The iPhone maker sues its biggest AI partner for trade secret theft, while the frontier model race gets a massive pricing shakeup.

Last Monday's edition told you to watch the EU AI Act's August 2 deadline and the infrastructure arms race building behind it. Both moved again this week — and so did the competitive alliances holding the whole thing together. Apple sued its own AI partner. OpenAI shipped the model family it built partly on hardware talent Apple says it stole. New York became the first state to tell the industry it cannot keep building without asking permission. And a $19 billion data center lease in Kentucky reminded everyone that the physical layer is where the real bets are being placed.

Three of those stories made every front page. The fourth — a new benchmark measuring which retailers are actually ready for AI shopping agents — is the one that will quietly reshape how products get discovered and sold. The lesson of the week is the same one it keeps being: the headline is the drama, but the constraint is the infrastructure, the regulation, and the readiness score your competitors are already tracking.

In a stunning reversal of their high-profile partnership, Apple has sued OpenAI in federal court, alleging a coordinated campaign to steal hardware trade secrets. The complaint accuses OpenAI, its Chief Hardware Officer Tang Yew Tan (a former Apple VP), and another former employee of systematically acquiring confidential information about unreleased Apple technologies, manufacturing processes, and supplier relationships to jumpstart OpenAI's consumer hardware ambitions.

The lawsuit highlights the growing tension as OpenAI moves from a software partner to a potential hardware rival. Apple claims that OpenAI directed job candidates still at Apple to bring "actual parts" to interviews for "show and tell" sessions, and that more than 400 former Apple employees now work at the AI lab. OpenAI, which acquired Jony Ive's hardware startup IO Products for $6.4 billion last year, denied the allegations, stating it has "no interest in other companies' trade secrets."

For the broader industry, this legal battle sets up a fight over who will control the future of AI devices that may bypass traditional operating systems. While the companies partnered just last year to bring ChatGPT to the iPhone, Apple has notably shifted its upcoming Siri overhaul to rely on Google's Gemini models instead. The outcome could delay OpenAI's hardware timeline and reshape the competitive landscape for consumer AI.

Read the full story →

The number of former Apple employees now working at OpenAI, according to Apple's own lawsuit filing. That is not a talent pipeline — it is a transfer of institutional knowledge at a scale that Apple's legal team argues could not have happened without a coordinated effort to extract hardware trade secrets. The complaint names two individuals by title, but the number is the story: when a company loses four hundred engineers to a single rival, the IP walks out with them whether or not anyone hands over a file.

Source: Apple Inc. v. OpenAI Inc., U.S. District Court, Northern District of California (July 2026)

OpenAI has launched the GPT-5.6 family into general availability, introducing three tiers: Sol (flagship), Terra (mid-tier), and Luna (budget). All three feature a massive 1.05M-token context window, making codebase analysis and long-document processing highly accessible. The pricing structure introduces a new dynamic for developers, explicitly charging for cache writes at 1.25x the input rate, meaning heavy caching workloads will pay more to fill the cache before realizing savings on reads.

In a massive bet on future compute needs, TeraWulf signed a 20-year lease with Anthropic for a 401-megawatt data center campus in Kentucky. The deal is expected to generate roughly $19 billion in contracted revenue over its lifespan, with service starting in late 2027. It is one of the largest AI infrastructure commitments to date, underscoring that physical capacity and power remain the ultimate bottlenecks in the frontier model race.

New York Governor Kathy Hochul signed an executive order creating a one-year moratorium on new "hyperscale" data centers that consume 50 megawatts or more. Driven by concerns over surging residential electricity bills and grid capacity, the ban makes New York the first state to halt AI infrastructure buildouts, setting a precedent that other states may follow as the tension between AI energy demands and local utility costs intensifies.

Meta is reportedly building an internal cloud business dubbed "Meta Compute" to rent out its excess AI capacity by the hour. CEO Mark Zuckerberg confirmed that selling compute "makes sense" as a way to monetize the company's $100 billion-plus infrastructure investments. The move would pit Meta against AWS and Azure, while directly threatening the business models of neocloud providers like CoreWeave that currently rent GPUs to the market.

Digital Commerce 360 and ReFiBuy have co-developed the AI Commerce Rankings, a first-of-its-kind benchmark measuring how prepared the Top 1000 retailers are for Agentic Commerce. The index evaluates bot-friendliness, AI source traffic, diversity of sources, and 90-day momentum to determine catalog readiness for AI shopping agents. Early leaders like Online Labels and Fashionphile prove that in the agentic era, scale alone does not determine product discovery success.

ChatGPT Work — OpenAI's new "super app" agent designed for white-collar workers. It combines ChatGPT with Codex capabilities to let non-coders create documents, presentations, and websites autonomously, positioning it as a direct rival to Anthropic's Claude Cowork.

Figma Make with GPT-5.6 — Figma has integrated OpenAI's new GPT-5.6 into its Make tool, allowing designers to generate complex UI layouts, responsive components, and interactive prototypes from text prompts with significantly higher fidelity and speed.

Grok 4.5 — xAI's newest frontier model, positioned as an "Opus-class" competitor. It features a 500K context window and is priced at $2 input / $6 output per million tokens, making it xAI's most expensive model to date, targeted heavily at developer pipelines.

Anthropic's CEO sat down for a wide-ranging conversation about what happens when you put "a country of geniuses in a data center." The discussion covers the scaling hypothesis, how AI will diffuse through the economy, and US-China competition — essential context for understanding why $19B infrastructure leases and hardware trade secret lawsuits are becoming the new normal.

If you only do one thing this week: audit your prompt caching strategy. With GPT-5.6 introducing explicit cache-write charges, your agentic workflows might be about to get significantly more expensive if they constantly rewrite system prompts.