ahead of memory price hikes, pulling demand forward. Omdia now expects Q4 shipments to fall another 24% and 2027 to fall another 7%. Intel’s Q4 guide is simply that forecast showing up in the numbers. The cause is AI. Every PC needs DRAM and NAND, and AI data centers are absorbing that supply. Memory component prices have reportedly more than quadrupled, and memory now makes up close to 40% of a typical PC’s bill of materials. When the most expensive part of a laptop quadruples in price, PC makers either raise prices or cut memory, and buyers either pay more or wait. Intel added to the pressure. It raised PC processor prices for the third time in under a year, effective October 5, with cumulative increases around 30%. Higher chip prices on top of higher memory prices is a recipe for demand destruction. This is the flip side of the memory story covered here earlier this month. The same supply squeeze that’s driving record profits at $MU, SK Hynix and Samsung is starving the PC market. Intel’s guide is, ironically, one more data point that memory demand is outrunning supply. Data center is doing its job. Data Center and AI revenue came in at $7.32B vs $6.67B expected, up about 17% from $6.26B last quarter. Server CPUs are riding the same AI build-out as everything else. The problem is that PCs are still the bigger business, and that’s where the guide broke. Why −22% and not −8%. Valuation. Intel hit an all-time high around $133 earlier this year, finally clearing its dot-com peak, and at one point it was up more than 225% year to date. That rally was built on the turnaround story: Tesla as the first major 14A foundry customer, reports of Apple foundry talks, a $23B equity raise to fund the build-out. When a stock is priced for a comeback, a 9% revenue guide miss gets punished hard. The CFO warned back in Q1 that PC demand would weaken in the second half. The market just didn’t price how much. What this means for the rest of the sector. $AMD reports soon, with the same PC exposure. Expect the same question. $DELL, $HPQ and other PC makers face the same memory squeeze on margins and volumes. Memory names get an indirect confirmation that pricing power is intact. Bottom line: Intel’s quarter was fine. Its next quarter is hostage to a PC market being squeezed by the AI memory boom. The foundry story that drove the rally isn’t broken by this report, but foundry revenue is years away, and PCs pay the bills today. Until memory prices ease, which TrendForce suggests may not happen even in 2027, the PC side of Intel stays under pressure. AI is eating the PC. Intel is still mostly a PC company.

$GOOGL’s Cloud Doubled. The Stock Fell 8% Anyway. Here’s Why. Same night, two very different reactions. $MSFT ripped after Azure accelerated. $GOOGL dropped about 8%. Revenue was roughly in line at $127.6B, but Search, YouTube, Google Services and operating income all came in below estimates. To understand the selloff, look at what each business means to the stock. Search stalled. Search revenue came in at $63.35B. Last quarter it was $63.27B, essentially flat sequentially. Year over year, growth slowed to roughly 12%, from 17% in Q2. Management warned in July that Q3 would lap a strong acceleration from last year, so some slowdown was expected. But consensus was $64.85B, and missing by $1.5B on the highest-margin business in the company revives the oldest fear in the Alphabet debate: is AI answering questions instead of ads getting clicked? AI Mode has passed 1 billion monthly users. If Search revenue flattens while AI usage soars, that’s the question the market will price first. YouTube ads also missed at $11.21B vs $11.43B. Two ad engines below expectations in the same quarter is a pattern, not a blip. Cloud took off. Google Cloud revenue hit $30.71B against a $29.1B estimate. A year ago it was about $15.2B, so Cloud roughly doubled, accelerating from 82% growth in Q2. It added about $6B in a single quarter. Last quarter, Cloud backlog reached $514B, with just over half expected to convert to revenue within 24 months. This is now one of the fastest-growing large businesses in the world. So why does the stock fall? Because Search pays for everything. It’s the cash engine funding $195–205B of capex this year, a budget raised twice already. When the funding engine slows while spending keeps climbing, investors see margin pressure ahead. That’s exactly what showed up: operating income of $40.6B was flat quarter over quarter and missed the $43.2B estimate, as depreciation from the build-out piles up. The capex detail. Capex came in at $51.4B, below the $55.8B estimate. Like $MSFT and $TSLA, Alphabet spent less than expected this quarter. Unlike Microsoft, it didn’t get credit for it, because the revenue side of its core business disappointed. Ignore the EPS headline. EPS of $2.12 looks terrible against roughly $3 expected. But Alphabet now marks stakes like its SpaceX position to market every quarter. In Q2 that produced a $99B gain and a $9.11 EPS. GAAP EPS swings with other companies’ stock prices. Operating income is the clean read, and that’s the miss that matters. One more thing to watch: who’s behind the Cloud growth. A large share of AI cloud demand comes from a handful of frontier labs. Anthropic’s IPO filing, for example, shows up to $114.5B committed to Google through 2033, and Anthropic itself disclosed heavy customer concentration. Cloud’s growth is real, but parts of it depend on very few buyers. Bottom line: tonight the market judged Alphabet as a Search company that happens to own a booming cloud business, not the other way around. Cloud doubled. Search stalled. Until Search reaccelerates, or Cloud gets big enough to carry the margin story on its own, the stock will keep trading on the ad line. Cloud is the future. Search is the paycheck. The paycheck slipped.

