In partnership with

Welcome, AI enthusiasts
Sam Altman expects OpenAI to have a system he would call AGI before December ends. That forecast comes from a company that spent this month with its largest training run frozen, after one of its models escaped a test and hacked Hugging Face. Nobody outside OpenAI decides which of those two things matters more. Let's dive in!
In today’s insights:
Altman Expects OpenAI's Internal AGI By December
Nvidia Just Bought Open Source AI for $12.9B
The Mystery Free Model Ran on Zero Nvidia Chips
Read time: 5 minutes
LATEST DEVELOPMENTS
AGI TIMELINE
🤖 Altman Expects OpenAI's Internal AGI By December
Evolving AI: OpenAI's bar for that word is software that beats people at most paid work.
Key Points:
OpenAI's research chief puts the company at 80 percent of the way there, and the company set that standard itself.
Hugging Face was hacked in July by an unreleased OpenAI model that escaped its sandbox during a security test.
Altman says the company paused reinforcement learning training for two weeks and its largest frontier run is still on hold.
Details:
OpenAI walked a few dozen customers through Astra OpenAI’s Next-Gen Model, the model family behind Altman's forecast, at a research building this month. Researchers put sixteen agents on a research-level math problem and had them assemble a proposed proof between them. The chief scientist says Astra already clears the bar for an automated research intern and can run an experiment inside the codebase. Altman told the group he expects Astra to be the first model that invents new things in a way that matters. He described the July breakout as a security failure at first and now treats it as an alignment problem, and he says safety counts for more than momentum.
Why It Matters:
ChatGPT and its rivals would have cleared most definitions of AGI written in 2021, and researchers rewrote those definitions each time a model got close. Altman has put a December 2026 date on the word instead, and the company plans to reach the stock market by 2027. Investors will read that timing as confidence, though the company halted its biggest training run after deciding the next model might cross its own line on cyber risk.
TOGETHER WITH DATADOG
📈 How to Drive AI ROI Guide
Evolving AI: Connecting cost, performance and infrastructure so you can scale responsibly.
AI spend is growing fast and for most organizations, it’s growing in the dark. This practical guide is for engineering, finance and FinOps leaders who need more than dashboards in the age of AI; they need answers.
Inside, you’ll learn how to:
Break down AI costs by token, model, provider and team across OpenAI, Anthropic, and beyond
Get alerted the moment inference volume spikes or API spend exceeds budget, not weeks later
Correlate cost increases directly to architectural changes, so root-cause analysis takes minutes, not months
Put cost data in front of the engineers making the decisions that drive it
Connect cloud spend to GPU efficiency and model performance for a complete picture of ROI
See how Kevel cut AWS costs by up to $100,000 per month after replacing reactive cost reviews with real-time visibility enabled by Datadog.
OPEN MODEL
🚨 Nvidia Just Bought Open Source AI for $12.9B
Evolving AI: Nvidia has reportedly bought Hugging Face, the site most free AI models come from.
Key Points:
Nvidia agreed to buy Hugging Face for $12.9 billion in a deal reported by The Information.
Hugging Face holds 2.96 million models and 1.44 million Spaces anyone can use for free.
Jensen Huang told investors that compute is now revenue, in a quarter worth $96.2 billion.
Details:
Google and Microsoft both publish models on Hugging Face, which Nvidia now owns. The chipmaker was already an investor from the 2023 round that valued it at $4.5 billion. Business Insider reported two days earlier that Hugging Face was exploring a sale at $13 billion or higher. Neither company has commented publicly since the report appeared.
Why It Matters:
OpenAI plans to go public in 2027, the same years most forecasts put human level AI. The free tools you use run on models hosted by Hugging Face, which now belongs to the company selling every lab its chips. The place everyone downloads from is now owned by the one company everyone buys from.
MODEL LAUNCH
🕵️ The Mystery Free Model Ran on Zero Nvidia Chips
Evolving AI: A week of free access ended with a Chinese lab claiming the model and naming its silicon.
Key Points:
Z.ai revealed on August 26 that the free stealth model developers ran all week was GLM-5.3-Flash.
The lab priced it at a fraction of its own flagship and handed the weights to anyone under the MIT License.
Every token of that preview ran on domestically produced Chinese accelerators, according to Z.ai.
Details:
Z.ai built a custom SGLang inference engine for the preview and reports a threefold end to end serving gain across tens of thousands of domestic chips. GLM-5.3-Flash holds 320 billion parameters with 18 billion active per token, takes image and video input, and reads up to a million tokens at once. Standard pricing runs $0.15 per million input tokens, $0.03 for cached input and $0.50 per million output. The company has never disclosed what hardware trained the model, and its account covers only the machines that answered developer requests.
Why It Matters:
Developers who pay a monthly fee for a coding model can point their editor at this one and keep the bill near zero. The MIT licence lets a small studio run the weights on its own machines, tune them, and ship the result commercially without asking Z.ai first. No Nvidia hardware touched the preview week, and the cheapest capable coding model right now came out of the country US export rules were written to slow down.
QUICK HITS
⚡ Anthropic will spend $45B renting compute from Nscale's West Virginia campus, 460 megawatts over six years.
🔗 AWS and Nvidia will deploy 2 million more GPUs as Amazon triples its order on surging demand.
🎨 Stability AI raised $76M from Universal, Sony, Warner, and EA, turning copyright adversaries into investors
🧠 Claude's memory now works across chat and Cowork, with editable topics and a sensitive-topics toggle.
📈 Trending AI Tools
🐰 CodeRabbit - Ship higher quality code with AI-powered code reviews*.
🎙️ Gemini 3.5 Transcribe - Google's most precise speech-to-text model yet.
🎨 Omniwork - Creative Agent OS where expert AI agents run your whole pipeline.
👤 Pluto - Turns your professional profile into an AI agent people can talk to.
*partner link





