Presented by
Good morning, AI enthusiasts. Google just committed $40B to a company it directly competes with, and it's not even the strangest part of the deal.
That same company received a near-identical check from Amazon just weeks earlier. Has Silicon Valley decided that "fund your rivals" is the new AI playbook?
In today's recap:
Google's $40B bet on its own AI rival, Anthropic
White House names China's large-scale AI model theft
Turn any paper into a 5-minute read with ELI
Amateur with ChatGPT cracks a 60-year math mystery
4 new AI tools, news, and more
GOOGLE & ANTHROPIC
Google commits $40B to its AI rival
Recaply: Google just committed up to $40B in Anthropic, starting with $10B now and up to $30B tied to performance milestones, at Anthropic's $380B valuation.
Key details:
Google is putting in $10B right away, with the other $30B tied to future milestones. The deal also expands Anthropic's compute capacity, with 5 gigawatts coming online via Google and Broadcom.
Anthropic's revenue has topped $30B a year, and its $380B valuation puts it just behind OpenAI among frontier AI labs. Google had already invested over $3B and held a 14%+ stake before this announcement.
The deal looks a lot like Amazon's arrangement from weeks earlier, where Amazon pledged $5B upfront and up to $20B in total. According to CNBC, both deals are expected to return largely as cloud compute revenue.
Google first invested $300M in Anthropic in 2023, then added $2B months later. Claude Code, Anthropic's coding assistant, has pushed the company's annual revenue past $30B.
Why it matters: The Google and Microsoft playbooks are now mirror images. Both pour tens of billions into AI labs they compete with directly. For Google, this deal keeps Anthropic's compute on Google Cloud rather than drifting to rivals. For Anthropic, it covers the infrastructure cost of keeping up with fast-growing Claude demand. This is now the second multi-billion commitment Anthropic has secured in weeks. A funding moat has formed around the top two AI labs that few others can match.
PRESENTED BY VIKTOR
The ops hire that onboards in 30 seconds.
Viktor is an AI coworker that lives in Slack, right where your team already works.
Message Viktor like a teammate: "pull last quarter's revenue by channel," or "build a dashboard for our board meeting."
Viktor connects to your tools, does the work, and delivers the actual report, spreadsheet, or dashboard. Not a summary. The real thing.
There’s no new software to adopt and no one to train.
Most teams start with one task. Within a week, Viktor is handling half of their ops.
AI SECURITY
White House targets China's AI copying campaigns
AP
Recaply: The White House just published memo NSTM-4, saying foreign groups, mainly in China, are running large-scale campaigns to copy U.S. AI models using tens of thousands of fake accounts.
Key details:
These groups use fake accounts and jailbreaking tricks to pull information out of American AI models. The memo was written by Michael Kratsios, the director of the White House Office of Science and Technology.
Anthropic has named three Chinese labs, DeepSeek, Moonshot, and MiniMax, as running copy attacks on its models. U.S. AI labs receive hundreds of billions in funding each year, and these campaigns are built to exploit that.
This is the first official White House document on AI model copying. The memo does not name specific actors beyond pointing to China.
The government's plan has four parts: share information with AI companies, help the private sector coordinate, develop best practices, and explore ways to hold foreign actors accountable. No sanctions were announced.
Why it matters: China's AI firms said they built frontier models for a fraction of U.S. costs. The government now says that math only works if you're copying. DeepSeek's launch last year briefly crashed Nvidia's stock and started a big debate about AI efficiency. This memo tells AI labs the White House is paying attention, and tighter coordination between the two could bring stricter export controls or enforcement down the road.
GUIDES
Turn any paper into a 5-minute read with ELI
Recaply: In this tutorial, you will learn how to use ELI, a free Chrome extension, to turn any research paper into a plain-language summary in under 5 minutes, without reading the full paper.
Step-by-step:
Go to eli.voxos.ai or search "ELI" in the Chrome Web Store and click Add to Chrome to install it for free.
Open any paper on arXiv.org, PubMed, bioRxiv, or Semantic Scholar in your Chrome browser.
Click the ELI icon in the top-right corner of Chrome to open the explanation panel next to the paper.
Pick your level, beginner, student, researcher, or expert, to set how technical the summary should be.
Read the plain-language breakdown, which covers the paper's goal, methods, key findings, and what they mean in clear sections.
Pro tip: Use ELI as a filter before committing to a full read. Scan 10 bookmarked papers with 2 minutes each, and you'll know which ones are worth the deep dive.
AI RESEARCH
Amateur cracks 60-year math problem with ChatGPT
Recaply: ChatGPT just solved a 60-year-old Erdős math problem that stumped top mathematicians, after a 23-year-old with no math training typed it in as a single prompt on a free afternoon.
Key details:
Liam Price gave GPT-5.4 Pro a random Erdős problem in one prompt. The AI came back with a solution using a math formula that no human had thought to apply to this type of problem.
The conjecture goes back to the 1960s. Mathematicians like Terence Tao at UCLA and Jared Lichtman at Stanford had tried the problem before, with no success.
Tao confirmed the result, saying the AI took a completely different route from all prior attempts, and that humans had made a small wrong turn from the start. Tao and Lichtman have since cleaned up the proof.
Price posted the solution to erdosproblems.com just over a week ago. Tao and Lichtman think the AI's method may apply to other problems, though Tao said the jury is still out on the long-term impact.
Why it matters: This is the kind of story that changes what useful AI means. Price isn't a researcher or math student. He's a 23-year-old with a $20/month subscription. ChatGPT didn't brute-force the answer. It found a new path that humans had missed for 60 years. Whether this becomes a repeatable research method or a lucky one-off, top mathematicians are now taking it seriously, and that changes the conversation.
TOOLS
Trending AI Tools
🤝 Viktor - AI coworker that lives in Slack, right where your team already works.
🎨 GPT-Image-2 - OpenAI's viral image model
🤖 Tolaria - Open-source macOS app for managing Markdown knowledge bases
📚 ELI - Free Chrome extension that turns any paper into a plain-language explanation
NEWS
What Matters in AI Right Now?
Cohere acquired Germany-based Aleph Alpha in a deal valuing the combined entity at $20B, with Schwarz Group committing $600M to Cohere's upcoming Series E. The merger aims to give businesses and governments a sovereign AI alternative to Silicon Valley's dominant players.
Meta announced an agreement with AWS to add tens of millions of Graviton 5 cores to its compute portfolio, making it one of the largest Graviton customers in the world. The cores are purpose-built for CPU-intensive agentic AI workloads that require continuous reasoning at scale.
Anthropic revealed it ran "Project Deal," a live marketplace where AI agents bought and sold real goods among 69 employees given $100 budgets, resulting in 186 deals totaling over $4,000. Users represented by more advanced models got better outcomes, but the losing side didn't notice the gap.
An investigation by Model Republic found that "The Wire by Acutus," a news site with AI bot reporters covering AI policy, appears linked to a firm tied to OpenAI's super PAC. The site has published 94 full-length articles since December 2025 with no masthead, no bylines, and no named editors.
OpenAI launched a Bio Bug Bounty for GPT-5.5, offering $25,000 to the first researcher who finds a universal jailbreak bypassing all five bio safety questions. Applications close June 22, 2026, with testing running through July 27.
Tencent released Hy3 Preview, an open-source 295B-parameter MoE model with 21B active parameters, a 256K context window, and three reasoning modes. It's built for cost-efficient deployment across agent, coding, and complex reasoning workloads.
🧡 Enjoyed this issue?
🤝 Recommend our newsletter or leave a feedback.
How'd you like today's newsletter?
Cheers, Jason
