Nasdaq rout spreads to Asia after Democrats win Congress

Learn how the Nasdaq's 5.5% drop spread to Asia and Europe, what OpenAI claims for GPT-6 Cyber, and how OpenAI and Meta answer the data center backlash.

GENERAL Newsletter
H
Hugenode

Investors have started pricing political risk into the AI buildout. Democrats won both chambers of Congress in Tuesday’s midterms after campaigning against data center expansion. On Thursday the Nasdaq fell 5.5%, and on Friday the selling spread to Asia and Europe, led by chip stocks. The companies building AI infrastructure are already adjusting: OpenAI is dropping non-disclosure agreements with public officials, and Meta’s president says building data centers will get harder. Separately, OpenAI released GPT-6 Cyber, a model it says finds and exploits software vulnerabilities for far less than bug bounties have paid.

Nasdaq drops 5.5% as the AI sell-off hits chipmakers in Asia and Europe

The Nasdaq had its worst day in 19 months on Thursday, according to The Guardian. Chip stocks led the slump: memory maker SanDisk lost 22% and hard drive maker Western Digital lost 17%. The Vix volatility index jumped 61% to its highest level since the early weeks of the Iran war. The BBC reports that the Nasdaq lost an estimated $2.5tn in market value and is now in correction territory, at least 10% below its recent high. The so-called magnificent seven shed a trillion dollars in a single session, and Nvidia fell more than 9%.

On Friday Japan’s Nikkei closed down 6%, with memory chip makers Kioxia and Tokyo Electron among the big fallers. South Korea’s Kospi fell more than 8% at one point, which triggered a circuit breaker. It closed down 6% after SK Hynix dropped more than 10%. European losses were smaller because those indexes hold fewer tech stocks. The reported figures depend on when they were taken: the Guardian had the FTSE 100 down 1.9% in mid-morning trading, while the BBC reported a 1.6% fall after the open.

Elon Musk’s companies were hit hard. The BBC has Tesla down nearly 11% and SpaceX down 8% on Thursday, while the Guardian’s live coverage puts the falls at 10.7% and 9%. Musk put $446m into Republican campaigns. Citing Bloomberg’s Billionaires Index, the Guardian says he has lost about $113bn in paper wealth over two days.

Analysts describe two worries arriving together. Susannah Streeter of Wealth Club told the Guardian that a slower data center build-out could undercut the spending behind the boom just as investors grow nervous about returns. Dan Coatsworth of AJ Bell told the BBC that investors see AI spending with “no clear idea of when, or whether, there will be payback.” Not everyone expects a lasting reversal. Jefferies economist Mohit Kumar told the Guardian he does not expect a quick rebound but does not see this as the start of the unwind of the AI trade. He argued that Democrats would be reluctant to let data center jobs move to other states or abroad. Even after the fall, the Nasdaq is still up 4.6% for the year.

What it means: No policy has changed yet. This move is about expectations. Our read is that the practical risk for founders and finance teams is the cost of capital, because investors are now attaching political risk to AI infrastructure. MIT Technology Review reports that banks and lenders were already getting pickier about which data center projects they back. Our Thursday edition covered the data center ballot measures in these midterms.

OpenAI says GPT-6 Cyber found 81 new Firefox vulnerabilities

OpenAI has released GPT-6 Cyber, a model trained to find vulnerabilities, build working exploits and suggest patches, The Decoder reports. The performance and cost figures below are OpenAI’s own. No independent testing has been published.

OpenAI’s headline claims:

  • Firefox: the model found 81 previously unknown severe vulnerabilities in a browser that had already been tested with earlier AI models. It also showed that 30 known issues were more severe than thought. Mozilla accepted the reports and has shipped fixes.
  • Linux kernel: it reproduced 1,724 vulnerabilities and built working exploits for 948 of them. GPT-5.6 Sol managed 166 under the same conditions. OpenAI reported 304 critical findings to the kernel maintainers with suggested fixes.
  • Cost: the kernel campaign cost $1,213,430 in compute, or about $1,279 per exploit. For comparison, programs such as Pwn2Own and TyphoonPWN have historically paid $20,000 to $70,000 for comparable findings.
  • Benchmarks: it scored 48.5% on ExploitGym, up from 32.1% for GPT-5.6 Cyber.

OpenAI rates the model “Critical” for cybersecurity under its own Preparedness Framework, the same rating it gave the base GPT-6 Astra. It says GPT-6 Cyber avoided boundary violations in 99.93% of cases without safety instructions, up from 94.49% for its predecessor, and in 100% of tests with them.

Access is limited for now. A select group of Daybreak Red customers can use the model through the API and Codex, and OpenAI says it will gradually extend access to all Daybreak Red customers. OpenAI also says abuse attempts are rising, including efforts to transfer model capabilities into other systems.

What it means: If OpenAI’s cost figures hold up, finding and exploiting bugs has become much cheaper. That helps defenders who get access and puts anyone who patches slowly at greater risk. The Decoder points out that Epoch AI researchers recently showed the open GLM model could have exploited known but unpatched Firefox vulnerabilities. Our inference is that limiting access to OpenAI’s model will not contain the risk on its own. For security leads, the practical step is the same either way: shorten the time between a fix being released and it being deployed.

OpenAI drops data center NDAs as Meta warns building will get harder

OpenAI says it will no longer use NDAs with public officials on data center projects, The Verge reports. It will keep only narrow protections for security information, trade secrets and competitively sensitive information. Amazon dropped NDAs with government agencies on its data center projects in October. OpenAI also says it will pay the full cost of the electricity it uses and of the infrastructure needed to supply it. It will publish how many jobs it expects to create and report on hiring. Every future “compute campus” it owns and operates will come with a “local compact” reflecting local priorities. Its president, Greg Brockman, said everyone had been surprised by the reaction to data centers, the BBC reports.

Meta president Dina Powell McCormick told the WSJ Tech conference on Wednesday that building data centers is going to be harder, per TechCrunch. She said the industry has not explained the benefits well, and that the opposition reflects how people feel about AI in general. She expects projects to face more review. Meta has data centers in 23 states, and five of them, including one in Texas, are planned at 1 gigawatt or more. She added that Meta gets calls every day from companies wanting to lease its compute.

Texas shows how much is exposed. MIT Technology Review reports that governor-elect Gina Hinojosa pushed for a full moratorium during her campaign. It also reports that outgoing governor Greg Abbott told a key state regulator on September 21 to halt data center permits. Projects under construction include Stargate Frontier, projected at 1,820 MW, and Nexus Hubbard, planned at 6,400 MW and intended mostly to serve Anthropic. The publication cautions that decisions are split across city, county and state officials, and that construction unions that backed some Democrats may resist a moratorium. Following through in office may prove harder than campaigning.

What it means: Our read is that local commitments on power costs, jobs and transparency are becoming part of the price of getting a data center approved. If your roadmap depends on new capacity coming online in Texas, plan for delays.

What to watch

  • Whether Hinojosa moves on a Texas moratorium once in office, and what happens to Abbott’s permit halt in the meantime.
  • Whether congressional Democrats take up Bernie Sanders’ call for a national data center moratorium.
  • Independent evaluations of GPT-6 Cyber, and how the Linux maintainers handle the 304 critical findings.
  • Whether other data center developers follow OpenAI and Amazon in dropping NDAs with officials.
  • Whether the selling continues. IG’s Chris Beauchamp warns there may be much more to come, with a possible resumption of the Iran conflict and the oil price as added risks.

Sources