Long durable scarcity. Short abundance.

Yesterday the market had its best day since April. This morning one inflation report took a piece of it back, and Nvidia fell again.

Prices ran hot. October consumer prices rose 0.4% against 0.3% expected. Core prices, without food and energy, rose 0.3% against 0.2%. The 10-year Treasury yield climbed about 8 basis points to near 4.95%. Traders now put the odds of a December rate hike above 80%, and the odds of a half-point hike went from zero to about 15%.

The Fed had warned us. Governor Waller said this morning, before the report, that a 0.3% core reading means "a 50 basis point hike in the policy rate should be considered." In his words, "Hiking only 25 basis points would put our credibility at risk."

This is the part of the AI story that people skip. The Fed does not have to touch a single chip to slow the build. Higher rates raise the cost of the debt that pays for data centers. Companies that pay with their own cash feel it least. Companies that borrow to build feel it most.

Nvidia, day two. The stock was down about 5% late this morning, near $169, after a 4% drop on Monday. The new piece is a report from The Information: at least $7.5 billion of Nvidia hardware reached Chinese companies through a network linked to Alice Huang since 2023, per two people familiar with the matter. This is a report, not a finding. Nvidia has said the Megaspeed diversion claims were investigated and debunked last year. A shareholder class action was also filed in San Jose.

Here is my honest read. On Sunday I said a drop on the headline alone looks like a sale price, and that the view dies on real evidence of a leak or an export failure. This report is aimed at exactly that second question. It is not proven, so my view stands for now. But the stock is no longer trading on gossip. It is trading on an export question, and that one can cost real money.

Anthropic set its price. It plans to sell 825 million shares at $70 to $78, raising up to $64.35 billion. At the top of the range, the fully diluted value is about $1.4 trillion, up from $965 billion in May. Seven companies, including Nvidia, AMD, Broadcom, Google, Microsoft, Micron and TSMC, plan to buy $26 billion more at the IPO price. Pricing is expected November 18, with trading November 19 under ANT on the Nasdaq. Our read: this is funded demand. The cash pays compute bills, and much of it flows back to the suppliers who are buying in.

Physical bottlenecks keep shipping. AAON's data center cooling sales rose 218% to $397.2 million last quarter, and the stock was up more than 20% this morning. One caution: AAON also restated its first-half results.

At 11:40 a.m. ET the S&P 500 was down about 0.75%, the Nasdaq about 1% and the Russell 2000 about 1.3%.

Tonight: CoreWeave reports after the close. Its own guide is $3.45 billion to $3.60 billion of revenue, with interest costs of $860 million to $940 million for the quarter. In a week like this, the interest line matters more than the sales line.

Research and opinion, not investment advice.

Stocks bounced after Thursday's crash, but the bounce is fading. By about noon ET the Nasdaq was up about 1.5% at 24,679, the S&P 500 up about 0.8% at 7,413 and the Dow flat at 51,539. Chips led the early move. The semiconductor index rose about 5% in the first hour after its 12% fall on Thursday, and Nvidia was up about 6.5% before 10 a.m. By noon Nvidia was up about 3.9%. The VIX, the market's fear gauge, fell about 16% to about 23.6. Trading was light, so this is not yet proof that big buyers are back. The Nasdaq is still about 10.6% below its October 6 high, so it is still in a correction.

The bigger news came at 8:30 a.m. The U.S. added 165,000 jobs in October. Forecasts were near 65,000. August and September were also revised up by a combined 84,000. Unemployment was 4.1%, and average pay rose 0.4% in the month and 3.1% over the year. That is good news for workers. For markets it raises the chance that the Federal Reserve raises interest rates again in December. Traders put those odds at about 70% to 78% this morning, depending on the source and the time. The 10-year Treasury yield sat near 4.86% to 4.88%.

Here is why that matters for the thesis. The AI build runs on borrowed money. Data centers, power deals and chip orders are paid for years before they earn anything. When rates rise, every one of those projects costs more, and a late project costs more again. On Thursday the market priced the risk of slower permits. Today it priced the risk of dearer money. A Barclays credit strategist put the new shape of the race well: scarce power "favors the biggest balance sheets." Local pushback makes the build more expensive, not impossible. The companies that can pay cash win. The ones that need loans pay more.

IREN is the test case, and it rose about 19% today. It said on Thursday evening that its AI cloud revenue reached $200 million in the quarter, up from $71 million the quarter before, with a $20.3 billion backlog of signed contracts. Its target of more than $4 billion in yearly AI revenue is contracted and funded, but the start moved to January and possibly February. The open question is the next layer. Of about $15 billion it still plans to spend this fiscal year, about $9 billion is in hand or committed. About $6 billion is still being arranged. One new five-year contract has the customer prepaying about 90% of the chip cost, which is the best defence against dear money. Real revenue earned the bounce. The $6 billion is the next test, and a rate hike makes that test harder.

The AI IPO got cheaper too. Anthropic aims to list as soon as the week of November 16. Investors now talk about a value between $1 trillion and $1.3 trillion, according to Bloomberg and The Wall Street Journal as reported by Yahoo Finance. Last month buyers talked about $2 trillion. Anthropic's filing shows $33.2 billion of revenue for the nine months to September 30. Nothing in that filing changed this week. The price people will pay for it did. That is the same lesson AMD taught on Thursday: a price is a guess about the future, and the guess just got more careful.

Firefly signed a rocket-engine deal. Firefly says AST SpaceMobile will pay more than $1.3 billion, in cash and AST shares, for 36 engines, a licence to build them and three years of engineering help. Firefly rose about 30%. It is a sign that launch engines are scarce outside SpaceX. Shares are part of the price, so the final value depends on AST's stock. SpaceX itself was up about 7%.

Cloudflare beat. Revenue rose 39% to $782.5 million, and the stock rose about 11%. It still reports a loss under standard accounting rules.

DraftKings fell about 27%. Revenue grew 16%, but it will spend $650 million to $700 million this year building a prediction-markets business, and it expects almost no revenue from it this year.

Shoppers feel a little better. The University of Michigan's early November sentiment reading rose to 51.0 from 48.0. Still, 57% of people said high prices are hurting them. The survey ended before the election result.

What to watch: the close, the Fed's next words on rates, Anthropic's final price, Firefly's call on November 10, and whether IREN closes its $6 billion.

Research and opinion, not investment advice.

Yesterday's election rally is gone. By late morning the Nasdaq was down about 5%, the S&P 500 about 3.4% and the Dow about 1,270 points, or 2.4%. The Philadelphia semiconductor index, the main chip-stock benchmark, fell about 11%. The VIX, the market's fear gauge, rose about 55%. On Wednesday the Dow had closed up 1,569 points, its best day since April 2025.

The selling hit the AI build hardest. Nvidia fell about 8%, AMD about 13% and Micron about 16%. Sandisk fell about 17%, so the memory trade lost most of its recent gains. The companies that rent out AI computing power fell further: in the morning CoreWeave was down about 9%, Nebius about 13% and IREN about 11%, and IREN was later down about 20%. The selloff reached Asia too, where SK hynix fell about 9% and Samsung Electronics about 7% in Seoul.

Here is what we know and what we do not. The fact is a large, fast repricing. No data-center contract was cancelled today, no chip order was withdrawn and no new law passed. The cause is the market's reading of the election. With Democrats now running both chambers, and Texas's new governor backing a pause on new data centers, investors now doubt how fast the AI build can move. One analyst, Dan Ives of Wedbush, estimated that 800 to 1,000 U.S. data centers could stall over the next two years. That is one analyst's estimate, not a count of real projects. Part of the chip drop has a separate cause: Qualcomm fell about 14% after a weaker smartphone forecast, and that has nothing to do with AI.

Tesla and SpaceX belong in this story, but only in part. SpaceX was down about 10% and Tesla about 9% to 10% (SpaceX about 7.6% at 9:56 a.m. ET). The main reason is political. Democrats have said they want to look hard at Elon Musk's role as a federal contractor, and SpaceX depends on government launch and satellite contracts. A planned Tesla and SpaceX merger also now looks harder to approve. The second reason connects to our thesis: SpaceX's AI ambitions need data centers, and those now face the same question about permits and power as everyone else. Musk also spent heavily in the Texas races, and his side lost. So the Musk drop is politics first and the data-center question second.

That data-center question is the thesis. We look for things the world needs and cannot quickly make more of. First the scarce input was chips. Then it was power. Today the market priced the next one: permission to build. A chip that cannot be plugged in earns nothing. A site with power connected and permits in hand can be leased for a premium, as Riot showed on Wednesday with a 15-year lease of its powered Texas campus worth about $28 billion over the initial term.