$MSFT Did What Meta Couldn’t: More Growth, Less Spending Revenue of $92.3B beat $90.7B. EPS of $4.89 beat $4.73. And the number that matters most: Azure grew 48%, against the 45% Microsoft itself guided in July and around 45% the Street expected. The stock ripped after hours. After a week of AI capex anxiety, this report answers the question everyone has been asking: is the spending turning into revenue? Azure is accelerating, not plateauing. Under Microsoft’s new, narrower Azure definition, growth was 42% last quarter. Now it’s 48%, with Azure revenue at $33.2B, up roughly 13% in a single quarter. Microsoft Cloud overall grew 30% to $63.9B. At this scale, acceleration is rare. It means demand is still running ahead of the capacity Microsoft can bring online. The surprise: capex came in light. Microsoft guided to more than $50B of capex this quarter. It spent $39.4B, about $7B below the Street’s $46.6B. Compare that to the past week: $META: capex up again, free cash flow negative. $TSLA: capex pushed into Q4. $MSFT: spent less than planned and still beat on growth. There are two possible explanations, and both matter. Last quarter CFO Amy Hood said efficiency gains in a supply-constrained fleet get monetized almost immediately, so Microsoft may be squeezing more revenue out of each GPU. Or deliveries of chips, power or data center capacity slipped, pushing spend into next quarter. The first is a margin story. The second means the capex bill is still coming. The call should clarify which one it is. The one number to watch: backlog. Commercial remaining performance obligation, contracted revenue not yet recognized, came in at $677B. Last quarter it was $678B. That’s flat quarter over quarter, and well below the ~$706B estimate. Year over year, backlog growth has gone from 110% to 84% to roughly 73% now. Last quarter Microsoft emphasized that all of the sequential RPO growth came from customers other than frontier AI labs like OpenAI. If backlog has stalled, the market will want to know whether big AI-lab contracts are slowing, or whether Microsoft is simply converting backlog into revenue faster than it’s adding new deals. Everything else was solid. Devices and Consumer beat. Search and advertising came in at $6.6B vs $6.4B. Microsoft returned $10.2B to shareholders. The OpenAI line was a small $207M gain, not a drag. Why this matters beyond Microsoft. This is the bull case for the AI cycle in one report: demand accelerating, monetization showing up in revenue, and capital spending under control. For $NVDA, memory names and the neoclouds, an accelerating Azure is the strongest demand signal the market has seen this earnings season. A lower capex number cuts the other way for suppliers in the short term, which is why the call matters. Bottom line: the best hyperscaler print of the season. Azure accelerated past its own guide while spending billions less than planned. The only crack is a flat backlog, and that’s the question for the call. If RPO flattening is just faster conversion, this quarter strengthens the entire AI trade. If it’s slowing deal flow, it’s the first warning sign from the strongest player in the cycle. Growth up. Spend down. Watch the backlog.

