GT Axis Daily, July 9, 2026

The company confirmed it will release GPT-5.6 to the public on Thursday, roughly two weeks after it quietly limited access to a small group of trusted partners at the government's request. GPT-5.6 comes in three tiers: Sol, the flagship built for hard problems, Terra for everyday work, and Luna, the cheap and fast one. The plain read: the newest, strongest AI was briefly kept on a short leash for national-security reasons, and that leash just came off.

SpaceXAI released Grok 4.5, which Musk describes as matching Anthropic's top model while running faster and cheaper. It was trained partly on data from Cursor, the popular coding tool SpaceXAI agreed to buy for 60 billion dollars last month, and it's aimed squarely at software and "agentic" work, meaning AI that carries out multi-step tasks on its own rather than just answering questions. On the company's own scorecards it wins about half its head-to-head tests, so treat the "best in the world" talk as a starting bid, not a settled fact.

Meta's Superintelligence Labs released Muse Image, its first in-house model for creating and editing pictures, now free inside the Meta AI app, Instagram Stories, and WhatsApp. You can hand it a photo, sketch on it, or type a detailed prompt, and it handles readable text inside images, even QR codes. The strategic tell is buried in the announcement: Meta says it plans to use this to reduce reliance on similar third-party technology. Same insourcing move as everyone else.

The company has started replacing some OpenAI and Anthropic models with its own MAI family inside its flagship Office apps, with tens of thousands of weekly requests that used to run on other people's AI now handled in-house. MAI is Microsoft's own model line, built so it doesn't have to pay a partner for every AI task. The point for leaders: the company that helped fund OpenAI is now routing its everyday work around it to save money.

The EU's General Court upheld the decision to classify Apple's App Store and iOS as "gatekeeper" services under the Digital Markets Act, the bloc's law that forces the biggest platforms to open up and give users more choice. Apple had argued its various app stores should be treated separately, and the court said no. Apple can still appeal, but for now the rules stand, and it's also fighting a separate 500 million euro penalty. The bigger message: in Europe, how Apple runs its own platform is no longer fully Apple's call.

GitHub made two upgrades to Copilot generally available in early July: its agents can now open a web browser to click, type, and read what's on a page, and they can see and reason about images and screenshots. GitHub also added Kimi K2.7, a powerful free model from a Chinese lab, to Copilot's lineup, the first China-built open model to make that roster. For a non-coder, the shift is simple: these tools are moving from "suggest the next line" to "go do the task and check your own work."

Apple signed a multiyear deal with Broadcom worth more than 30 billion dollars to produce over 15 billion US-made chips, including the custom silicon Apple needs to build its first dedicated AI server as soon as next year. Broadcom is putting 1.5 billion dollars into expanding a Colorado plant, and the partnership runs through 2031. Translation: Apple wants to design and source the guts of its AI hardware at home instead of depending on distant suppliers, the same own-your-stack logic showing up everywhere this week.

The robots that spent years doing backflips for the camera are now clocking in: Tesla is ramping production of its Optimus robot, Figure's machines are working sequencing tasks at a BMW plant, and Boston Dynamics is deploying its Atlas fleet to Hyundai and Google. Powering a lot of this is NVIDIA's new Jetson Thor computer, essentially a brain small enough to sit inside a robot. The story isn't one launch, it's that "physical AI" is starting to earn a paycheck.

A single tiny device that creates, steers, and reads light-based signals points toward computers that are far faster and sip far less power than today's electronic ones. It's early-stage research, not a product, but the direction matters: "photonics," or computing with light, is one of the credible answers to how we keep scaling AI without melting the power grid.

Together AI raised 800 million dollars at an 8.3 billion dollar valuation, led by Saudi Aramco's venture arm with NVIDIA also in the round. Together runs "inference," the workhorse job of actually answering the millions of daily requests that hit an AI model, and it does it on cheaper open models instead of the expensive closed ones. The company says it plans to grow its computing footprint roughly 50-fold over five years. The signal for leaders: the steady AI money isn't in the headline apps, it's in the plumbing that keeps them running.

Alphabet moved to raise roughly 80 billion dollars, later upsized past 84 billion, specifically to expand the data centers behind its AI, with Warren Buffett's Berkshire Hathaway putting in 10 billion. Alphabet flat out said demand for its AI already outstrips the computing power it has. When a company this size taps the markets this hard, it's telling you the buildout is nowhere near done.