PLUS: NVIDIA's massive $500B memory deal, Cloudflare's new AI bot controls, and AI for bovine banking
Happy reading
A widely-used AI training technique is now at the center of a major controversy. The debate over "distillation" has been ignited by the rapid rise of Chinese models like Kimi K3, which are quickly catching up to U.S. competitors.
The clash pits national security concerns against the principles of open innovation. This raises a critical question: should this powerful technique be regulated to protect intellectual property, or will that stifle the very competition it helps create?
In today's Next in AI:
China's Kimi K3 sparks distillation debate
NVIDIA's massive $500B memory deal
Cloudflare's new AI crawler controls
AI helps farmers tokenize cows for loans
The Distillation Debate
Next in AI: A firestorm is brewing over "distillation," a controversial AI training method now at the center of a clash between national security concerns and open innovation. The debate intensified after the release of Chinese models like Kimi K3, which are rapidly challenging U.S. leaders in performance.
Explained:
Distillation is a widely used technique where a smaller "student" model learns from the outputs of a larger "teacher" model, helping create more efficient and accessible AI.
The current conflict ignited after Anthropic accused Moonshot AI of copying its Fable model, describing the effort as "industrial scale" distillation.
In response, a coalition of tech giants including Nvidia, Microsoft, and Meta is urging policymakers to avoid "premature restrictions" on open-weight models, arguing it would stifle competition and innovation.
Why It Matters: This debate forces a critical choice between protecting intellectual property through regulation and fostering global innovation through open access. The outcome will set a major precedent for the future of AI development worldwide.
NVIDIA's $500B Power Play
Next in AI: NVIDIA is securing its AI hardware dominance with a massive $500 billion deal with SK Hynix. The partnership locks down the supply of essential high-bandwidth memory (HBM) for its next-generation AI systems.
Explained:
This is more than a simple supply order; SK Hynix's affiliate SK Telecom will build a 2-gigawatt cloud business using NVIDIA’s future Vera Rubin systems, with the first data centers coming online in 2027.
High-bandwidth memory has been a critical bottleneck for AI hardware, and this deal ensures NVIDIA has a stable supply while co-developing the next generation of HBM.
The entire semiconductor industry is making similar moves, highlighted by a recent partnership between Samsung and Broadcom to support next-gen AI infrastructure.
Why It Matters: This strategic move shows that securing the AI supply chain is now as important as designing the chips themselves. By locking in its memory pipeline, NVIDIA is ensuring a more predictable and powerful roadmap for the AI systems that companies will rely on for the next decade.
Cloudflare's AI Gatekeeper
Next in AI: Cloudflare is giving every website owner new granular controls to manage AI traffic, moving beyond a simple “block all” approach to a more nuanced system.
Explained:
Website owners can now manage AI bots based on three distinct categories: Search, Agent, and Training. This allows for fine-tuned permissions, such as allowing bots that help with discoverability while blocking those that scrape content for model training.
Starting September 15, 2026, Cloudflare will block Training and Agent bots by default on any pages that display ads for new domains. This new setting aims to protect monetized content while still allowing search indexing.
For Enterprise customers, Cloudflare launched BotBase, a comprehensive directory for tracking known bots, and is testing new content-use signals in robots.txt to let owners specify how their content can be stored or reused.
Why It Matters: This shift gives content creators more direct control over how their intellectual property is used in the age of AI. It also pressures the broader industry to adopt more transparent practices, forcing bot operators to be clear about their intent when accessing web content.
AI for Bovine Banking
Next in AI: Farmers in Brazil are turning dairy cows into digital assets to secure loans. This novel approach uses AI-powered collars to tokenize their dairy cows on the blockchain, unlocking new lines of credit previously out of reach.
Explained:
Agtech firm Cowmed equips each cow with an AI-powered smart collar that creates an encrypted digital identity by continuously monitoring its health, behavior, and location.
This real-time data allows the livestock to serve as movable collateral, preventing fraud and enabling farmers to secure a rural credit note on Brazil's B3 stock exchange.
With 100,000 cows already in its network, Cowmed projects this model could unlock over $77 million in new agricultural financing.
Why It Matters: This is a powerful demonstration of how AI can verify and digitize real-world assets, making them bankable. It shows a future where industries previously limited by physical constraints can access global capital markets through technology.
AI Pulse
Realbotix deployed a humanoid robot named Sally in a New York high school, sparking controversy among teachers and residents due to the company's previous acquisition of a realistic sex doll manufacturer.
Granola sparked a debate on privacy as a growing number of professionals use AI transcription services to record all conversations, from work meetings to first dates, often without disclosure.
AI cracked a decades-old math problem with a simple prompt, triggering what some mathematicians are calling a "profound spiritual crisis" about the future of human discovery in the field.
Anthropic revealed its new best practices for context engineering, advising developers to use less rigid rules and instead design better interfaces to let advanced models like those in the Claude 5 generation use their own judgment.