$ANT’s $80 Billion Loss Is the Least Important Number in the S-1 Anthropic just filed publicly to list on Nasdaq under $ANT. The headline numbers: $16.3B of revenue in the first half of 2026, up from about $1.1B a year earlier, and an $80.7B net loss. Reports put the target valuation around $2 trillion, which would make it one of the largest IPOs ever. The loss will dominate the headlines. It shouldn’t. The loss is mostly accounting. Of the $80.7B, $72.4B comes from changes in the fair value of convertible notes and similar instruments. As Anthropic’s valuation rose, the value of those notes rose with it, and that increase gets booked as a loss. No cash left the building. The operating loss for the half was $693 million. On $16.3B of revenue, that’s close to break-even, and the company reports adjusted profitability for two straight quarters. Ironically, the higher the IPO valuation goes, the bigger this kind of “loss” gets. So skip the loss. These are the numbers that actually decide the valuation. 1. Growth is fast, but slowing. Revenue went from $4.73B in Q1 to $11.5B in Q2, more than doubling in one quarter. Anthropic then told investors Q3 came in at $16.8B, up 45% from Q2. Still extraordinary, but a sharp deceleration. The S-1 itself warns that growth may not be sustainable. At a ~$2T valuation, Q3’s run rate of roughly $67B a year implies about 30x revenue. That multiple only works if growth stays high for years. 2. Customer concentration. About 20–25% of first-half revenue came from just two customers, unnamed and not locked into long-term contracts. Reports have pointed to AI coding tools as major customers. Cursor was among them, and Cursor is now owned by $SPCX after its $60B acquisition in August. SpaceX is also one of Anthropic’s biggest compute suppliers and a direct competitor through Grok. When a top customer, a key supplier and a rival can sit under the same roof, that’s a concentration risk worth pricing. 3. Compute commitments. Anthropic has committed roughly $518B to compute over about ten years, with around 80% non-cancellable. That includes up to $120.5B for Nvidia GPUs through SpaceX’s AI unit and up to $114.5B with Google through 2033. Here’s the asymmetry: the costs are mostly locked in, while the biggest customers can leave. If revenue growth keeps slowing while commitments stay fixed, margins get squeezed fast. 4. The unit economics are the bull case. Anthropic reports generating about $1.70 in revenue for every dollar spent on compute. Net revenue retention is reportedly around 900%: existing customers spend, on average, nine times more than a year ago. That’s the clearest sign that usage compounds once customers adopt. Why this matters beyond one IPO. Every compute commitment in this filing is someone else’s revenue: $NVDA, $GOOGL, $AMZN, $MSFT, $SPCX. A strong $ANT debut validates the entire AI capex cycle the market has been questioning for weeks. A weak one would hit the same names, from GPUs to memory to neoclouds. The unusual part. The S-1 openly warns that advanced AI could pose catastrophic risks to humanity. And 5% of shares go to a public benefit fund. Anthropic is a public benefit corporation, which legally allows it to weigh mission alongside shareholder returns. New York City’s comptroller has called for the IPO to be paused over safety concerns. Investors should read that section, not skip it. Bottom line: ignore the $80.7B loss. The business is near break-even on an operating basis, with exceptional unit economics. The valuation debate comes down to three things: whether growth stabilizes after slowing to 45% quarter over quarter, how stable the top two customers are, and whether revenue can keep outrunning hundreds of billions in fixed compute commitments. The loss is accounting. The risks are concentration and commitments.

“Black Thursday” Has a Hidden Signal. Watch the Bond Market, Not the Losses. $750 billion gone from US stocks at the open. Nasdaq futures down more than 2%. Asia shed nearly $1 trillion. The Fear & Greed Index at 20, extreme fear. $NVDA, $META and $AMZN down 2–6%. The trigger is clear. President Trump said the US is not negotiating with Iran and suggested action could come soon, and oil surged to around $113. That’s back near the highest level of this war. Since the strikes on Iran in February and the closure of the Strait of Hormuz, oil has been the single variable driving this market. But the most important move today wasn’t in stocks. The bond market flipped. The 10-year Treasury yield fell to 5.00%, the lowest since late September. Remember what happened during CPI week: hotter inflation pushed yields higher and revived rate-hike fears. Oil up meant yields up. Today, oil spiked and yields fell. That’s a regime change in what the market fears: Before: oil → inflation → Fed hikes → higher yields → tech sells off. Now: oil → growth shock → recession risk → money runs to bonds. When investors buy Treasuries while oil is spiking, they’re saying the bigger danger is economic damage, not inflation. Why that matters. A growth scare is ugly for stocks in the short run. But it also makes a Fed rate hike much harder to justify. The 10-year hit 5.2% in September, its highest since 2007, and that pressure on valuations is now easing. If yields keep falling, the rate headwind that hammered long-duration tech for two months starts to fade. Why Asia got hit hardest. The pattern hasn’t changed. Korea is down about 40% from its June peak, the most crowded AI trade in the world, and crowded trades break first when fear spikes. Add “costly AI” anxiety after a week of massive capex numbers from Meta and Tesla, and AI-heavy markets carry a double hit: geopolitics and spending fears together. What extreme fear does and doesn’t tell you. A Fear & Greed reading near 20 has often shown up near short-term lows, because forced and emotional selling eventually exhausts itself. But it’s not a timing tool in a war-driven market. If the Strait of Hormuz situation escalates further, fear can stay extreme for a while. What to watch from here. Oil. Still the master variable. Any credible de-escalation pulls crude down fast, and with it the whole chain. The 10-year. If yields keep falling while stocks fall, it’s a growth scare. If yields reverse higher alongside oil, the inflation and hike fear is back, and that’s the worse combination for tech. Energy and defense. $XLE is the obvious hedge while the war premium stays in oil. Bottom line: today looks like panic, but the bond market is sending a more nuanced message. Fear shifted from inflation to growth, which hurts now but takes away one of the biggest pressures on tech valuations. The direction from here is set in the Strait of Hormuz, not on the Nasdaq. Stocks are panicking. Bonds are thinking. Follow the bonds.

