Tech stocks plunge after the US midterms

On Thursday 5 November, US tech shares had their worst day since April 2025. The Nasdaq 100 fell 6.3% and the S&P 500 fell 4.4%, which together wiped out about $5.1 trillion in value. The Philadelphia semiconductor index dropped 12.3%, its biggest fall since March 2020. Nvidia lost 9.2% and memory makers did even worse: a DRAM ETF fell nearly 18%.

Three things drove the selling. Democrats won the midterms, and some of the winning candidates had promised to restrict data center construction. Investors are also unsure AI spending will pay off. And Arm and Qualcomm reported earnings on Wednesday night that the market didn't like, sending Arm down 13% and Qualcomm down 9.5%. In Seoul, the Kospi opened 8% lower and trading was briefly halted. SK Hynix fell more than 10%.

For builders: If your compute comes from debt-funded providers, expect financing and capacity plans to get more cautious. It's worth checking how exposed your suppliers are before you sign multi-year commitments.

Democrats set out an AI agenda for the new Congress

Reporting on 6 November described what Democrats plan to do with AI once they take control of Congress in January. Rep. Lori Trahan wants bills that would require independent experts to test leading models against safety standards, and that would spell out when AI companies are liable for harm. Chuck Schumer named data center electricity costs for consumers as a priority. Rep. Greg Casar wants AI executives to testify under oath. In October, House Democratic committee leaders had already asked OpenAI, Anthropic, Meta, Google and SpaceX to describe their security practices and report cases where AI tools behaved in ways their designers didn't intend.

These bills will struggle to become law while President Trump, who has rejected mandatory AI rules, can veto them. Democrats might try to attach AI provisions to must-pass annual legislation. Meta president Dina Powell McCormick said on 4 November that the election results would make data centers harder to build.

For builders: Expect more document requests and hearings in 2027. Teams that already log incidents and run third-party evaluations will find it easier to answer them.

Anthropic's IPO meets a nervous market

Anthropic made its S-1 public on 26 October. It plans to list Class A shares on Nasdaq under the ticker ANT, with Goldman Sachs and Morgan Stanley leading 22 banks. The filing shows first-half 2026 revenue of $16.3 billion, up from $1.16 billion a year earlier. Its first operating profit came in the second quarter, at $974 million. The half-year net loss was $80.7 billion, mostly because of a $72.4 billion non-cash charge on convertible notes. Through a single Class F share, the co-founders keep 50.1% of votes on ordinary matters.

The Wall Street Journal reported that the listing was aimed at the week of 16 November. On 6 November, Bloomberg reported that some potential buyers value the company at as little as $1.1 trillion. A month earlier, expectations had been around $2 trillion.

For builders: A public Anthropic will have to disclose its finances and risks every quarter. That gives customers a clearer view of pricing pressure and how stable the company is.

OpenAI's funding round comes in smaller than planned

TechCrunch reported on 6 November, citing Bloomberg, that OpenAI has raised about $11 billion at a $1.25 trillion valuation. The round hasn't closed, and it's expected to reach about $20 billion within weeks. In September, OpenAI was reportedly seeking $30 billion at $1.4 trillion. On 5 November, OpenAI president Greg Brockman said the company could be ready for an IPO in 2027.

For builders: If funding gets tighter, the big labs will be pushed to make money from what they already sell. API pricing and rate limits may change sooner than they would have.

OpenAI releases ChatGPT Cyber for vetted researchers

OpenAI announced ChatGPT Cyber on 5 November. It's only available to Daybreak Red, the top tier of OpenAI's application-only security program, which is aimed at offensive researchers who hunt for vulnerabilities. OpenAI said the model flagged 948 potential issues in the Linux kernel and found 81 new vulnerabilities in Firefox. It also found a severe bug affecting Ubuntu-based servers, which Debian has patched.

For builders: The strongest security models are being kept behind vetting programs. If your team does security work, apply early and plan around the general-access models in the meantime.

Google takes over Crusoe's Wyoming campus

Crusoe confirmed on 6 November that it has sold its unfinished data center campus in southeastern Wyoming to Google. The project was announced in 2025 and was eventually meant to reach 10 gigawatts. Google didn't disclose the price, but it plans to invest $17 billion in the site. Google had reportedly been worried about cost and schedule while Crusoe was in charge.

For builders: More large-scale capacity is moving to the hyperscalers, so independent neocloud contracts need a close look at delivery dates.

Dropbox buys Tigris Data for $350 million

Dropbox signed an agreement on 3 November and announced it on 5 November to buy Tigris Data, an object storage platform built for AI workloads, for about $350 million in cash. The deal is expected to close in the fourth quarter. Tigris will form the base of DBX Agent Cloud, which starts as S3-compatible storage for AI developers.

For builders: That adds another S3-compatible option for training data and agent files, with a large and established company behind it.

Sources

Talk about it with people building with AI

Our Auckland meetups are open to everyone. Members get a ForgeBox and the members' projects area for NZ$599 a year.

Become a member

Get the daily AI news by RSS