On June 18, Noam Shazeer, a Google DeepMind vice president of engineering and co-lead of the Gemini models, said he is leaving Google to join OpenAI as Lead for Architecture Research, a role focused on the basic design of future models. Shazeer co-wrote the 2017 paper 'Attention Is All You Need,' which introduced the transformer, the design behind essentially every major large language model (the kind of AI behind chatbots) today. Google paid roughly $2.7 billion in 2024 to bring him back from his startup, Character.AI.
Why it matters
It shows how fiercely the top labs compete for a tiny group of researchers who shaped the field, and it hints that OpenAI is investing in model designs that could go beyond today's transformer.
John Jumper, who shared the 2024 Nobel Prize in Chemistry for AlphaFold, said he is leaving Google DeepMind after nearly nine years to join the AI startup Anthropic. AlphaFold predicts the three-dimensional shape of a protein from its building blocks, and more than two million scientists in 190 countries have used it. Neither company has said what Jumper will work on at Anthropic.
Why it matters
It signals Anthropic pushing harder into AI for science and biology. Landing the same week as Shazeer's exit, it also points to a wider talent drain from Google's research lab.
Days after its record-setting Nasdaq debut, SpaceX said on June 16 it will buy Anysphere, the startup behind the AI coding tool Cursor, in an all-stock deal worth $60 billion. Cursor had reached roughly $4 billion in annualized revenue (its current yearly pace), about $2.6 billion of it from business customers, and the deal gives Elon Musk's xAI its first major foothold in AI tools for developers. Widely described as the largest acquisition of a venture-backed startup ever, it is expected to close in the third quarter of 2026 if regulators approve.
Why it matters
It is a textbook case of a newly public company using its high-flying stock as currency to buy capability instead of building it. It also brings a coding tool that many students and developers already use under SpaceX's control.
On June 18, the Federal Energy Regulatory Commission (FERC) sent tailored 'show-cause' orders, which make a party justify its rules or change them, to the six regional grid operators it oversees: PJM, MISO, Southwest Power Pool, CAISO, ISO New England and NYISO. Texas is not included. The operators have 60 days to defend or rewrite their rules for connecting data centers and other large power users, and 30 days to report how they will ensure enough electricity is generated. FERC says the goal is faster access to power without ordinary customers absorbing the cost.
Why it matters
AI's growth is increasingly limited by electricity, not just chips. These rules will influence where data centers get built and whether their costs land on ordinary people's power bills.
Odyssey, a Palo Alto AI lab founded by self-driving veterans, announced a $310 million Series B (its second major funding round) on June 17 at a $1.45 billion valuation, led by Natural Capital with Amazon, AMD Ventures, GV, EQT and the CIA-linked fund In-Q-Tel joining. It builds 'world models,' AI that learns physics and cause and effect to simulate the physical world rather than only generate text. As part of the deal, Amazon Web Services becomes its preferred cloud provider and will supply its Trainium chips.
Why it matters
Venture money is moving beyond chatbots toward AI that simulates the real world, a field with uses in robotics, gaming and science that members may build on in the years ahead.
The Reuters Institute's Digital News Report 2026, published June 16 and based on nearly 100,000 interviews across 48 markets, found weekly use of AI chatbots for news rose from 7 percent to 10 percent worldwide in a year, and reached 17 percent among 18- to 24-year-olds. Only about 4 percent of users say they always or often click through to the original source. Trust in news hit a record low of 37 percent, and social media and video networks overtook TV and news websites as the top news source worldwide for the first time.
Why it matters
News discovery is shifting to AI tools that summarize answers without sending readers to the source. That raises real questions about accuracy and accountability, and about whether the publishers who report the facts can survive.
This week the AI contest moved off the model leaderboard and onto three harder battlegrounds: people, power and the physical world. Two of the best-known minds behind Google's research left within days of each other, a reminder that as model designs mature, scarce human expertise becomes the real prize, and that the labs writing the biggest checks do not always keep their stars. A freshly public SpaceX showed another route, turning its soaring stock into currency to buy capability outright, while new funding for world models signals that investors see AI that understands reality, not just language, as the next frontier.
The stories also cut against each other in useful ways. The boom driving talent wars and mega-deals is colliding with stubborn physical limits, which is why a regulator, not a lab, may quietly set AI's pace by deciding how fast data centers get electricity. And the news-habits data is a sober counterweight: AI is already rewiring how ordinary people learn about the world, faster than trust or accuracy can keep up.
What to watch
Watch FERC's clock: the six grid operators have 30 days from June 18 to report on power supply and 60 days to justify or rewrite their rules for connecting AI data centers.
Test yourself
A few quick questions on this week. Just for you, nothing is recorded.
1. Why does Noam Shazeer's move to OpenAI matter beyond a single hire?
2. How is SpaceX paying for Cursor maker Anysphere, and what made that possible?
3. According to the Reuters Institute, how many chatbot news users say they always or often click through to the original source?
4. What makes a 'world model,' like the ones Odyssey builds, different from a chatbot?
