Nvidia's Huang Out, Musk Merger Looms, and Rust Finally Gets `!`
8 min read · 15 sources
- Jensen Huang resigned as Nvidia CEO over two extramarital affairs; CFO Colette Kress is interim CEO.
- SpaceX and Tesla merger talks are progressing, with an announcement possible as soon as December.
- Rust 1.100.0 stabilizes the never type (!) after 10 years and five failed attempts.
- Nearly 80% of incident.io's PRs now include agent-generated code, with 4.8x larger PRs.
- A Microsoft 365 outage (MO1550821) hit Outlook, Teams, and Copilot worldwide starting at 1:26 PM ET.
Jensen Huang is out at Nvidia, and not by his own choosing. The founder-CEO resigned after the revelation of two extramarital affairs, according to the Wall Street Journal. Finance chief Colette Kress takes over as interim CEO while the board hunts for a permanent replacement, and long-time director Harvey C. Jones becomes the company’s first-ever chair.
This is the biggest shakeup in the company that owns the AI hardware market. Huang had famously said he doesn’t believe in succession planning, arguing leaders should continuously pass on knowledge. That bet just came due. The board is weighing internal and external candidates, and President Trump has already floated Apple chairman Tim Cook for the job on Truth Social. Cook has the resume – he succeeded Steve Jobs – but he stepped down as Apple CEO just two months ago. The next few weeks decide whether Nvidia’s leadership vacuum becomes a crisis or a clean transition.
The board has appointed Harvey C. Jones as Nvidia’s first-ever chair, and Huang had previously said he doesn’t believe in succession planning.
The Biggest Merger in History Is Getting Real
Elon Musk is doing it again. Bloomberg reports that SpaceX and Tesla are progressing in merger negotiations, with an announcement possible as soon as December. This would be the biggest corporate combination ever – a carmaker and a rocket company under one roof.
For engineers, this is less about the vehicles and more about the balance sheet. Tesla’s cash flow could fund SpaceX’s Starship program without external capital, and Starlink’s satellite network becomes a captive data pipeline for Tesla’s autonomy efforts. The aggressive timeline is classic Musk: boards and advisers are working out governance structure now, with the midterm elections as a political backdrop. If this closes, it reshapes both the EV and launch markets overnight.
A DeepSeek Moment for Biotech
Chinese drugmakers are no longer just copycats. Bloomberg’s feature highlights Akeso’s lung cancer treatment as proof that Chinese pharma has evolved from producing low-cost generics to original breakthroughs. Chinese companies now account for at least a third of the world’s pipeline of innovative drugs and run more clinical trials than any other country.
The catch: China-originated novel treatments have yet to reach overseas markets at scale. Global ambitions run through partnerships with Western firms, who handle late-stage development and sales outside China. Some US lawmakers want to restrict those partnerships, but analysts say the business momentum is too strong to break. The question isn’t whether China invents drugs anymore – it’s whether the West can afford to keep refusing to sell them.
The Next Chapter of mRNA
Moderna’s bet on mRNA is paying off in ways the skeptics didn’t predict. CEO Stéphane Bancel told TIME that the technology, once dismissed as too unstable for medical use, saved an estimated 2 million lives with its COVID vaccine and earned $46 billion from 2021 to 2024. The company has since launched the first approved mRNA flu shot, and in October, Moderna and Merck announced their mRNA cancer vaccine, Intismeran, successfully prevented melanoma recurrence in mid- to late-stage patients.
The speed advantage is the real story: a vaccine can be created against any pathogen within six to eight weeks – less time than it takes to run the safety studies. That compresses the traditional drug development timeline from years to months for infectious diseases, rare diseases, and cancer. For anyone who operates on six-month release cycles, that kind of iteration speed is the point.
Slop Is Moving Up the Stack
Source: gabeorlanski.github.io ↗
AI agents are writing cleaner code, but the mess is migrating. New benchmarks from Gabe Orlanski show frontier agents now solve significantly more SlopCodeBench checkpoints while producing less slop at the code level. The problem: slop has moved up the stack into design decisions. The new LibraryDesignBench and LibraryUseBench reveal that agents lack foresight about how their code will be used – they show poor abstraction design and completely disregard reusing library-provided code.
Solve rates on SlopCodeBench continue climbing across all three test tiers (strict, isolated, and core). But the benchmarks expose a different failure mode: an agent that writes correct code with terrible architecture is arguably worse than one that writes sloppy code you can review. The abstraction problem is the harder one to fix, because it requires judgment about future use cases, not just syntax.
Bad Ideas That Aren't Bad Anymore
A 15-year veteran engineer argues that several “bad ideas” now work, and the most radical is eliminating peer code review entirely. The team now has engineers work with Claude, which opens the PR, and the human reviews the AI’s work before merging once CI is green and Greptile review is clean.
