GT Axis Daily, July 23, 2026

Gemini 3.6 Flash and 3.5 Flash-Lite are Google's "Flash" tier, meaning the fast, cheap, everyday-use models rather than its top-of-the-line thinker. Both are faster and use fewer tokens per task (tokens are the chunks of text AI processes, so fewer tokens means cheaper answers) than their predecessors. The story underneath: Gemini 3.5 Pro, the flagship model Google was counting on to lead in coding and complex reasoning, remains delayed after internal testing showed it falling short of expectations.

The accusation: Moonshot AI ran a hidden system to copy capabilities out of Anthropic's model (called "distillation," essentially training a smaller AI to mimic a bigger one's answers) and separately got access to Nvidia's most advanced chips through a workaround in Thailand, even though those chips are barred from sale to China. Treasury is now raising the possibility of sanctions. Why it matters: this is the clearest sign yet that AI competition is turning into an actual trade and national security fight, not just a product race.

Presence combines OpenAI's models with a company's own rules and data so an AI agent (software that doesn't just answer questions, it actually resolves the issue end to end) can handle customer calls, chats, and claims, then hand off to a human only when it's stuck. OpenAI is already using it for its own English-language phone support, where it now resolves about 75% of calls with no person involved. For now it's invite-only for select enterprise customers, not something any business can sign up for yet.

Apple alleges OpenAI's hardware chief, a former Apple product design VP, and other ex-Apple employees fed OpenAI confidential supplier and design information as OpenAI built its own consumer hardware push. The complaint claims over 400 former Apple staff now work at OpenAI. Why it matters for any leader: this is the visible edge of a much bigger talent war, and the terms of who owns what an employee learned on the job are about to get tested in open court.

Its "Managed Agents" system lets a company plug Claude into its own tools and let it work independently on multi-step jobs, with built-in memory (the AI actually remembers what it did last time) rather than starting fresh every conversation. The latest update tightens how that memory gets searched and organized at scale. This is the unglamorous plumbing work that decides whether a company can trust AI to run unattended for hours, not days of hype.

These "always-on subagents" can hand off pieces of a coding job to specialized helpers and keep context the whole way through, cutting down the copy-paste babysitting developers usually do with AI tools. Small update, but it's part of the broader shift toward AI that manages a whole project rather than answering one question at a time.

Hitachi, Intel, and Japan's national research institute AIST are building a government-backed project to take "silicon spin-qubit" quantum computers (a quantum chip design built on the same silicon manufacturing Intel already knows how to do at scale) from lab experiments to a factory-buildable product. The targets: a 100-qubit (qubit is quantum computing's basic unit of information) prototype by 2028 and a 1,000-qubit machine by 2030. Translation: instead of hand-building one quantum computer at a time, this is the plan to build them like normal chips.

Infleqtion's "neutral-atom" approach (using individual atoms trapped by lasers as the computer's building blocks, one of several competing quantum designs) will be applied to problems grid operators already struggle with today, like deciding which power plants to run and preventing cascading outages. The company also won three federal energy research contracts this week. It's one of the more grounded quantum stories out there, an actual customer problem instead of a future promise.

BlackRock's infrastructure arm and Abu Dhabi's MGX fund completed their purchase of Aligned Data Centers, which owns and is building more than 6 gigawatts of data center capacity (a gigawatt is roughly enough electricity to power 700,000 homes, so this is serious industrial scale) across the US and Latin America. Days after closing, the buyers committed another $5 billion in growth capital. The point for any leader watching the AI boom from the outside: the real money isn't chasing the flashiest app, it's locking down the power and buildings everything else needs to run.