GT Axis Daily, June 30, 2026

It previewed GPT-5.6 in three flavors: a heavyweight for hard problems, a balanced everyday version, and a cheap fast one for high volume. The catch is that neither the public nor most companies can use it freely, because it now has to clear Washington's new review process for frontier AI, meaning the most powerful models get a government check before wide release. Translation: the cutting edge of AI is starting to come with a permission slip.

For two years the message was "use as much AI as you can." Now OpenAI and Anthropic are selling the opposite: get more done with fewer "tokens," the tiny billable units that measure how much AI you actually consume. The reason is simple. Business buyers stopped writing blank checks and started asking what they get back. The era of spending for the sake of spending is ending, and that is healthy.

Meta tried to rent more of Google's Gemini AI capacity and Google said no, because it doesn't have enough computing power to go around. Several other Google customers got squeezed too, but Meta felt it most because its appetite is the biggest. The reaction inside Meta says everything: staff were told to use AI more sparingly to stretch what they've got. When the second-richest tech company gets rationed, the shortage is real.

Two more senior researchers tied to Google's Gemini model are leaving for Anthropic, on top of recent exits including a Nobel laureate. The pull is money: Anthropic and OpenAI are both racing toward stock-market debuts, and joining before the IPO is a rare shot at a life-changing payday. For Google, the worry isn't one departure, it's the pattern.

With AI power scarce and budgets tightening, the smart teams are squeezing more output from every token, the billable unit of AI usage. Even Meta is coaching staff to be frugal with it. For any leader running a business, the lesson is the same one that always wins: the edge goes to whoever gets more done per dollar, not whoever spends the most.

The President signed two orders aimed at building the first quantum computer powerful enough to do real science, and at protecting America's data before quantum machines can crack today's encryption. Quantum computers use the strange rules of physics to solve problems normal computers can't touch. The second order sets hard deadlines for federal agencies to switch to "post-quantum" security, meaning encryption built to survive a quantum attack. Washington is treating this like a national race, and that pulls private money in behind it.

The government, with Samsung and SK Hynix, committed to four new chip factories and a wave of memory-chip and data-center capacity, built at home. Memory chips are the short-term, high-speed storage that AI systems lean on constantly, and they're in dangerously short supply right now. The plan is a national bid to control the part of the AI supply chain everyone else depends on.

Anthropic, the maker of Claude, confidentially filed to go public after a funding round valued it at $965 billion, topping OpenAI for the first time. Confidential filing just means the early paperwork is in with regulators before any public stock sale. Both companies are eyeing debuts as soon as this fall. The race is why Google keeps losing researchers, and it's a signal that the AI money is moving from private hands toward the public markets.