AI Arms Race Volume: 1
Author: Brod Justice, Ryan McClure
Published Date: July 23, 2025
Weekly AI Signals — Curated by Builders, for Builders
Every week, we round up the most relevant developments in AI, tech, and the business strategies shaping them. Born from internal conversations among senior engineers, product leads, and AI researchers, this digest simplifies the noise into clear signals. Authored by Brod Justice and Ryan McClure, these summaries are built for real practitioners—people shipping code, scaling platforms, and thinking past the hype.
Volume 1
1. Google Hooks Windsurf Stars to Dodge Anti‑Trust Waves.
Google poaching the top Windsurf talent after the OpenAI multi-billion bid failed suggests that Google is nervous about anti-competition and monopoly regulation. See also point 3.
2. AWS Feels the Heat as AI Firms Build Their Own Racks.
Amazon cloud servers are suffering as AI companies build out their own hardware (rumours suggest that this is because AWS is far too expensive). What does that mean for Amazon that is often rumoured to be loss making without AWS?
3. DeepMind‑Powered Google Surges Past OpenAI—Bad News for NVIDIA?
Google was written-off last year, but is making a massive come-back, it may have overtaken OpenAI (see chart). Mistral (Europe’s hope) is losing ground 🙁 If Google is doing so well, and they are doing all of this without NVIDIA, what does that mean for NVIDIA? Also, Google’s efforts are driven by DeepMind, its UK-ish AI division. Hope for Europe?
4. Search Wars Escalate: Perplexity Accuses Google of ‘Mafia Tactics
Perplexity CEO implicitly saying Google is like a Mafia organization is mostly true, but utterly hypocritical coming from a company whose business model is essentially stolen content re-sold as their own.
5. EU Funding Crawls While UK AI Group Ghosts Start‑ups
UK AI group still not responding (been 6 months now). EU/European government bodies that are supposed to support AI start-ups are hopelessly slow, and in the case of some (eg, “AI Munich”) totally unfit for purpose.
6. Kimi K2 Lands: 1 T Tokens, 2 M Context, Open Source
The new Chinese Kimi K2 AI model is pretty good. Not quite another “deepseek moment” but still got to be worrying for all those billions invested in OpenAI and others. It’s a kind of reasoning model and open source. A massive 1 trillion tokens, and a 2 million token content window. Try it here: https://www.kimi.com
7. Clients Flock to Gemini & Mistral; Claude Priced Out
Our customer favorite LLMs are Gemini-2.0 and Mistral Small and Medium. Claude for some, but it’s 30x more expensive and often down. Mistral also having server problems is pushing more clients to Google (which is very fast and cheap)
8. OpenAI Quietly Delays AGI—Microsoft Left Hanging
Looks like OpenAI is pushing back the date for “AGI” (whatever that might mean). This idea of “AI better than humans” seems to be dying like it should do. What does that mean for Microsoft’s agreement with OpenAI though? That’s got to be painful for OpenAI. Altmas claims e.g. “No AI scaling law”, look weaker all the time. Meanwhile ChapGPT 4.5 is effectively dead and Altman seems to want to classify AGI as something you can “feel”. Pity no OpenAI LLM can create new memories, like a 1 week old baby can.
9. Cloudflare Slams Gate on AI Crawlers, Hits OpenAI & Co.
Cloudflare’s announcement of intention to block AI crawlers has got much less publicity than it deserves – this could really damage OpenAI, Perplexity, etc. Fit’s very well with our VSP protocol!
10. Study Finds AI Pair‑Programming Actually Slows Devs Down
A new paper suggests that AI makes coders & developers slower, and they don’t even realise it. This fits with some of our own experience.
11. Musk’s Grok ‘Waifu’ Hints at Avatar‑Filled Future
Musk and his Grok “waifu” is slightly disturbing, e.g. does she really undress? But avatars are inevitable and maybe even ubiquitous by this time next year at the latest.
12. Thinking Machines Bags $2 B—Albania Bets Big on AI
Thinking Machines Lab seems to have finalized raising $2 billion ($12 billion valuation). But what exactly they will do with their AI research isn’t quite clear. They did get $10 million directly from the Albanian government – is Albanian being bolder with AI than the EU?
13. Leaked Memo: OpenAI Aims to Replace the Web
Leaked (and alleged, not confirmed or denied, from June 2025) OpenAI Strategy document is further evidence of what we suspected (and aligns with Perplexity’s goal). As they say “ChatGPT … will be the way we interact with everything”, “it will replace more and more things, search engines, browsers, you name it”, “priority is growth not monetization”. See more here.
14. HeyGen’s 7× Growth, But Ireland’s Anam Steals the Show
HeyGen is doing really well with AI Avatars (x7 revenue increase), but our money is on a little known start-up from Ireland called Anam. We have an interactive AI Avatar demo working already. Perhaps a world first?
15. Call It “Artificial Usefulness,” Not Artificial Intelligence
We should rename “Artificial Intelligence” to “Artificial Usefulness” (hat-tip to Alberto Romero, his Stubstack is great). It’s not truly intelligent, and calling it such seems to lead businesses to dismissing it. AI is more like a really, really good useful tool, like a screwdriver. Everybody needs to know what you can do with a screwdriver, and likewise everybody needs to know what they can, cannot, and should not do with “Artificial Usefulness”. (We are only half-joking here)
16. ChatGPT Agent: A Paywall Wrapped Around Your Content
OpenAIs new “ChatGPT Agent” looks very much like #13 is true. It steals content and charges you for it. Read more.
17. VS Protocol Plugs the API‑Docs Gap in Vibe‑Coding
Documented limits of so-called “vibe-coding” is the inability of AI (like Claude Code) to get the very latest documentation on e.g. API calls. The VS Protocol fixes that AND optionally makes the AI vendors pay. Is the VS Protocol the vital missing part of vibe-coding?
18. Brussels calls Big Tech’s bluff.
The EU released its final General‑Purpose AI Code of Practice on 10 July and reiterated that the AI Act’s first hard deadline (2 Aug 2025 for new GPAI models) will not be delayed. The code obliges model providers to log training data, honour “do‑not‑scrape” signals, disclose energy use and run continuous risk audits. That hands publishers fresh leverage over LLM builders and undermines the “train‑now‑pay‑never” culture. A very tidy narrative fit for our VSP idea of making AI vendors pay for content.
19. Netflix just moved the goal‑posts for VFX.
In the upcoming series El Eternauta, Netflix’s Eyeline Studios generated a key building‑collapse scene with generative‑AI in “a tenth of the time and cost” of traditional pipelines—the first time the streamer has admitted using AI in a flagship drama. If one marquee shot can be done this fast, mid‑tier post houses (and the Hollywood unions that struck over AI in 2023) have a whole new negotiating landscape.
Meta began to officially filter synthetic content. On 15 July Facebook began demonetising pages that repeatedly repost AI‑generated or lifted material—after already purging 500 k spam accounts—while, in the same product cycle, pitching advertisers shiny new AI‑video formats for Reels. When the platform both supplies the brush and bans the paint, creators and brands will start looking for neutral distribution layers… exactly the gap a VSP‑style, publisher‑controlled search layer can fill.
ChatBar AI Website Assistant
New to AI? Take Our Free Course.
ChatBar AI Learning is the fast, practical way for business owners and teams to get up to speed on real-world AI.
Want More Like This? Just Ask.
Tap below and ask ChatBar to recommend more high-signal AI content and use cases—no fluff, no noise.
Ask ChatBar to show similar content
Ready to Try ChatBar AI on Your Site?
Apply now to join our early access program and see how ChatBar can supercharge your site’s content, discovery, and engagement.