GT Axis Daily: July 17, 2026

Gemini 3.5 Pro is Google's rebuilt reasoning model, meaning it works through problems in steps instead of guessing the first answer, and it's rolling out with a massive 2 million token context window, essentially how much text or data it can hold in mind at once. The launch is six weeks later than promised, which tells you how hard it is to stay ahead right now.

Moonshot AI's Kimi K3 is "open weight," meaning any company can download it and run it themselves instead of paying rent to a big lab. It's the largest model of its kind ever released and Moonshot claims it holds its own against Anthropic and OpenAI's top models. Translation: the gap between US and Chinese AI labs just got a lot smaller.

Thinking Machines Lab's Inkling can understand text, images, and audio together, and lets users dial its "thinking effort" up or down depending on whether they want speed or accuracy. It's open weight too, following the same playbook as Moonshot: give away the model, make money on the tools built around it.

Xi Jinping delivered the keynote at Shanghai's World AI Conference, calling for AI to be "a symphony of international cooperation" rather than one country's show, a not-so-subtle jab at US export controls. Every prior year, a deputy handled the opening. Xi showing up himself is Beijing saying it now sees AI leadership as a national priority worth his personal time.

Internal plans show the chip, code-named Iris, moving into production in September, part of an effort to double Meta's computing capacity by next year. Think of it as Meta trying to grow its own food instead of buying all of it at the grocery store, cheaper over time, but only after a lot of upfront work.

Claude Sonnet 5 is built to plan out multi-step tasks and actually use tools like a web browser or a coding terminal on its own, not just answer questions. It's priced to be accessible, roughly a third of what the top-tier model costs, and it's now what Free and Pro users get by default. The bet: make the capable version the standard one, not the upsell.

The update lets businesses connect custom AI models, including Claude, directly inside their coding tools instead of being locked into one vendor's assistant. For a non-technical leader, the takeaway is simple: less vendor lock-in, more competition among AI providers to win developer trust.

Agility Robotics, maker of the Digit warehouse robot already working shifts for companies like Toyota and Mercado Libre, is merging with a shell company to list on public markets in a deal valuing it near $2.5 billion. That's a faster, cheaper path to Wall Street than a traditional IPO, and it puts real money behind the bet that warehouse robots are past the science-project stage.

IonQ, D-Wave, and Rigetti are speculative, pre-profit companies that trade heavily on sentiment. When bigger, safer tech names get sold off over spending worries, the smaller, riskier names in the same neighborhood tend to fall even harder, exactly what's playing out now.

TSMC's revenue jumped 36% from a year ago, profit jumped 77%, and it raised its full year sales outlook. Normally that's a stock-popping report. Instead shares fell more than 3% because the company also raised how much it plans to spend building new factories, and Wall Street is nervous that even TSMC's customers can't keep writing checks forever.

NVIDIA, AMD, Intel, and infrastructure names all fell alongside it, part of a broader reset that has erased roughly $3.3 trillion in chip sector value since late June. The one name bucking the trend, Micron, rose slightly, a reminder investors aren't abandoning AI, they're getting choosier about who's actually converting the spending boom into profit right now.