The key insight: the protection code review provided – a second set of eyes – is still in place. It’s just from the AI rather than a human teammate. The author notes that the intuition you develop over a career is worth revisiting as the tooling changes. What was a bad idea in 2015 might be the efficient path in 2026, especially when your reviewer is a model that has read more code than any human ever will.
AI Company Employees Should Unionize
Safety-conscious AI researchers have more leverage than they think. An essay on Astral Codex Ten argues that skilled AI researchers are scarce enough that a dozen top researchers threatening to walk would force companies to negotiate without needing formal union recognition. The National Labor Relations Act protects informal “concerted activity,” so workers can organize without a formal union vote.
The argument gets sharper with the timeline: if recursive self-improvement happens as soon as next year, it would automate AI researcher positions anyway. That means this is the maximum leverage moment. Safety issues like whistleblower protections, auditing requirements, and safety team staffing double as workplace issues. For engineers who worry about what they’re building, the essay makes a pragmatic case that collective action is the only lever that moves companies.
Google's Smartglasses Are Here, Starting at $395
Google is back in the wearable game with Gemini-powered smartglasses developed with Samsung, Gentle Monster, and Warby Parker. Preorders start today. Warby Parker offers three frames starting at $395 in the US; Gentle Monster offers six styles starting at $449 in the US and South Korea. The frames weigh about 41 grams, have no screens, and are controlled via “Hey Gemini” voice commands or a touch-sensitive area on the frame.
The privacy angle is the pitch: Google claims features and limitations to prevent creepy behavior, both from users and the device. The camera feeds Gemini for tasks, and the assistant has access to user data and preferences for personalized answers. At 41 grams, these are light enough to wear all day. The question is whether the privacy guardrails hold up in practice – and whether anyone actually wants to talk to their glasses in public.
First Strike
George Hotz is worried about the endgame. In a short post, he argues that if trends continue, China will have 10x the electricity of the US in 10 years. His proposed solution: sell AI chips to China and use the money to build solar and battery factories. He’s concerned that the US “AI race” mentality could lead to a belief in using ASI for a “decapitating first strike,” which he doubts would work. Hotz says his likelihood of imminent WWIII went up significantly with the nationalization of NVIDIA.
Age and Accelerationism
Source: herecomesthemoon.net ↗
The accelerationist argument has an age bias. An essay on Here Comes the Moon examines the Pascal’s wager-style logic: since death is inevitable without an intelligence explosion, one should support accelerating superintelligence. The author notes this argument is compelling for people in a narrow range of takeoff timelines – particularly those who are older or have short timelines. Modern AI systems can now write novel mathematical proofs and develop software with minimal human supervision, making the future arrive faster than expected.
incident.io's AI Spend, In Numbers
Source: blog.pragmaticengineer.com ↗
incident.io shared detailed AI spending data with a new “effort point” concept to analyze ROI. Nearly 80% of PRs shipped to production now include agent-generated code. Compared to a year ago: 3.4x more PRs shipped, 4.8x larger PRs (from 22 to 111 lines), and 6x more “effort points” shipped. The shift began in June 2025 when Claude Agents launched, worktree tooling improved, and engineers started running multiple agent sessions in parallel.
Rust 1.100.0: The Never Type Finally Ships
Ten years and five failed attempts later, Rust 1.100.0 stabilizes the never type (!). This is the canonical uninhabited type for diverging computations, and now you can name it anywhere, including in generic parameters like Result<T, !>. The release also changes Infallible to be an alias to the never type, which required a breaking change to the “never type fallback” from () to ! in editions before 2024. The release also includes the Allocator and DropGuard APIs and conjuring zero types. If you’re on an older edition, check the migration notes before upgrading.
Microsoft 365 Goes Down, Worldwide
Source: bleepingcomputer.com ↗
A massive Microsoft 365 outage (tracked as MO1550821) hit Outlook, Teams, Copilot, Exchange Online, SharePoint Online, and Defender XDR worldwide, starting Thursday at 1:26 PM ET. Microsoft determined the cause was servers unexpectedly becoming unavailable, not increased traffic, which strained remaining infrastructure. The company paused deployments, redistributed connection traffic, and restarted affected servers. A separate earlier Teams incident (TM1550488) lasted from 9:45 AM to 10:53 AM ET due to resource contention. If your org runs on M365, today is a good day to test your offline fallbacks.
You May Also Like
Nvidia's Board Hunts for Huang's Replacement as Agents Take Over AWS
Nvidia's board has hired an outside firm to identify CEO candidates as it weighs ousting Jensen Huang following reports of extramarital affairs, including one …
Nvidia's CEO is gone, $1T vanished, and Apple wants Siri to actually do things
Nvidia CEO Jensen Huang resigned after admitting an affair with a woman US intelligence identified as a Chinese spy, wiping out over $1 trillion in market value …
Anthropic ships Claude Duet, Musk pushes Tesla-SpaceX merger, and zero-days hit PHP and Node
Anthropic launched Claude Duet, an enterprise-oriented autonomous agent connecting to over 6,000 tools with built-in financial guardrails via Stripe and …




