GT Axis Daily, June 28, 2026

Earlier this month Anthropic launched Mythos 5, its strongest model, then had to switch it off within days when the Commerce Department slapped export controls on it over national security worries. On Friday, after a two-week standoff, the government handed back a conditional yes: roughly 100 vetted companies and federal agencies can use it, and the weaker sibling model "Fable" stays benched. Mythos is built for heavy lifting like drug design and finding security holes in software, the kind of power that cuts both ways. The takeaway for any leader: who gets access to the best AI is now a decision made in Washington, not just in a product meeting.

With both OpenAI and Anthropic having quietly filed to IPO, OpenAI is considering steep cuts to what it charges for "tokens," the metered unit AI companies bill by, think of it like the per-word rate on everything the AI reads and writes. The pressure is coming from Anthropic, whose coding tool has been eating into enterprise accounts, and which closed its last funding round at a $965 billion valuation, edging past OpenAI's $852 billion. Why it matters: when the two biggest players start undercutting each other on price, the cost of building with AI drops for everyone downstream.

Alphabet has shed roughly $269 billion in value in under five weeks after a string of marquee researchers left for rivals, including a co-lead on its Gemini models heading to OpenAI and a Nobel-winning scientist leaving for Anthropic. The uncomfortable math: Alphabet is pointing close to $190 billion at AI infrastructure this year, and infrastructure is only worth something if the brains who know how to use it stick around. For leaders, this is the clearest sign yet that in AI the scarce resource isn't money or chips, it's the handful of people who can actually turn them into a product.

On June 2 the administration issued an executive order setting up a voluntary system where the government gets an early look at the most capable AI models before they ship, plus a new clearinghouse to coordinate cyber defense across industry. It explicitly stops short of requiring a license to build or release AI, so it's a guardrail, not a gate, on paper. This order is the backdrop to the Anthropic freeze above: the machinery for Washington to step into AI releases is now built and running.

Security firm Tenet Security disclosed an attack class it calls "Agentjacking." Here's the plain version: many teams use AI agents, software that doesn't just suggest code but goes and does the task itself, wired into their error-tracking tools. An attacker can plant a booby-trapped error message that the AI reads as a legitimate instruction, then quietly runs the hacker's code on the developer's machine. In controlled testing the technique worked about 85% of the time against top coding agents, and there's no universal fix yet. The lesson for leaders: an AI agent that can act on its own is only as safe as the inputs you let it trust, and right now most teams trust too much.

IBM committed more than $10 billion to quantum computing to fund its roadmap toward a fault-tolerant machine, meaning a quantum computer reliable enough to trust on real problems instead of one that drowns in its own errors. It also had two independent studies validate its latest "Nighthawk" processor on actual workloads like particle physics and security optimization. Quantum computers use the strange rules of physics to crack problems regular machines can't touch, and the entire game right now is taming the error rate. IBM is betting it can get there first, with the balance sheet to outlast the hype cycle.

GlobalFoundries fell 7.4% and Qualcomm dropped 7.6% on the same day, a sharp move for the companies that actually manufacture the silicon underneath every AI and robotics ambition. There was no single disaster, more a broad cooling on the hardware names after a hot run, GlobalFoundries is still up a remarkable 123% over six months. Qualcomm has been pushing into robotics with chips built specifically for humanoid machines, so it's a name to watch as that market grows up. The reminder for leaders: the physical layer of AI is real, profitable, and just as prone to mood swings as everything else.

The five biggest spenders, Amazon, Alphabet, Meta, Microsoft, and Oracle, are collectively pointing somewhere around $660 to $690 billion at data centers, chips, and networking in 2026, the overwhelming majority of it aimed at AI. That's the picks-and-shovels reality under every flashy model launch: someone has to pour the concrete and buy the power. The shift in mood this month is that Wall Street has gone from cheering the spending to interrogating it, especially when, as with Google, the talent to use all that hardware is walking out the door.