GT Axis Daily, July 16, 2026

Thinking Machines Lab, founded by OpenAI's former chief technology officer, released its first model, Inkling, after roughly 18 months of near-silence. It's "open-weight," meaning anyone can download the actual model and run or modify it on their own machines rather than renting access through someone else's website. It handles text, images, and audio natively, and it's built as a "mixture of experts," a design that holds 975 billion parameters, the internal dials a model learns during training, but only wakes up about 41 billion of them for any single task, which keeps it fast and cheap to run. The lab is explicit that this isn't the strongest model available. It's meant to be the best starting point for companies that want to customize one.

According to the Financial Times, NVIDIA built a "white list," an approved-buyer roster that companies only join after passing in-person audits: someone physically visits the data center, verifies the contracts, and interviews the people actually using the machines. Over half of its previous Asian buyers, especially smaller cloud providers, failed and were dropped, though they can reapply. This follows a US Commerce Department ruling in May that widened export controls across the Asia-Pacific supply chain. The plain-English version: NVIDIA is now doing the government's enforcement work itself, because the alternative is being blamed when chips turn up somewhere they shouldn't.

A day after Governor Hochul signed the country's first statewide moratorium on large data centers, the President posted that the state should reverse it "IMMEDIATELY," called it a terrible decision, and said the projects are already being courted by Alabama, Florida, Texas, and Arizona. Hochul's answer: "We hit pause because the communities powering AI should share in its success." Set aside the politics and look at what it reveals. Data centers are now a state-versus-state competition for jobs and tax base, which means where AI gets built is becoming a bidding war between governors.

Cursor, one of the most widely used AI coding tools, shipped version 3.11 with "Automations," agents that stay running in the background and trigger themselves off events, like someone commenting on a code review, instead of waiting to be asked. There's a plain-language setup where you describe the chore you want handled and the tool wires up the triggers itself. The shift worth noticing: this is the move from a tool you open to a process that runs, which is exactly how software quietly turns into headcount.

QuiX Quantum delivered "Carina" to Germany's aerospace agency, and the interesting part isn't the physics, it's the plumbing. Most quantum machines need to be chilled to near absolute zero in specialized facilities. Carina uses light, single photons, and runs at room temperature in a standard equipment rack. QuiX also showed "below threshold" error suppression, meaning it can keep mistakes rare enough that adding more hardware makes the machine better instead of noisier, which is the gate every quantum company has to pass to matter. It now goes through months of integration and benchmarking before anyone can claim it works.

Chairman C.C. Wei said the additional $100 billion will fund at least four more fabs running "2-nanometer" processes, industry shorthand for the current leading edge, where smaller means more computing power per watt of electricity. That brings TSMC's total Arizona commitment to roughly $265 billion, on top of the $165 billion pledged last year. Wei declined to give a timetable, saying it depends on customer demand. For a country that spent a decade worried its most critical chips were made in one geopolitically exposed island, this is the hedge finally getting funded.

The company that manufactures the chips for NVIDIA and Apple posted $40.2 billion in quarterly revenue with profit up 77% from a year ago, beating expectations, and guided next quarter to $44.6 to $45.8 billion, which would be another record. Gross margin hit 67.7%, meaning it keeps roughly 68 cents of every dollar of sales before overhead, extraordinary for a manufacturer. If you have wondered whether AI demand is a story companies tell or a thing that shows up in the books, this is your answer. The stock fell 2.3% anyway.

Governor Hochul's executive order pauses state permits for any new data center drawing 50 megawatts or more, roughly the electricity of 9,000 to 40,000 homes, for up to a year while the state studies the environmental impact and writes new rules. The number that explains the whole decision: nearly 12 gigawatts of data center power requests are sitting in New York's grid queue, with more than 8 of those gigawatts arriving in 2025 alone. That's a decade of demand growth showing up in about a year. The order also floats making data centers pay upfront for grid upgrades so ordinary ratepayers don't. Research labs, universities, and hospitals are carved out.

Intrinsic Power closed the first tranche of its seed round from Kyocera Ventures, Drive Catalyst, Boost VC, and RPV Global, to commercialize software that orchestrates how data centers draw and manage power. It's a small round on a day full of enormous numbers, and that's exactly why it's worth noticing. When New York's grid queue and TSMC's $100 billion land in the same week, the scarce resource isn't chips or capital. It's a place to plug in.