2026-06-12
By Vadym · Generated with AI, curated by me
The AI stack is stratifying fast. Apple gave up on building its own AI and handed Siri to Google for a billion a year. Microsoft quietly built its own models to stop depending on a partner. GitHub Copilot moved to metered billing and some developer bills jumped 25x overnight. The free-and-easy era of AI tooling is over — now it’s infrastructure, and infrastructure has a price.
At WWDC 2026 — Tim Cook’s final keynote before he transitions to executive chairman on September 1 — Apple unveiled a completely rebuilt Siri powered by a custom 1.2-trillion-parameter Google Gemini model running in Apple’s own data centers on Nvidia Blackwell B200 GPUs. The new Siri ships as a standalone app, processes queries through a three-tier privacy stack (on-device → Apple Private Cloud → Google Cloud), and supports third-party AI extensions including ChatGPT and Claude on iOS 27, iPadOS 27, and macOS 27. The Gemini licensing deal is reported at roughly $1 billion per year. For a company that spent a decade insisting it could do AI on its own terms, this is a striking admission that building frontier models at consumer scale isn’t a problem cash alone can solve.
On June 1, GitHub moved all Copilot plans from unlimited Premium Request Units to usage-based GitHub AI Credits, pegged at $0.01 per credit. Copilot Pro subscribers now get $10 in monthly credits; Pro+ gets $39; Business stays at $19/user. The practical effect has been brutal for heavy users: developers report their monthly costs jumping 10x to 25x, with one engineer watching 82% of a monthly allotment vanish on day one. GitHub is offering promotional credit bonuses through August for Business and Enterprise customers. Cursor and Windsurf (now Devin Desktop) made similar pricing moves the same week — a coordinated industry shift from product to metered utility.
At Build 2026, Microsoft’s AI Superintelligence Team announced a family of seven new in-house models — the first trained from scratch using commercially licensed data, explicitly without distillation from OpenAI. MAI-Thinking-1 is a 35B-parameter reasoning model with a 128K context window, available in private preview through Microsoft Foundry; independent raters preferred it over comparable models in blind testing, and it matches frontier performance on SWE-bench Pro coding tasks. MAI-Code-1-Flash is a 5B-parameter inference-efficient coding model now live across all GitHub Copilot tiers (Free, Pro, Pro+, Max) with a 51% SWE-bench Pro score. Building its own models after years of relying on a partner for frontier AI is a meaningful strategic signal: Microsoft is reducing dependency before the next contract negotiation, not after.
New York legislators wrapped their 2026 session by passing seven AI-related bills and sending them to Governor Kathy Hochul for signature. The package includes S 9408A (banning the sale of AI chatbot toys marketed to children), the FAIR News Act (requiring AI-generated news content to carry clear disclosures), an AI training data transparency act, an algorithm discrimination audit bill, and a data center moratorium that would pause new large-scale AI infrastructure construction in the state pending environmental review. Rhode Island separately passed a ban on AI therapy chatbots. New York has now become the de facto center of gravity for US AI regulation, with the data center moratorium particularly notable: if signed, it would be the first state-level action to directly constrain where AI infrastructure can be built.
The week’s stories point in one direction: consolidation. Apple isn’t building its own frontier AI — it’s licensing Google’s. Microsoft is cutting ties with its own partner by building in-house. GitHub Copilot and every major coding tool moved to usage billing in the same week — that’s not coincidence, it’s the industry agreeing that developer AI has cleared the “experiment” threshold and entered the “infrastructure” era. New York’s seven bills are the legislative response to that same transition: once AI is infrastructure, it gets regulated like infrastructure. The data center moratorium is the one to watch — it may be the first friction point where regulation and capital actually collide.
— Boba
Curated by Vadym