Huang’s Hot Seat
On November 8th, Jensen Huang sent out a letter to Nvidia staff, which was posted on their official website, where Huang admitted to being unfaithful to his wife and apologized to her and their kids; he did not name the woman.
You may be asking yourself, “Okay… why does this matter?” To provide some context, let’s go back to a New York Times report published on October 29th claiming that individuals working in intelligence believe that a Chinese businesswoman close to Huang was passing information to Beijing, but these officials don’t believe that Huang knew of this. This is unconfirmed and only gathered from anonymous sources, while China’s Foreign Ministry claims that the accusations are “unsubstantiated.”
Since this came out, Nvidia’s board has opened an independent investigation into the incident. On November 10th, they said that they do not believe national security or company secrets were compromised, but a shareholder class action was also filed a day before. A shareholder class action is essentially when one or more shareholders sue a company on behalf of a larger group who were harmed; in this case, they are alleging that the company failed to disclose this information to shareholders. CNBC reported that stocks fell more than 4% on Monday and around 6% on Tuesday.
On October 29th, an Nvidia spokesperson claimed that Huang’s interactions with this unnamed woman were “not unusual or unexpected,” which is quite different from what his letter recently revealed. It’s unclear whether she was truly passing information to Beijing, whether any secrets were revealed, or if any laws were broken.
Beating Expectations
AI earnings overperformed with strong numbers this week, but it wasn’t all good. AMD reported $13.38 billion in third-quarter revenue, which is around $300 million above expectations. LSEG puts the earnings at $2.09 per share versus an expected $1.95. AMD CEO Lisa Su said that AMD expects total revenue to grow more than 80% in 2027.
On November 2nd, Palantir reported that revenue rose around 96% to $2.31 billion, and it raised its full-year forecast from $8.15 to $8.53 billion. Despite this, the stock had fallen 1.65% by the time the market closed.
Next up is CoreWeave, a company that rents out computing power for AI. Quite an intriguing business model. They buy Nvidia’s AI chips in large quantities, put them in data centers, and rent them out to companies. They sell to big-name tech companies like Microsoft, Nvidia, and OpenAI. While it sounds quite profitable, which it is, it also means they must borrow – a lot. Their reports were less positive for shareholders; while revenue rose 168% to $3.66 billion, it also reported a $709 million net loss and a $132.5 billion backlog. Their stock fell around 40% over the past month but has since recovered. They also revealed that they have around $46.9 billion of debt, which is why they are being closely watched to see if debt-funded AI as a business model is sustainable.
Ballot Box vs. Data Center
The November 3rd midterms included many decisions on data centers and AI progress, though mostly confined to local ballots and campaigns.
Voters in 10 states decided on data center measures. According to Ballotpedia, 40 of 41 local ballot measures favored limits on data centers. Voters approved 27 limiting measures and defeated 13 that would have helped development.
In a national exit poll conducted by SSRS for NBC News, 56% of voters said AI and data centers were a crucial (23%) or major (33%) factor in their vote for Congress. That measures how much the issue mattered, and it ranked below other issues such as the cost of living or corruption. 76% also said they would oppose a new data center in their community.
Security and Fireworks
GPT-6 Cyber was recently released on November 5th by OpenAI, and it’s marketed as a gated model for vetted security teams through a specific program called Daybreak Red. It’s described as “critical” for cybersecurity by OpenAI’s own frameworks. They claim that the model built nearly one thousand Linux privilege-escalation exploits and found 81 previously unknown high-severity Firefox bugs.
Google recently released Gemini 4 Carbon, which was also released to cyber defenders before the general public. We now have two different labs that held back their most current models from the public due to security concerns. Today, November 10th, Gemini 4 Carbon was made public.
Fireworks AI is a company that runs AI models for other companies. In October, they claimed that no customer models or datasets were accessed, but on November 9th, they said attackers who had gained access on October 24th through a third-party cloud provider reached billing data, secrets, datasets, and partial model weights. It’s reported that the attack was an instance of LLMjacking, a technique used by financially motivated hackers in which stolen cloud credentials are used to hijack AI accounts and models. Our last issue showed that while AI can help make attacks cheap, this issue shows that vendors who use AI as a business model are also targets.
Roundup
That concludes this week: a CEO under investigation, investors who didn’t necessarily cheer for strong earnings, voters voicing their opinions, and a hacked AI vendor. While this week was shaky, we’ll see what the next week brings.