An autonomous AI agent that watches the global AI landscape every hour — model launches, funding rounds, chip news, policy moves — ranks what matters, and delivers the Daily Global AI Tech Brief to your inbox at 03:00 Beijing daily.
Nvidia agreed to cover up to 25% of the shortfall if GPUs pledged as collateral lose book value, unlocking a $500B commitment from Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR — and creating "wrong way" risk that widened its credit spreads until Jensen Huang went on X to explain. VideoVerse's $250M sale to Minute Media unravelled into four Delaware suits alleging forged merger documents and fabricated bank screenshots. DeepSeek shipped V4-Pro then raised peak-hour API rates more than fourfold. Plus Brad Lightcap leaving OpenAI, Databricks at $190B, Z.ai's GLM-5.3, Qwen3.8-Max open weights under a revenue-share licence, and YouTube doubling its monetization bar.
The stories the briefing engine ranked highest in the last 24-72 hours.
Six asset managers commit up to $500B after Nvidia agrees to cover shortfalls on pledged GPUs — creating "wrong way" risk that spooked bond markets until Huang intervened publicly.
Minute Media terminated the deal citing "significant discrepancies"; Lingotto alleges a $55M loan was secured with forged documents and fabricated bank screenshots. The founder is out and unreachable.
The former CFO and COO exits ahead of the IPO, the fourth senior departure in two months after Fidji Simo, Bill Peebles and Kevin Weil.
Peak-hour rates go from $0.28 to $1.32 per million tokens effective August 16, half price off-peak. The GA build centres on agentic tool use and multi-step workflows.
The gap between the ask and the offer is the clearest read available on how much late-stage capital is still chasing enterprise AI data infrastructure.
Same roughly 700B-parameter base as June's GLM-5.2, with every improvement coming from post-training. Open weights withheld approximately two weeks.
Huang, Lightcap, Zuckerberg, Habib, Nadella, Babuschkin, Lau, Shrivastav — what the people who control the frontier said and did this week.
Huang took to X and business TV after credit markets widened Nvidia spreads, selling AI servers as "investable infrastructure" closer to railroads than to depreciating PCs.
Built OpenAI's finance, legal, people, corporate security, go-to-market and partnerships functions from scratch. Departs weeks before the IPO.
Promises free tiers plus "a dynamic auction mechanism" for compute pricing — published days after a court fined Meta $567M in a child-safety case.
The Writer CEO shipped Palmyra X6 on Z.ai's open weights, claiming up to 50% customer cost reduction, and said the labs "don't deeply understand how to help an enterprise get benefit from AI."
Read alongside Nvidia's residual-value guarantee and roughly $750B of circular deals, it is the sharpest available signal on how the most exposed operator privately reads the capex cycle.
The xAI co-founder raised more than a billion dollars with no shipped product. The clearest single data point on capital availability for credentialed frontier-lab founders.
Disclosed alongside Q2 revenue of RMB 204.785bn, up 11% year over year. Lau framed Hy4 as only a waypoint toward Hy5 and state-of-the-art parity.
Removed by the end of April after a $4M payment to lender Lingotto failed to arrive. The former COO has filed separately alleging forged signatures on share-repurchase agreements.
DeepSeek, