Not in Washington

The biggest market move of the night did not come from the Senate. It came from Texas, where the Associated Press projected Gina Hinojosa the winner over Governor Greg Abbott. Futures on the Nasdaq 100 fell as much as 2.2% and contracts on the S&P 500 as much as 1.2%, Bloomberg reported. On Robinhood's 24-hour market CoreWeave was off 7%, Oracle and Nebius 4% each. CNBC put it plainly: the Texas governor's mansion going to Democrats “appeared to be the biggest blow to tech stocks.” In Tokyo the Nikkei 225 fell 4.45% to its lowest since May.

Then the night repaired most of it. By about 3 a.m. in New York, Nasdaq 100 futures were down 0.22% and S&P 500 futures 0.14%. The Nikkei closed 1.42% lower; Korea's Kospi, which had swung from a 3.2% gain to a loss of about 4%, closed up 0.62%. The havens kept their gains. The 10-year Treasury yield fell from 4.974% before polls closed to 4.897% in Asian trading, the dollar index dropped 0.6% to a one-month low of 101.27, and spot gold rose 2.2% to $4,016 an ounce. Credit Agricole's David Forrester read the dollar as an AI asset: “The US dollar's Achilles' heel is being exposed.”

Overnight, the market sold the race that can slow more data centers than any other.

Underneath, the market had been pulling two trades apart for weeks. The Goldman Sachs Broad AI basket fell 15% in October, before anyone voted, and stood more than 24% below its peak on Wednesday, according to Vantage Point's Nick Ferres. Banks went the other way: from the end of September to November 3 the KBW Bank Index moved +2.0% while the Nasdaq 100 moved -7.3%. On Wednesday morning Europe's banks were among the best sectors, up more than 2%, and in Milan a trader read the vote as a rates story: it “could ease pressure on the Fed and therefore create more room on interest rates.”

History offers no clean signal. BofA's Michael Hartnett had named this exact pairing, Democrats winning the Senate combined with the Texas governorship, as one that “could lead to a ‘big slump’ in stocks of more than 10% into year-end.” Both conditions were met. Against that, Ned Davis Research counts five of eight midterm cycles since 1930 in which the S&P 500 rose through year-end after the president's party lost the House.

Democrats now hold the House and 54 Senate seats. That is a majority, not a governing one. They lack a filibuster-proof Senate, and every bill still needs the signature of a president who, in the words of JPMorgan's Jahangir Aziz, “has openly said that he's going to turn into Mr Veto.” For banks that splits the policy map in two: what regulators decide on their own, which the election barely touches, and what needs a law, which it mostly freezes.

Capital is the clearest case. Regulators re-proposed the Basel rules in March with a 4.8% cut in core equity requirements for the largest banks, and the Fed's Michelle Bowman said in October: “we're ready now to finalize those rules before the end of the year.” The looser leverage ratio for the biggest banks is already final and has applied since April 1. A Democratic House “can hold hearings and write angry letters,” as Trefis put it. “It can't block a final rule without a law the President signs.” TD Cowen's Jaret Seiberg reached the same verdict on election day: regulators keep easing capital, liquidity and tiering rules whichever party wins.

What changes is the hearing room. Maxine Waters is expected to take the gavel at House Financial Services and Elizabeth Warren at Senate Banking. American Banker spelled out what that buys: the chairs “will control the hearing schedule, can block future Trump administration nominees, dictate most of the witnesses for hearings and issue subpoenas.” Crypto is expected at the top of the list. For banks the cost is time and headlines rather than capital, and the merger window is where it bites. Large deals closed in an average of 126 days in the first five months of this year, against 369 in 2024. Sellers will say “it's clear we've got a two-year window,” Olsen Palmer's Christopher Olsen predicted in October. Waters, Covington notes, has a record of personally testifying against large bank mergers.

For banks, the election moved the hearing room, not the rulebook.

The laws that could still pass are the populist ones. In January President Trump called for a one-year 10% cap on credit-card interest rates; Trefis calls it the one bill that could actually become law, though the odds are low, and Seiberg sees room for rate caps, limits on bank buybacks and a government credit score. Card lenders were the weakest names on Tuesday, which we read as the rate-cap channel: Synchrony fell 2.13% and Capital One 0.98% while the S&P 500 rose. The CFPB, which in September asked the Fed for about $273 million to cover the rest of fiscal 2026, gets hearings rather than money: a Democratic House “can't fund or rebuild the bureau without a signed law,” Trefis notes. Crypto runs the other way. The CLARITY market-structure bill stalled 49-50 in a September Senate procedural vote, dies if it is not passed before the new Congress convenes on January 3, and TD Cowen does not expect it to pass in the next two years. The stablecoin rules under the GENIUS Act are not in play: the OCC's final rule takes effect on January 18, 2027.