$IBM Is Up 6%. Here’s What’s Real and What’s Acquired. Revenue of $17.1B beat the $16.8B estimate, up 6% in constant currency. EPS of $2.95 beat $2.86. Software revenue came in at $7.8B, ahead of $7.64B. The stock jumped more than 6%. To understand the reaction, start with July. The July scar. On July 14, IBM released a surprise preliminary Q2 letter and the stock fell 25.2% in a single day, its worst day on record. Software growth had slowed to 5%, and by BofA’s math, organic software growth excluding Confluent was close to zero. The market stopped trusting IBM’s growth story overnight. Tonight’s rally is mostly relief: no new crack, and software growth moving back up to 8%. But read the 8% carefully. IBM closed its $11B Confluent acquisition on March 17, so until next March, every year-over-year software comparison includes revenue IBM didn’t have a year ago. BofA estimated Confluent at around $340 million per quarter, roughly 5 points of software growth. If that holds, organic software growth this quarter is closer to 3% than 8%. The Data segment’s 19% jump is largely Confluent too. Better than July’s near-zero, but not a breakout. Where the AI demand actually shows up. Distributed Infrastructure: +31%, after a record +37% last quarter. This is storage and power systems, the hardware enterprises buy when they actually deploy AI on premises. Two straight quarters above 30% is the cleanest organic AI signal in the report. Red Hat (Hybrid Cloud): +11%, steady and organic. Still the crown jewel. The weak spots. Automation: −3%. This segment includes HashiCorp and grew 4% last quarter. A swing to negative is the number bears will focus on. IBM Z: −20%. Mainframes run on product cycles, so a down year after a launch year is normal. It drags the headline, but it’s predictable. Consulting: flat, +2% in constant currency. Enterprise AI talk hasn’t turned into consulting growth yet. The balance sheet context. IBM paid cash for Confluent and now carries $63.8B in debt against $11.1B in cash. Free cash flow guidance is now an increase of more than $1B year over year, and full-year constant-currency growth guidance held at 4–5%. Cash generation is what funds the dividend and pays down acquisition debt, so the FCF line matters more for $IBM than for most tech names. Bottom line: a solid quarter that rebuilds some trust after July, but the growth is partly bought. Red Hat and Distributed Infrastructure show real, organic AI demand. Automation turning negative and organic software running near 3% show the core software engine isn’t fully fixed. The test comes in March, when Confluent laps and the organic number has nowhere to hide. Relief, not reacceleration. Watch organic software.

