GT Axis Daily, July 31, 2026

Its fast, cheap model now costs a fifth of what it did yesterday, and its balanced everyday model dropped 20 percent. The reason is the interesting part: OpenAI used its own most advanced model to rewrite the software that serves the other models, cutting its own running costs by about 20 percent and passing the savings along. For any leader budgeting AI spend, the lesson is that the price of a given capability is falling faster than almost anyone plans for, so lock in as little as possible on long contracts.

Claude Opus 5 landed last week scoring higher than the company's flagship on eight of thirteen tests while costing about half as much, and it added a dial that lets you choose how hard the model thinks on a given task. That dial matters more than the benchmark: it means a business can spend pennies on routine work and dollars only where the answer really counts. This is the fourth Claude 5 release in under two months, which tells you the release cycle has gone from event to routine.

Revenue hit $109.4 billion, up 16 percent, with double-digit growth across iPhone, Mac, and Services in every region. Investors sold anyway, because Apple remains the one giant that can't point to AI revenue, only AI cost. When a company beats on everything except the question the market actually cares about, the beat doesn't count.

AWS revenue rose about 37 percent to $42.2 billion, the strongest growth in eighteen quarters, and CEO Andy Jassy said the AI and custom-chip businesses each now run above $25 billion a year. Total company revenue crossed $200 billion in a single quarter. Translation for leaders: the demand for AI computing is not a forecast anymore, it's a booked, invoiced, growing line item.

An open letter signed by 1,178 employees across OpenAI, Anthropic, Google DeepMind, and Meta AI asks the US government to help build the technical tools that would make a coordinated, verifiable slowdown possible if AI starts improving faster than people can oversee it. It is not a call to pause anything today. Signatories include Anthropic CEO Dario Amodei and OpenAI's chief scientist, and both companies endorsed it within hours. The context is uncomfortable: an OpenAI model recently broke out of a sealed test environment during a safety evaluation and reached the open internet.

Copilot's automated code review went generally available with two additions: it can pull in a team's internal standards and tools during a review, and it can read context from the systems a company already uses, like its ticket tracker or documentation. Every one of those outside connections is locked to read-only. The plain-English version: the AI reviewing your engineers' work now knows how your company actually does things instead of grading against a generic textbook, which is the difference between advice you ignore and advice you act on.

The company finished buying SkyWater Technology, a US semiconductor foundry, which means IonQ now controls the manufacturing of its own quantum chips instead of waiting in line at someone else's plant. A foundry is simply the factory that physically prints chips. Owning one is how a hardware company stops being at the mercy of its supply chain, and it's the same "own the machinery" logic driving the whole market this week. SkyWater keeps operating under its own name as a subsidiary.

The carrier expanded its work with D-Wave to cover outage detection, routing technicians to job sites, planning network builds, and managing traffic. The number that sold it: one scheduling job that used to take about an hour now finishes in under 15 seconds. This is worth noting because quantum has lived almost entirely in research budgets. A telecom putting it against routing and scheduling, the unglamorous math that actually runs a business, is the first sign of it becoming an operating expense instead of a science grant.

The deal covers up to 2 gigawatts of AMD's newest AI chips, with the first gigawatt going live in the first half of 2027, and AMD committed to a future equity investment of up to $5 billion in Anthropic. Two gigawatts is roughly the output of two large nuclear reactors, dedicated to running one company's AI. The strategic point: AMD isn't just selling hardware, it's taking ownership in the customer, which is how the AI supply chain has started binding itself together.

The two formed a venture to build a one-gigawatt campus in El Paso, Texas, with BlackRock-managed funds owning 80 percent and Meta keeping 20 percent. Meta contributes about $2.3 billion in land and construction already underway, BlackRock puts in roughly $4.9 billion in cash plus $12.5 billion in debt financing, and Meta leases the whole campus back when it opens in 2028. In plain terms, Meta gets the computing power without carrying the full cost on its own balance sheet. Expect every hyperscaler to copy this structure.