Washington cannot stop a data center. A governor can slow one down, and the biggest market just changed hands. Texas holds 27% of planned US data-center gigawatts, according to Wells Fargo, and nearly 30% of the capacity set to come online in 2028, according to SemiAnalysis. Its grid operator, ERCOT, has been weighing about 474 gigawatts of large-load requests, roughly 90% of them from data centers and more than five times the system's record peak. Only 5.9 gigawatts of large loads were actually energized.

The voters had turned first. In August 57% of Texans told the University of Texas/Texas Politics Project that they oppose a data center in their own community; 30% supported one. Nationally, 56% of voters called data centers a crucial or major factor in their decision in the CNBC voter poll, and those who called it crucial backed Democrats 62-34. Quinnipiac found 72% of adults opposed to an AI data center in their community in September, up from 65% in March. On the ballot itself, Stateline counted 33 measures to restrict data centers passed or leading; in Trenton, Ohio, a ban aimed at an Amazon Web Services campus passed 80% to 20%. Behind the anger sits the power bill. In the PJM grid the capacity price rose from $28.92 per megawatt-day in 2024-25 to $329.17 for 2026-27, and the market monitor ties 63% of the 2025-26 increase, or $9.3 billion, to data centers.

The Texas pause predates the vote. The grid connection queue has been under audit since August 3, the state froze environmental (TCEQ) permits on September 21, and in October Governor Abbott extended a de facto moratorium on data-center approvals until at least December 10. What the election changed is who holds the switch. The pause belonged to a Republican governor who could lift it. Hinojosa campaigned on a moratorium “until legislation is passed that gives communities a voice, protects household utility bills, and ends the tax incentives for datacenters.” She is sworn in after the regular session opens on January 12, only Abbott can call a special session before then, and the Legislature stays Republican. Her pledge is a moratorium that lasts until that Legislature passes the law she wants.

Washington has a veto. Data centers get built under governors.

How much is exposed is a matter of opinion. GLJ Research's Gordon Johnson puts roughly 13 gigawatts in Texas immediately at risk: projects with power, permits and steel keep going, while those with power but no state environmental permits slip 12 to 18 months. Jefferies said before the vote that a Hinojosa win would mean an immediate statewide pause and move Texas from Tier 3 to Tier 4 in its framework. PIMCO's Libby Cantrill expected “a lot more sand in the gears of the AI buildout regardless,” even without statewide moratoriums. Texas is not alone either. Bloomberg's analysis of the results in Texas, Ohio, Georgia and Pennsylvania says they “could leave half of proposed AI data centre additions across the US vulnerable to stricter local scrutiny and pushback.” Georgia holds 7% of planned gigawatts, Ohio and Pennsylvania 5% each, by Wells Fargo's count. Federal relief is not coming: JPMorgan's Aziz expects a national moratorium to be vetoed and points to the states instead, where “there have been many states in which limited one-year or two-year moratoriums have been put in place.”

We put the chain from the ballot to the loan on one page: The Texas Switch, every number sourced.

The clearest line from the ballot to a bank runs through Hubbard, near Waco. Lenders led by Morgan Stanley lined up around $15 billion to build a data center there, leased to Anthropic. At its core sits a $14.2 billion one-year facility from Citigroup, Intesa Sanpaolo, Mizuho, Morgan Stanley, MUFG, Royal Bank of Canada and Santander, alongside a five-year revolver, and the bridge has already been funded. The plan was to sell $5 billion to $7 billion of it to high-yield bond investors in September. Then Texas put a hold on new projects. The sale was delayed, Bloomberg reported, and may now be spread over the next year, which would leave the banks holding the debt on their balance sheets longer than they planned.

A bridge loan is a promise to sell. When the sale stalls, the bank becomes the investor.

