Democrats Won Congress. The AI Buildout Still Needs Parts

Democrats took the Senate late Tuesday night. They took the House before sunrise. By the time most of us had coffee, the AI trade had already been sold on the headline & bought back. Nasdaq futures dumped overnight. CoreWeave $CRWV got hit for as much as 8% in the pre-market, the Asia chip names got smoked, and then the bid showed up. Dow futures soared a thousand points. The tape panicked, read the fine print & changed its mind.

You can likely guess where we're going with this. The fine print is better for component makers than for anyone else in the AI complex. Here's why.

How we got here

The AI trade was bleeding before a single vote got counted. The Nasdaq-100 fell into a correction on Monday. Goldman's broad AI basket lost 15% in October and sat 24% off its high going into Wednesday. Hyperscaler earnings last week were a nobody-raised story. Microsoft trimmed its capex number to $170 billion from $175 billion and Alphabet held its range. So the market walked into Election Day nervous & then got a political headline dropped on its head.

Here's what the headline really says. Democrats run both chambers starting January 3, 2027. Not Wednesday. January. President Trump keeps the veto, and overriding one takes two-thirds of both chambers. Nobody has that. Nothing with teeth gets signed in the next eight weeks except a holiday card.

The scary version, told straight

We're not going to pretend the bear case doesn't exist. There's a real wing of the Democratic Party that wants to pause data center construction outright. Bernie Sanders filed the AI Data Center Moratorium Act in March. Alexandria Ocasio-Cortez has her own version. Frank Pallone, next in line to run the House Energy and Commerce Committee, called for a national moratorium at a markup over the summer. And Texas just elected Gina Hinojosa, who ran on a data center moratorium & beat Greg Abbott in the state the Street had penciled in for a quarter or more of the next wave of data center power.

That's the asteroid. Now the math.

Why a moratorium is a 2029 problem, not a 2027 one

A national moratorium has to pass a House run by Hakeem Jeffries, who hasn't backed it. Politico's E&E News reported in July that the idea had few takers even inside his own caucus. Then it needs 60 votes in the Senate to beat a filibuster, and Democrats hold 54 with a couple of races still being counted. Then it needs Trump's signature, which it won't get, or two-thirds of both chambers, which it won't get either. Fifty-four isn't sixty, and sixty isn't two-thirds.

Jeffries' own answer to the AI question tells you where leadership sits. He stood up a House Democratic commission on AI in December, put the industry-friendly wing of his caucus in charge of it & took a beating from the left for doing so. The biggest AI bill in the House this Congress, Ted Lieu's American Leadership in AI Act, was written with a Republican, Jay Obernolte. It's a standards & research bill. It doesn't pause anything.

What a Democratic Congress can do in 2027 is hold hearings, fire off subpoenas, make the platforms sweat & move a bill that makes data centers pay for their own grid hookups. That last one has bipartisan legs. It raises the cost of a US data center at the margin. It doesn't change how many optical ports a rack needs.

A moratorium moves a rack. It doesn't cancel one.

This is the part the Election Day sellers skipped. The data center backlash isn't blue. It's local. Abbott, a Republican, paused new data center grid approvals in Texas back in August because voters were furious about water & power bills. This is a NIMBY fight over who pays for the substation. It's bipartisan, and it has exactly one effect on the buildout: it moves the building.

A pause in Texas sends the racks to Ohio or Arizona or wherever the power is. It doesn't send them to zero, because the buyers just told you what they're spending. The four hyperscalers guided to roughly $720 billion to $740 billion of 2026 capex on their own numbers, Amazon's $220 billion included. Microsoft's CFO said spending accelerates in the first half of 2027 because of higher component costs & delayed deliveries. Read that twice. The biggest buyers on earth say their problem is getting parts, not getting permits. As a group they're expected to clear a trillion dollars of capex next year.

If you make the parts, that's your world. Backlogs. Pricing power. Customers who need you more than they need the next zoning board.

The platforms are the ones under the lights

The regulatory heat that's actually on paper is aimed at the app layer. In January, Ed Markey wrote to the CEOs of OpenAI, Google, Meta, Microsoft & three others and asked how they'd keep ads inside chatbots from working users over, kids most of all. A Democratic Senate turns letters like that into hearings. Meta Meta Platforms (META) and Alphabet Alphabet (GOOGL) have ads, teens, chatbots & antitrust files. That's a regulatory surface. A laser doesn't have a privacy policy. An optical engine has never served an ad to a 14-year-old. Nobody hauls a transceiver in front of a committee.

And yet in Wednesday's pre-market the tape bought Meta, Alphabet & Microsoft Microsoft (MSFT) and sold the infrastructure names. The market sold the layer that gets ordered and bought the layer that gets regulated. That's not analysis. That's a reflex. Reflexes are where the trade lives.

Where POET sits

POET Technologies POET Technologies (POET) builds optical engines & light sources designed for the links between AI chips inside a data center. It's a component supplier two or three steps removed from any user, any ad & any chatbot. Its regulatory exposure in Washington is about the same as a fiber maker's. When the next racks get built, a component supplier doesn't care whether they land in Texas, Ohio or overseas.

The indirect risk is real & we'll say it plainly. If a customer's project slips because a permit did, the order slips with it. For a company at POET's stage that's a timing risk, not a thesis risk. The thesis is still the thesis: customer qualifications, shipments, the manufacturing ramp & recognized revenue. The next look at all four comes Monday, November 16, when POET reports Q3 at 7:00 a.m. ET, pre-market, with no call. The first conference call comes with the year-end numbers in March.

Then there's the balance sheet. POET had $796.3 million in cash & short-term investments at June 30. That war chest came from selling stock, not from the bond market, so a Washington fight doesn't touch the funding. In October we ran the math on what the market pays for the business versus the cash. It worked out to 98 cents a share for the entire business at $5.38. We called that crazy then. A regulation scare aimed at the wrong layer doesn't make it less crazy now.

What we're watching

First, the 2027 capex numbers. If the hyperscalers hold or raise them at the next round of guidance, the parts-shortage story holds, and Wednesday's selling in the infrastructure names gets reversed.

Second, Pallone's committee in January. If the first data center bill out of Energy and Commerce is a ratepayer bill and not a moratorium, the bear case loses its teeth.

Third, the Senate count. Fifty-four, fifty-five or fifty-six, none of them is sixty, and sixty is what a moratorium needs. Short of that, it stays a press release.

And November 16. Shipments & the ramp are the story for POET. The party label on Congress isn't.

Subscribe to StockAlpha.ai for the next read on POET and the photonics trade.