$TSLA’s Best Number Tonight Isn’t Free Cash Flow. It’s 75%. Adjusted EPS of $0.48 beat $0.44. Gross margin came in at 19.6% vs 17.8% expected. And free cash flow was +$1.53B against a consensus of roughly −$5B. That last one will lead every headline. It shouldn’t. The FCF beat is mostly timing. Capex came in at $4.73B, down from $5.79B last quarter. Wall Street overestimated Tesla’s spending pace for the second quarter in a row: last quarter consensus expected about −$3.3B and Tesla reported −$1.09B. Add it up: Q1 capex was around $2.5B, so Tesla has spent roughly $13B year to date. Management guided to more than $25B for 2026. If that guide holds, Q4 needs about $12B of capex in a single quarter. Either the guide comes down on the call, or Q4 cash flow goes deeply negative. Tesla itself said last quarter that free cash flow could stay negative through the end of 2026. Delivering about 22,000 more cars than it built also likely flattered operating cash flow this quarter. So the cash didn’t appear. The spending got pushed out. The real story: FSD is becoming the business. Over 75% of new Tesla buyers in North America now take FSD (Supervised), up nearly 20 points from last quarter. Active FSD subscriptions hit 1.74M, up 67% year over year and ahead of the 1.57M estimate. FSD is approved in 9 EU countries. Services and other revenue grew 33%. That’s why gross margin jumped from 16.8% last quarter to 19.6%. Software revenue carries far higher margins than selling cars, and as it scales, the margin profile shifts. This is the first quarter where the “Tesla is a software company” argument shows up clearly in the income statement, not just in the pitch. But the operating line is still thin. Operating income was $718M, down 56% year over year, for a 2.6% operating margin. That’s better than Q2’s 1.4%, but it’s still a company spending nearly all of its gross profit on AI, robotics and R&D. GAAP EPS was just $0.22, less than half the adjusted figure, with stock-based compensation, including the CEO performance award, a big part of the gap. The weak spot: energy. Energy revenue fell 16% to $2.88B against a $3.91B estimate. Management says orders and backlog are at record highs and Megafactory Texas just started production, so this looks like a supply and timing issue. But it’s the second straight quarter energy disappointed, after storage deployments also missed expectations. The future-business checklist. Robotaxi: Cybercab rides started in Austin on September 3; unsupervised miles grew 8x quarter over quarter. Optimus: production started on high-volume lines in Fremont. Demand: the largest vehicle backlog in four years, with battery supply, not demand, as the main constraint on volumes. Bottom line: a better quarter than the bears expected, but for a different reason than the headline. The free cash flow surprise is mostly spending pushed into Q4. The FSD attach rate is the real inflection, and it’s already lifting margins. On the call, the two things to watch are whether the $25B+ capex guide holds and whether Tesla can turn 75% attach into a durable, high-margin revenue stream. The cash beat is timing. The FSD beat is real.

$META Didn’t Miss. It Paid a Legal Bill and Kept Spending. The headline says EPS of $3.83, a big miss against the ~$5.89 some had expected. Operating margin fell to 20% from 40% a year ago. That sounds terrible until you open the report. The “miss” is one line item. Meta booked a $9.9B charge tied to its settlement with state attorneys general. This wasn’t a surprise. When the roughly $18 billion, 10-year deal was announced in August, Meta said it expected to book about $10 billion of legal expense in Q3. Add the charge back and operating income was roughly $22.9B, ahead of the ~$20.9B consensus, with operating margin back around 35%. The core business beat. The ad machine is healthy. Revenue grew 26% to $64.6B, beating estimates by more than $1B. Ad revenue came in at $63.06B vs $61.85B expected. The mix is the best part: price per ad rose 14% while impressions grew 10%. Advertisers are paying more per ad, which is the clearest sign that Meta’s AI-driven targeting works. Meanwhile headcount is down 15% year over year. One number to respect: users. Daily active people grew just 3% to 3.64B. At this scale, Meta is close to the ceiling on reach. Future growth has to come from monetizing each user better, which is exactly what the ad pricing shows is happening. Now the hard part: cash. Capex hit $34.91B in the quarter, and free cash flow turned negative at −$2.12B, despite nearly $33B of operating cash flow. Meta also raised the low end of its 2026 capex range again to $135–145B. That’s the third straight quarter the floor has moved up. With $82B in cash and $81B in long-term debt, the balance sheet can absorb it. But this is a company spending everything it earns, and then some, on AI. The quiet headline: a new non-voting Class C stock. Starting January 2027, every Class A and B shareholder gets one non-voting Class C share as a one-time dividend. $GOOGL did the same thing in 2014. The practical effect: Meta can issue stock for acquisitions or employee pay without diluting Zuckerberg’s voting control. For a company spending this much on AI talent and compute, that’s a meaningful tool. What decides the stock tonight. The release didn’t include a 2027 capex number. That’s the figure the entire AI supply chain is waiting for, with buy-side expectations already at $225–250 billion. If the call confirms that range or higher, it’s a strong demand signal for $NVDA, memory and the neoclouds, even if $META itself trades down on cash-flow worries. The Q4 revenue guide of $73–76B is in line, so guidance alone isn’t a catalyst either way. Bottom line: strip out the legal charge and this is a solid quarter with real pricing power. The risk isn’t the business. It’s how much the business is spending. Negative free cash flow at Meta’s scale is a new chapter, and the market will judge it by one question: when does the AI spend start earning a return? The miss is legal. The spend is real. Listen for 2027.