Banks know what that feels like. In May the Financial Times reported that JPMorgan and MUFG had spent more than six months distributing $38 billion of construction debt for Oracle-leased campuses in Texas and Wisconsin. Man Group's Matthew Moniot put the pattern bluntly: “Banks very quickly start choking.” The exposure sits with the names that won the business. JPMorgan led a $7.1 billion construction loan for Crusoe's Stargate campus in Texas in 2025, and Morgan Stanley's chief executive Ted Pick has called AI financing “existential” for the firm.

So the lenders are buying protection. HSBC has been working on a significant risk transfer tied to more than $1 billion of data-center loans, Barclays is looking for insurance on more than $10 billion of AI-infrastructure credit, and Deutsche Bank on a risk transfer tied to about €2 billion of project-finance debt. The sellers of that protection are filling up: “some insurers have indicated their capacity is close to being exhausted,” Bloomberg reported on October 30. The IMF ran the bad version in October. In its scenario with an AI crash, 22% of global banking assets would sit in 110 weak lenders whose core capital ratio falls below 7%.

The exit runs through the junk market, and the junk market has split in two. The ICE BofA CCC-and-lower index yielded 18.17% at its October 23 peak, with its spread the widest in the three years of daily data FRED publishes, while BB yields, the top tier of high yield, stood at 6.77% on November 2. Investment-grade demand is still deep: Meta sold $15 billion of bonds on election day against about $58 billion of orders. The question for Hubbard is not whether buyers exist but what they charge for a construction timeline that now runs through the Texas Legislature.

Texas reaches the hyperscalers through other doors too. Meta's El Paso campus is financed by a vehicle that issued about $12.6 billion of amortizing notes for a 960-megawatt site, and Meta must begin paying rent on December 1, 2028, with very limited exceptions, “regardless of whether the project is complete.” Our reading: in that structure a Texas delay is Meta's risk, not the bondholders'. S&P had already moved Meta's outlook to negative on October 30. Oracle's five-year credit default swap stood at 292 basis points on November 2, after it filed to sell $10 billion of stock and $10 billion of notes. And Anthropic, the tenant in Hubbard, is heading for a Nasdaq listing as ANT, with first trading possible from November 16; Bloomberg's read of its amended filing put its spending commitments at $895 billion. We have followed each link before: who carries the debt of AI data centers, why their cash flow turns negative first, and how the financing loops back to Nvidia.

The Fed held at 3.75% to 4.00% on October 28 with two dissents, Lisa Cook and Beth Hammack, both preferring a quarter-point hike. On Wednesday morning CME's FedWatch put the odds of a December hike at about 77% to 79%. What eased was the path after December. The hikes priced by the June 2027 meeting fell 3.3 basis points to 55, Deutsche Bank's Jim Reid noted, and the curve split: the 1-year yield rose 2.6 basis points to 4.42% in the early hours while longer maturities fell, and by mid-morning in London the 2-year was down 3.7 basis points at 4.597% and the 10-year down 4.7 at 4.918%. Our reading: the market held on to December and trimmed what gridlock makes less likely after it.

The fiscal clock runs faster than the legislative one. The government is funded only through December 11 under stopgap law P.L. 119-103, with a funding gap from December 12 if the lame-duck Congress does nothing. Divided government raises the odds of a shutdown or a debt-ceiling standoff, TD Cowen's Seiberg wrote. The debt limit, raised to $41.1 trillion in the 2025 reconciliation law, is projected to bind at some point in 2027; estimates of the exact date run from spring 2027 to early 2028 depending on whether they measure the limit or the cash. Some Republicans want a party-line increase in the lame duck, but Senate Majority Leader John Thune said before the vote: “Right now, we don't have the votes to get on a third recon bill,” adding that they might after the midterms. Meanwhile the borrowing keeps coming: Treasury expects to borrow $618 billion in privately held net marketable debt this quarter and $800 billion from January to March.

Congress lost the pen. It kept the purse, and the purse runs out on December 11.

The long end has been the pressure point all year, as we covered in The Term Premium Is Back and Four Billion Against Forty Trillion. A divided government does not change that arithmetic. It changes who gets blamed when a deadline slips.

The election moved power to places with dates attached. These are the ones that decide whether the night's sell-off was a preview or a misread.

Read the list as one story. Washington's part is mostly deadlines and hearings: a funding cliff, a debt limit, a crypto bill running out of time, committee chairs with subpoena power. The parts that move money sit elsewhere: in a Texas session that opens on January 12, in a bridge loan that was meant to be sold in September, and in a junk market deciding what a delayed data center is worth. That is where to look for the midterms' effect on AI. Not in Washington.