GT Axis Daily, July 19, 2026

Moonshot AI, a Beijing startup, released Kimi K3, an "open-weight" model, meaning anyone can download and run it themselves instead of paying to rent access. At 2.8 trillion parameters it's the largest open model ever released, and Moonshot says it performs close to Anthropic's best paid model while badly beating everything else, all at a fraction of the price. Investors immediately drew the comparison to last year's "DeepSeek moment," when a similar free Chinese release made everyone question whether US labs can charge what they've been charging.

Starting July 20, Anthropic is folding Claude Fable 5 into its Max and Team Premium subscriptions at half the usual usage limits, while Pro and Team Standard users lose free access entirely after a one-time credit. Translation: the company ran short on the computing capacity needed to keep giving away its best model and is now rationing it like a scarce resource, because it is one.

Gemini 3.5 Pro was promised in June, then delayed again after internal testing showed it falling short on coding and multi-step reasoning, the kind of AI that plans and executes a task rather than just answering a question. Google is now testing it quietly with partners alongside an upgraded lighter model, with no new public date.

The stock fell 25 percent in a single session, the steepest one-day drop since the company's founding in 1911, after CEO Arvind Krishna warned that quarterly earnings and revenue would miss expectations. The reason matters more than the number: clients shifted their spending away from IBM's software and services toward buying physical hardware, memory chips and servers, before an expected price hike, and Krishna specifically said Anthropic's newest model release had stalled several large deals while customers waited to see what it meant for them. It's a reminder that being in the AI conversation isn't the same as being the company customers pay first.

Xi Jinping formally stood up the World AI Cooperation Organization, a 29-country body headquartered in Shanghai that includes Brazil, Indonesia, Russia, and South Africa. Xi framed it as AI development shouldn't be "a solo performance by a single country," and pledged 5,000 training slots for developing nations. Read plainly: China is positioning itself as the AI partner for the parts of the world Washington isn't courting as hard.

Apple's complaint accuses OpenAI's Chief Hardware Officer, a former Apple VP, of asking job candidates to bring actual Apple parts to interviews, and alleges a former Apple employee who joined OpenAI stole a company laptop on the way out. The dispute centers on OpenAI's device ambitions with former Apple design chief Jony Ive. It's a sharp reversal for two companies that were partners on Siri just two years ago.

Dave Brown, who ran AWS's computing infrastructure for nearly 19 years, is joining Meta to help build out its data centers and a new effort called Meta Compute, widely read as Meta testing whether it wants to rent out AI infrastructure the way Amazon, Microsoft, and Google already do.

The latest update lets Copilot predict follow-up edits anywhere in your file instead of just near your cursor, added the ability to pull full pull request context straight into chat, and rolled out automated security reviews inside the Copilot app. The shift across nearly every coding tool this quarter is the same: less "autocomplete," more "check my work before it ships."

The AI coding tool now splits usage into two separate pools, one for its own Composer models and one for outside models like Claude or GPT, with a new higher-usage "Premium" seat alongside the standard one. Cursor says the changes should lower costs for 90 percent of teams. For any leader who's had a developer tool bill balloon out of nowhere, this is the industry finally building in guardrails.

The world's dominant chipmaker reported $40.2 billion in quarterly revenue, up 34 percent from a year ago, and net profit up 77 percent, the ninth straight quarter of double-digit profit growth. AI chips now make up two-thirds of its wafer business, and the company raised its full-year growth outlook and committed another $100 billion to its Arizona plants. None of that stopped the stock from falling the same week, because Kimi K3 raised the question of whether the industry needs quite as many of the most expensive chips as everyone assumed.

IonQ, Rigetti, and D-Wave are all sitting 60 to 76 percent below their 52-week highs, with another 17 to 20 percent shaved off in just the last week, even though the underlying science, better error correction, longer-lasting qubits, keeps genuinely improving. It's the clearest example in tech right now of a sector where the stock chart and the lab results are telling two completely different stories.

The industrial giant is scaling up production of cooling and power-management technology built for the data centers hyperscalers are racing to build, in a new strategic partnership with Microsoft. It's a good reminder that a lot of the money in this buildout is going to companies that don't say "AI" anywhere in their name.

Plug Power, which has spent years burning cash on hydrogen fuel technology, signed agreements to sell land and infrastructure assets to Stream US Data Centers for up to $132.5 million, part of a plan to free up more than $275 million in total liquidity. It's a sign of how much AI-driven land and power demand is worth right now, even a company outside the AI industry can cash out by selling into it.