The AI boom is sold out through 2028, and it isn't the models
The chip toolmaker everyone depends on can't keep up. Plus: Gemini 4 Argon goes wide, and Huawei quietly closes in on Nvidia in China.
The one
Chips & InfraWidely covered
ASML says it can't build enough chipmaking machines for AI until at least 2029.
ASML, the only maker of the machines that print the most advanced chips, said today its planned output for 2027 and 2028 will not fully meet demand. Customers are already asking about orders for 2029 and 2030. It raised its 2026 sales outlook after a €11.7 billion quarter.
ASML machine supply, 2026 to 2030
Our take
Everyone is arguing about which model is smartest. This sets the ceiling on all of them. If the machines are sold out for two years, compute stays scarce, and the labs that locked in capacity early keep their lead.
Google priced it at $2/$10 per million tokens until November 30, then double, and it gives wrong answers far less often than GPT-6 Astra (15% vs 51%) but also gets fewer right. Great for work where a confident wrong answer costs money; test it on your own code before switching.
IDC says Nvidia's share of China's AI chip market fell to 34% in the first half of 2026, with Huawei at 28%, and Nvidia's shipments there dropped 44%. Export controls are building Nvidia's toughest rival in real time.
4. Meta's Muse Wonder, day two: the reviews are split.
We led with it Monday and it was still everywhere today. Coders like it, but reliability is uneven and Meta's AI chief has acknowledged rollout problems. At $4/$20 per million tokens it's cheaper than the top tier, not cheap.
Some models handled individual steps of copying themselves to other machines, and one finished the full chain only in a simplified test that skipped some security barriers. Real enough to patch your smart devices, not a reason to panic.
6. Exposed AI infrastructure jumped 63% in nine months.
Censys counts nearly 268,000 internet-facing systems running AI tools, many wired into sensitive company data. Your team probably stood up one you don't know about.
8. Microsoft is betting on its own chip, Maia 300.
Microsoft claims 40% more output per watt than top GPUs and plans multi-gigawatt deployments starting in 2027. Read it next to the ASML news: every big cloud wants off the Nvidia waitlist.
Safety & SecurityOnly security outlets caught this
9. LiteLLM flaws could hand your AI gateway to any logged-in user.
Check Point found bugs that let low-privileged users take admin control and steal provider API keys. If LiteLLM sits in front of your models, check Check Point's report for the fix and rotate your provider keys today.
10. Google opens its science models to 200 European researchers.
Access to AlphaFold 3, AlphaGenome and other frontier tools comes with $3 million in compute credits. Small money, but a smart play for goodwill as Europe writes its AI rules.
