GT Axis Daily, June 29, 2026

GPT-4.5, the model many businesses still lean on, is being switched off inside ChatGPT, with everyone pushed onto the newer GPT-5.5 family. Next up are three more models nicknamed Sol, Terra, and Luna, going out first to a small group of trusted partners and the US government before a wider release. The plain read: OpenAI is moving fast enough that the tool you standardized on six months ago is already yesterday's news, so build for change, not for a fixed version.

Zhipu, also branded Z.ai, has pushed an open model (one anyone can download and run for free) to within striking distance of the top US systems, even with Washington's export limits making life harder for Chinese AI firms. For a business, "open and nearly as good" is a budget conversation: why pay premium prices when a free model does 90% of the job. The catch is data and trust, where the model runs and who can see it.

The maker of Claude raised money at a $965 billion valuation, passing OpenAI for the first time, and quietly filed the early paperwork to go public as soon as this fall. It told investors its revenue is on pace to top $50 billion a year, driven largely by businesses paying for AI that writes software. The takeaway: the company that helps write the code is, for now, winning the money race.

Two of the company's most important AI scientists quit for rivals in a single week, one for OpenAI and one for Anthropic, and one of them co-wrote the original research that nearly every modern AI model is built on. Investors panicked, and Alphabet had its worst day in over a year, shedding roughly $250 billion in value. The lesson for any leader: in this race, your smartest people are the asset, and unlike a factory, they can resign on a Tuesday.

On the same day the big names bounced, chip stocks and some AI infrastructure plays slid, with investors fixating on how much all this computing power actually costs to build and run. It's a shift from "buy anything with AI in the name" to "show me the contracts and the margins." For leaders, that's healthy: the hype is cooling into real accounting.

For two years the move was "tokenmaxxing," basically throwing as much computing at AI as you could afford. Now leaders want the same result for less, and some are walking. One AI startup moved all of its traffic off Anthropic's Claude to a cheaper Chinese open model and watched its costs collapse. The point for any operator: AI is becoming a line item to manage, not a trophy to brag about, and the vendor with the best price-for-quality wins the renewal.

The pitch is simple: a lot of what businesses pay for is AI re-reading the same context over and over, and this company cuts that waste by giving AI a smarter memory. Investors writing a check this size for a cost-cutting tool tells you where the market's head is at. When the hot money chases efficiency, the boom is maturing.

Micron makes memory, the short-term workspace every computer (and every AI system) needs to think. Revenue more than quadrupled from a year ago to about $41 billion, and the company says its high-end AI memory is completely sold out through the rest of the year. When the supplier can't make enough no matter the price, that's the clearest sign of where the real demand is.

ON Semiconductor agreed to buy Synaptics, a company whose chips help devices sense and respond to the physical world, the kind of hardware that powers robots, cars, and smart machines. The industry calls this "physical AI," meaning AI that touches and moves things, not just chats. The signal: the next frontier isn't a better chatbot, it's AI that does work in the physical world.

Researchers at IonQ and Duke linked three separate quantum processors into one working network using light to carry the signal between them. Why it matters: today's quantum machines are small, and the path to useful ones is wiring many together, the same way regular data centers stitch thousands of computers into one. Separately, IBM's newest quantum chip got validated by independent studies. Real progress, still years from your business.

Elon Musk's rocket company signed a deal worth up to $6.3 billion to rent out computing power from its Colossus data center to an AI startup, which will pay about $150 million a month through 2029 for access to top-tier Nvidia chips. It's the fourth giant compute deal SpaceX has struck for that site, after even bigger ones with Anthropic and Google. The point for leaders: owning the warehouse full of chips is now its own business, and a very lucrative one.

Blackstone put $5 billion into a new venture to build data center capacity, aiming to bring 500 megawatts (enough to power a mid-sized city) online by 2027. The twist worth noting: it lets AI customers tap Google's specialized chips without going through Google's cloud. The signal for leaders: the power and real estate behind AI is now so expensive that even the giants are bringing in outside investors to share the load.

The EU's roughly $23 billion plan to build five massive AI data centers is bogging down over funding delays and unclear demand, frustrating the companies it needs as partners. The contrast with the US, where private deals keep closing fast, is the story. The lesson: in the AI buildout, speed and clear demand beat grand announcements, and committees move slower than capital.