GT Axis Daily, July 12, 2026

GPT-5.6 comes in three sizes: Sol for the hardest problems, Terra for everyday work at a lower cost, and Luna for fast, high-volume tasks. OpenAI says Sol beats rival models while using fewer computing resources per answer, which in plain terms means cheaper answers at the same quality bar. The family is also now the default AI running inside Microsoft 365 Copilot, tightening the two companies' ties even as their broader relationship gets more complicated.

Elon Musk described it as being in the same league as Anthropic's top model, just faster and cheaper to run. The model was trained partly using data from Cursor, the AI coding tool SpaceX is in the process of buying for $60 billion, which is why this release doubles as a preview of what that deal is meant to produce.

Muse Spark 1.1 is Meta's new model built for "agentic" work, meaning it can carry out multistep tasks on its own instead of just answering one question at a time. Developers pay per use, at roughly a quarter of what OpenAI and Anthropic charge for comparable models. Mark Zuckerberg broke a three-year silence on X to announce it himself, a signal Meta wants this noticed.

The lawsuit, filed in federal court, claims OpenAI coached departing Apple employees on evading security checks and used Apple's own project code names during recruiting, all while building consumer AI devices through the former Apple design studio it bought last year for $6.4 billion. Two years after Apple built ChatGPT into the iPhone, its former partner is now a named defendant. OpenAI has denied the allegations.

Tens of thousands of weekly requests inside Excel and Outlook that used to go to outside AI are now handled by Microsoft's own MAI models instead. Microsoft says the harder, more demanding jobs will still go to OpenAI's models, this is about routing the cheap, high-volume stuff to software Microsoft owns and doesn't have to pay a partner for.

Kimi K2.7 Code, built by Moonshot AI, is now a selectable option inside Copilot, and it's "open-weight," meaning anyone can download, inspect, or run it themselves instead of only renting access. It's a lower-cost alternative sitting right next to Copilot's usual paid options in the same picker, giving developers and the companies that pay for their seats a cheaper lever to pull.

GPT-5.6 replaced the prior model as the preferred option in Microsoft 365 Copilot, the AI assistant bundled into Word, Excel, and Teams for paying business customers. For a leader deciding what AI tool to standardize on at the office, this is the version showing up by default starting now.

Quantum computers need somewhere to hold information while they work, the equivalent of a regular computer's RAM. ETH Zurich built that memory out of tiny vibrating components instead of the usual magnetic fields, packing more storage into less space and holding information for longer before it degrades. Early days, but "more memory, smaller chip" is exactly the kind of unglamorous engineering that turns lab demos into real machines.

SK Hynix listed its US shares on the Nasdaq, raising $26.5 billion in the second-largest share sale ever, behind only SpaceX's IPO last month. The same day, NVIDIA and SK Hynix announced a multiyear deal to build faster memory chips together for AI data centers and robotics hardware. NVIDIA's CEO called SK Hynix its largest memory partner and said that continues.

Iris, built with Broadcom and manufactured by TSMC, passed testing and enters production in September, part of a plan to roughly double Meta's AI computing capacity by 2027. The point of building a chip in-house is simple: less money paid to NVIDIA and AMD for every unit of computing power Meta needs.

The Alberta facility will draw 1 gigawatt, roughly enough for a small city, and costs more than $13 billion Canadian. Meta says it will run on a closed-loop cooling system that uses no water for operations and will match its electricity use with clean power. The project is Meta's 33rd data center worldwide and its biggest outside the US.

The 20-year deal, called Project Kilby, pairs a roughly $7 billion power plant in West Texas with a Microsoft-run data center campus, delivering about 2.67 gigawatts once built out, enough for well over a million homes. Power starts flowing in 2028. The arrangement sidesteps a strained public power grid by building dedicated generation next door instead.