When a building is late, someone pays. There are three parties: the tenant (an AI lab or a giant like Microsoft), the builder or landlord (a neocloud or developer), and the lender. The tenants have the power. According to The Information, as reported by several outlets in October, tenants now ask for more than $5 million a day in late-delivery penalties on new AI data-center leases, up from about $1 million in the spring. Insurers have started to sell delay policies to cover those fees. A neocloud is a company like CoreWeave or Nebius that rents out AI computing power and borrows to build it.

The big tenants also protect themselves. In September, Oracle sent a force majeure notice on its New Mexico campus. A force majeure clause excuses a party for delays caused by events it does not control. Oracle's notice lets it pay lower, pre-opening rent for longer if power arrives late. Oracle says the project is on schedule, and its landlord says the financial commitments are unchanged. Google now backs some of Anthropic's lease payments at TeraWulf's Kentucky site, but the support starts only when each data hall is finished. So even with a strong partner, the builder still carries the cost of being late.

This is the split we flagged last week. A hyperscaler like Amazon or Google pays for data centers from its own cash flow. If one site is late, it waits or builds elsewhere. A neocloud borrows against each site opening on time. Its interest is due whether the building is ready or not. A delay is an inconvenience for the first and a financing problem for the second.

What would prove us wrong: projects keep their permits, sites open on time, and the neoclouds keep raising money at today's cost. If that happens, today was a panic, not a repricing. We will watch Texas permit decisions, utility connection approvals, the next neocloud debt raise and whether any signed lease is delayed or cut.

Research and opinion, not investment advicw.

Democrats won both the House and the Senate, according to AP projections. The official count is still under way. President Trump said he would accept the result. By late morning the Dow was up about 1,480 points, or 2.9%. The S&P 500 was up about 1.8% and the Nasdaq about 2%. The VIX, the market's fear gauge, fell about 17%.

The simple reason is gridlock. When one party no longer controls Washington, fewer big laws pass, and markets like fewer surprises. Removing the risk of a disputed result helped too. One strategist put it plainly: earnings matter more than elections. That is the right frame for the rest of the year.

The more important story is underneath the index. The rally was not even. The biggest cash-generating tech companies jumped. Apple and Microsoft each rose about 6% to records, Amazon about 7% and Alphabet about 5%. The companies that rent out AI computing power on borrowed money fell. CoreWeave was down about 8%, and Nebius and IREN also fell. Nvidia slipped.

The reason is Texas. Democrat Gina Hinojosa won the governor's race and has pledged a pause on new data-center construction. Texas holds roughly a quarter of America's planned data-center capacity. A pledge is not a law, and nothing has been enacted. But investors now have to price a new risk: that building a data center depends on a permit, not just on money and chips.

This is the split we flagged last week. A hyperscaler is a company like Amazon or Google that runs its own giant data centers. It pays for them from its own cash flow and has sites around the world. If Texas slows down, it builds somewhere else. A neocloud like CoreWeave rents out computing power, borrows to build, and needs each site to come online on schedule to pay its lenders. A delay is an inconvenience for the first and a financing problem for the second. Credit markets already see it: the cost of insuring AI borrowers' debt has risen as much as 80% since August.

For the thesis, this adds a new kind of scarcity. We look for things the world needs and cannot quickly make more of. Chips were scarce, then power. Now a data-center site with power connected and permits in hand may be the scarcest input of all. Owners of ready sites gain pricing power. Builders who still need approvals carry more risk.

The rest of the market reads the same way. Solar stocks jumped, with First Solar up about 10%, on hopes of friendlier energy policy. Hospital operators and health insurers rose. Defense stocks fell. The 10-year Treasury yield eased to about 4.90%.

What would prove us wrong: Texas projects keep their permits, construction stays on schedule, and the neoclouds keep raising money at today's cost. If that happens, today's selloff was an overreaction, not a repricing. We will watch Texas permit rulings, utility connection approvals and the next neocloud debt raise.

Research and opinion, not investment advice.

This afternoon we said the quarter was not the test, the forecast was. AMD passed both.

Third-quarter revenue was $13.4 billion against about $13.1 billion expected. Adjusted earnings were $2.09 a share, 15 cents above estimates. Data-center revenue nearly doubled to $8.5 billion, up 97% from a year ago. The weak spot was PCs and gaming, down 8%.

The forecast is the bigger number. AMD guided fourth-quarter revenue to about $17 billion, roughly $700 million above what analysts expected, which would be about 66% growth from a year ago. Management called it accelerating growth driven by data-center demand.

