In partnership with
In this edition
Happy Friday, fam 👋👋
Anthropic spent the week raiding Google's trophy case — it hired a Nobel laureate, poached two Gemini leads, and shipped a new product — while the rest of the industry threw billions at chips and compute. Blink, and Google got smaller.
🧠 Anthropic poaches a Nobel laureate (and Google's stock felt it).🧠 Anthropic poaches a Nobel laureate (and Google's stock felt it).
🏗️ The compute land-grab goes parabolic: SpaceX, Qualcomm, Broadcom.
🛠️ Five high-impact tools you've (probably) never heard of.
🪦 Fresh graveyard arrivals + an AI cold-war ethics read.🪦 Fresh graveyard arrivals + an AI cold-war ethics read.
🧠 Anthropic poached a Nobel laureate
John Jumper — who shared the 2024 Nobel Prize in Chemistry for AlphaFold — left Google DeepMind for Anthropic after nearly nine years, and Alphabet's stock fell about 6% on the news. Days later, two of Google's Gemini leads, Jonas Adler and Alexander Pritzel, followed him to Anthropic — on top of Noam Shazeer's defection to OpenAI (the researcher Google paid ~$2.7B to bring back just last year). Google is hemorrhaging the exact people who built its lead, and Anthropic is catching most of them.
🏗️ The compute land-grab goes parabolic
SpaceX leased $6.3B of compute to Reflection AI ($150M/month through 2029) — the fourth big outside tenant of its Colossus supercluster in Memphis, alongside Anthropic (~$45B) and Google (~$30B); committed outside revenue there now tops $80B. And Qualcomm opened talks to buy Jim Keller's Tenstorrent for $8–10B, a direct shot at Nvidia. The model race is now just as much a real-estate-and-silicon race.
🚀 Anthropic's week kept going
Beyond the talent haul, Anthropic shipped Claude Tag for Slack — tag @Claude in a channel and it works like an always-on teammate (it reportedly already writes ~65% of Anthropic's own product code).
OpenAI countered with GPT-5.5-Cyber, a record 85.6% on CyberGym, gated to vetted security teams. The two-house consolidation only gets more lopsided.
Our take: this is what “the moat is people” looks like in real time — and right now, Anthropic is the one doing the hiring.
🌮 Fail of the week
Snap's $2,195 face computer. Snap debuted Specs, its long-promised AR glasses, at $2,195 — Evan Spiegel's bet that you'll strap a computer to your face in a “post-smartphone” world. They've got OpenAI and Gemini built in. They also cost more than a laptop and ship “this fall.” Bold. Brave. Beleaguered.
@cnbc Snap is launching Specs, the company’s first AR glasses geared toward the broader public instead of developers. "They're both highly weara... See more
Honorable mention: Google paying ~$2.7B to re-hire a researcher who left 22 months later (see up top). The most expensive faceplant of the week — and it wasn't even close.
⚡ Rapid fire
OpenAI + Broadcom unveil “Jalapeño” — the company's first custom inference chip, a bid to lean less on Nvidia.
Google pushes Gemini 3.5 Pro to July — its flagship slips out of June as the talent keeps walking.
Anthropic's revenue reportedly passed OpenAI's (~$30B vs ~$25B run-rate) — with a fraction of the users.
A satellite ran Gemma 3 in orbit — Loft Orbital's YAM-9, the first vision-language model in space.
SpaceX's stock had its worst day since its IPO (down ~16%) — the same week it booked a $6.3B compute deal. Winning is weird.
/
You Already Have a Take on What AI Does Next
OpenAI or Anthropic? Which model leads the next benchmark? Which company ships the next major breakthrough?
If you follow AI closely, you already have opinions on where the industry is headed. Kalshi lets you trade on real-world AI and technology events, with markets that move as models launch, benchmarks drop, and announcements happen.
The people who follow this space most closely often see the story developing before everyone else. Put that knowledge to work and trade what you think happens next.
Bonus credit varies from $15 to $500. Terms apply.
You know ChatGPT and Claude. Here are five high-impact tools flying under most people's radar:
Glean — enterprise “Work AI” that searches across all your company's apps and answers like a coworker who's read everything.
Harvey — the legal-AI platform quietly running inside top law firms: drafting, review, and research at firm scale.
Cartesia — real-time, ultra-low-latency voice AI (Sonic); the engine behind a lot of the voice agents you've talked to.
Decagon — AI customer-support agents that actually resolve tickets end to end.
Granola — the AI notepad power users swear by; turns back-to-back meetings into clean, searchable notes.
This week's biggest mover deserves the spotlight: Noam Shazeer. Beyond co-authoring “Attention Is All You Need,” he's behind a startling share of the plumbing modern AI runs on — Mixture-of-Experts, multi-query attention, and early large-scale language modeling at Google. He left to build Character.AI, got bought back by Google for ~$2.7B in 2024, and just walked to OpenAI to lead “Architecture Research” — a mandate to find whatever comes after the Transformer he helped invent. If anyone's qualified to make his own paper obsolete, it's him.
Four more names checked into the AI Graveyard recently:
Phind — the AI answer engine devs loved, shut down.
Langfuse — the LLM-observability favorite, acquired by ClickHouse.
PlayHT — the AI voice generator, domain gone dark.
DreamStudio — Stability's consumer image studio, sunset.
Want the autopsy? We dug into what actually kills AI tools — funding, acquisition, or just a dead domain — in our data report.
When the only moat is people
This week made the quiet part loud: in frontier AI, the durable advantage isn't the model — models get matched in months — it's the few hundred people who can build the next one. Google learned that the hard way, paying billions to acquire talent only to watch it walk across the street.
That raises uncomfortable questions. If a handful of researchers can swing the balance of power between trillion-dollar companies, how much leverage should any one of them have? And what happens to everyone downstream — the startups, the open-source projects, the smaller labs — when the talent and the capital keep pooling into two or three houses?
There's a public-interest angle, too. The same week the talent consolidated, 42 states subpoenaed the market leader and Washington reached for export controls. Concentrated capability invites concentrated scrutiny — and concentrated risk.
None of this is an argument against ambition. It's a reminder that “the moat is people” cuts both ways: a flex when you're hiring, a liability the day they leave.
Prompt of the week: the key-person risk audit
If this week proved anything, it's that the people and vendors behind your AI stack can move fast. Here's a five-minute exercise — paste it into your model of choice and run it on your own setup:
Here is my current AI tool + vendor stack: [paste your tools]. For each one: (1) how exposed am I if the company gets acquired, pivots, or shuts down in the next 12 months, and why; (2) what would switching cost me in time and money; (3) the single cheapest hedge I can put in place now. Rank the list from highest to lowest risk.
On the calendar:
AI Engineer World's Fair — June 29–July 2, San Francisco (Moscone). 6,000+ builders; Coding Agents headlines Day 1.
RAISE Summit — July 8–9, Paris.
ICML 2026 — July 6–12, Seoul.
Ai4 2026 — August 4–6, Las Vegas.
Until next time 👋
Building something interesting, or know a tool that belongs in the directory (or the Graveyard)? Hit reply, or email [email protected] — we read everything.
— The ToolDirectory.AI team
