GT Axis Daily, June 22, 2026

Sensor Tower, a firm that measures app usage, put ChatGPT's share of the AI assistant market at about 46%, the first time it has held less than half. Google's Gemini climbed to roughly 28% and Anthropic's Claude to about 10%. The headline isn't that ChatGPT is shrinking, it still has more than a billion monthly users, it's that the field finally has real alternatives people switch to on purpose. The number that should grab a leader: Claude reportedly converts the highest share of its users into paying customers, which matters more for a real business than raw sign-ups.

Anthropic launched Fable 5 on June 9 as the strongest model it had ever shipped, then pulled it three days later under a government export-control order (rules that limit which advanced tech can leave the country). It's been dark for ten days. The twist today: June 22 was the last day of the free-trial window for paying subscribers, so they lost the trial and the access at the same time. The lesson for leaders is blunt: with the most capable systems, the government now has a hand on the switch, and "we shipped it" no longer means "it stays shipped."

A coalition of 42 state AGs opened a formal investigation, with New York's office serving the subpoena (a legal order to hand over records). They want documents on OpenAI's advertising, how it handles user and health data, how it treats minors and seniors, and a quirk called "sycophancy," the tendency of a chatbot to tell you what you want to hear instead of what's true. The timing is the story: OpenAI filed confidentially for an IPO in early June, and a coordinated 42-state probe is a heavier class of legal risk than the usual tech scrutiny. For leaders, this is the moment regulation stopped being a someday problem for the model names and became a front-page one.

Snap opened preorders for Specs, see-through glasses that lay digital images over the real world, priced at $2,195 and shipping this fall. The hook for builders: the glasses run on OpenAI's and Google's AI, and Snap is courting developers with support for AI coding agents to build the apps. At that price it's an early-adopter and developer device, not a mass-market one, but it ships before Meta's comparable product, which is the bragging right Snap wanted.

Android 17 shipped with Gemini built in at the operating-system level rather than as a separate app, including tools that edit video by conversation and generate music from a text prompt. The quiet strategic move: by tying its AI to the Android release itself, Google reaches a billion-plus phones at once and hands third-party developers the same building blocks. Owning the platform the assistant lives on is exactly the distribution edge a standalone app can't buy.

The takeaway flips a common assumption: what decides whether an AI coding session succeeds isn't how good a programmer you are, it's how well you know the actual problem. Across the data, people made the planning calls (what to build) and the AI handled execution (how to build it), and experts got far more useful work per instruction than novices. The practical lesson for any team rolling out these tools: the person in the loop doesn't need to be your best engineer, they need to be the one who deeply understands the business problem. If your AI coding rollout is underwhelming, the fix is probably a domain expert in the chair, not a fancier model.

Quantum computers use the strange rules of physics to attack problems today's machines can't touch, think new medicines or unbreakable encryption. The hard part has always been making them bigger without making them fall apart. Duke University and IonQ linked three separate quantum machines into one working network and shared a fragile quantum state across all three, the building block for stitching many small machines into one large one instead of trying to build a single giant. That "connect many, not build one" approach is how this field most likely gets to a useful scale.

The Nasdaq-100, the index that anchors a huge chunk of passive investing, added five companies today, and the list reads like a map of the AI buildout: CoreWeave and Nebius (which rent out AI computing power), Astera Labs (which makes the connector chips inside AI servers), plus Rocket Lab and Teradyne. Getting added matters beyond prestige: hundreds of funds that track the index now have to buy these names automatically. The signal for leaders is that the "boring" infrastructure layer has graduated into the core of the market, not the speculative fringe.

Applied Digital raised $1.59 billion in secured notes (corporate IOUs backed by specific assets) to fund a new building at its North Dakota data-center campus, adding 150 megawatts of computing capacity, enough to power a small city's worth of AI servers. The point for leaders is the same one this newsletter keeps making: the AI gold rush isn't only apps and models, it's the unglamorous race to build the power and the warehouses everything else runs on, and that race is being funded in the billions.