That settles the Intel comparison. Intel's quarter was hurt by PCs and by the cost of its factories. AMD's PC business also shrank, but it owns no factories and its AI chip sales are growing faster than anyone modelled. The two stocks fell together last week. The businesses did not.

For the thesis, this is funded demand showing up in a supplier's numbers. Somebody is paying for this much AI hardware, and AMD is now capturing a larger share of it. The open question stays the same: whether the buyers, many of them borrowing to build, earn enough on that hardware to keep ordering at this pace.

Our earlier instinct that AMD would miss because Intel missed was wrong. The call we published, that the forecast was the real risk, held up, and the forecast was strong.

Research and opinion, not investment advice.

AMD reports after the close today. The stock fell about 18% over the two sessions before, then bounced about 5% this morning. Our read: the market priced in a miss before the numbers existed.

The bounce came from a report, not from AMD. Korea's Maeil Business says AMD and Nvidia have booked Samsung's second plant in Taylor, Texas, to make the base layer of their high-bandwidth memory, the stacked memory that feeds AI chips. Samsung would not comment. Equipment goes in from 2029 and volume production starts in 2030, so this changes nothing tonight. It does show AMD lining up a second U.S. supplier besides TSMC.

Intel set the mood. Its data-center chip sales rose 78% last quarter, but weak PC sales and the cost of its factories dragged the forecast down, and it is raising $23 billion in new stock. AMD is a different business. It owns no factories, so it does not carry that bill. It has beaten revenue estimates for fifteen straight quarters, and it guided to about $13 billion this quarter.

So the quarter is not the test. The forecast is. If AMD guides the fourth quarter below what investors hoped for its AI chips, the selling was early, not wrong. A strong guide would mean the market sold AMD for Intel's problems.

SpaceX is reportedly in talks with EchoStar about Boost Mobile. Nothing is signed. Boost has about 7.5 million subscribers but rents its network from AT&T and T-Mobile, and those contracts may stand in the way. The logic fits what SpaceX has done all year. It keeps buying spectrum, the licensed airwaves no one can make more of. Customers would give that spectrum something to carry.

The Treasury reset Series I savings bonds to 5.51% for the next six months. The part that matters is the fixed rate, 1.40%, the highest in about twenty years. That is a guaranteed return above inflation for as long as the bond is held. When savers can lock that in without risk, a stock has to offer more to win their money. It is the same message as Monday: capital has a price again.

Research and opinion, not investment advice.

The market split in two today. The S&P 500 fell 0.5%, but the average stock did better: the equal-weight S&P 500 rose 0.8%. Money moved out of AI hardware and into companies that sell to consumers.

The backdrop turned friendlier first. U.S. crude fell 7.6% on renewed hopes for U.S. and Iran talks. Long Treasury yields eased. The October ISM manufacturing index rose to 55.5 against 54.6 expected, its tenth straight month of expansion, while its prices index fell 2.7 points. Cheaper energy without weaker growth is an unusually good mix for stocks.

The AI build took the other side. Oracle fell 8.8% after reports of a roughly $20 billion raise, half stock and half bonds, to fund its data centers. Marvell agreed to buy Lumentum in an all-stock deal near $115 billion. Marvell fell 12.7% and Lumentum 18.6%. AMD, Micron, Dell and Intel fell between 5.8% and 8.8%.

So where did the money go? Consumer stocks rose 2.4% as a group. Home Depot gained 3.6%, Walmart 3.4%, Costco 2.8%, Amazon 2.2% and Wingstop 8.9%. The market paid for revenue that does not depend on someone else's capital spending.

Palantir reported after the close and beat. Revenue rose 96% to $2.31 billion against about $2.18 billion expected, and full-year guidance rose to about $8.53 billion. Palantir gets compared with the consumer AI names, but it sells to governments and enterprises. U.S. government revenue was $953 million and U.S. commercial revenue $984 million, up 148%. Those customers buy software to run operations, and defense budgets do not move with consumer sentiment.

Inovio rose 215%. On Friday evening the FDA granted accelerated approval to VOCERYQ, its treatment for recurrent respiratory papillomatosis. It is the company's first approved product. Inovio expects to take orders in about 60 days and to finish its rollout in the first quarter of 2027. The stock closed near 60 cents on Friday, priced for failure, and a binary event went its way.

The thesis has not changed. Scarcity still wins over time. Today showed that the price of scarcity includes the cost of the capital that builds it, and that when oil and rates fall, investors have other places to find growth.

Research and opinion, not investment advice.