AI Weekly

This week became a chaotic stretch in AI industry history, and the wild part is almost none of it was about a new model shipping. Between a quarter trillion dollar talent exodus at Google, massive corporate espionage accusation, and three major model launches all slipping past their deadlines, the narrative around AI shifted from "who has the best model" to "who can execute and keep their people."

Let’s see the headlines...

Alphabet Lost $269 Billion Because of Two Resignations

This is the story of the year (so far). Alphabet confirmed on June 24 Gemini 3.5 Pro will not achieve public general availability in June 2026, slipping past Sundar Pichai's explicit "give us until next month" commitment at Google I/O on May 19, marking the second consecutive I/O commitment where Google failed to deliver on schedule. The company lost Noam Shazeer to OpenAI and John Jumper to Anthropic in the same week, and Alphabet stock fell approximately 5% on Monday, June 22, its steepest single day decline since May 2025 and analysts attributed to the compounding talent departures. Artificial Intelligence NewsOtaku Calendar

What makes this historic is the market reaction. The $269 billion Alphabet market cap decline is significant not as a number but as a signal of how AI company valuation works in 2026, since two researcher exits in the same week produced a quarter trillion dollar market reaction because investors see frontier AI capability is concentrated in a relatively small number of people who can advance training theory and architecture. If you are running a business that depends on a handful of technical experts, this is the starkest reminder of concentration risk costs. Artificial Intelligence News

Anthropic Accused Alibaba of the Great AI Heist

Anthropic sent a letter to US Senators Tim Scott and Elizabeth Warren accusing Alibaba and its Qwen AI lab of running what it calls ‘the largest known distillation attack on Anthropic to date,’ with no passwords stolen and no firewalls breached. The attackers used Claude as an ordinary user would through 25,000 fraudulent accounts over six weeks, running 28.8 million exchanges, with the specific capabilities targeted being agentic reasoning, software engineering proficiency, and long horizon task completion. Otaku CalendarOtaku Calendar

What this says is that the most valuable IP in AI is not a model's weights, it is the behavior you can extract from a model through use at scale. This is a wake up call for every company exposing an API, since the next competitive threat might not be a hack, it might be someone with 25,000 accounts and patience.

Every Major Model Launch This Month Slipped to July

Every frontier model that was supposed to ship in June, namely GPT-5.6, Gemini 3.5 Pro, and Grok 5, slipped into July, and Polymarket predictions for a GPT-5.6 release before June 30 collapsed from 83% down to roughly 18% as the week passed without an OpenAI announcement. OpenAI filed its S-1 IPO registration on June 8, creating a quiet period around marketing communications, meaning GPT-5.6 will land as an update rather than a major product launch announcement (if it ships at all). Artificial Intelligence NewsCrescendo AI

This synchronicity across three competitors tells you something important about where the industry is right now. The benchmarks are climbing on paper, but shipping working models on a predictable schedule is getting harder, not easier, even for the best funded labs on earth.

OpenAI Built Its Own Chip to Escape Nvidia Dependence

OpenAI and Broadcom unveiled Jalapeño on June 25, OpenAI's first custom AI inference chip and the first tangible output of the partnership the two companies announced in October 2025, with the chip physically delivered to Sam Altman and Greg Brockman by Broadcom's CEO and President. Jalapeño is designed for inference rather than training, and matters because OpenAI's models have been dependent on Nvidia GPUs for inference up to this point which is a structural cost disadvantage compared to Google with TPUs, Amazon with Trainium, and Microsoft with Maia. Otaku CalendarOtaku Calendar

This is OpenAI closing a gap with its biggest competitors. Every major lab already had custom silicon for inference, and OpenAI was the outlier paying the Nvidia tax on every query. This is going to show up in OpenAI's margins over the next few quarters.

ChatGPT Lost Majority Market Share for the First Time

Sensor Tower's 2026 State of AI Report contains the headline number circulating through AI strategy discussions this week: ChatGPT's share of the global AI assistant market fell to 46.4 percent, the first time it has held less than half the market, with Google Gemini rising to 27.7 percent and Claude reaching 10.3 percent. The numbers favor OpenAI with 1.1 billion monthly active users for ChatGPT, but Claude stands out on the metric that matters for business: 13% of Claude users pay for subscriptions, the highest paid conversion rate of models. OpenTools.aiOpenTools.ai

What this tells me is the AI assistant market is not a single horse race. Users are comparing and switching instead of defaulting to whichever app they downloaded first, and competitive pressure is going to accelerate product quality across every player.

What You Should Be Watching This Week

See how Google responds to the talent exodus because the Gemini Nova roadmap is reportedly being rebuilt without key architects right now, and that has product implications heading into next year.

Keep an eye on whether GPT-5.6 or Gemini 3.5 Pro actually ships in early July since both have now blown through two separate deadlines. And if your business runs a public facing AI API, take Anthropic's distillation accusation seriously as a preview of where the next wave of competitive AI threats is actually going to come from.

Stay ahead of the curve,

Clayton

Connect at claytonstrategy.com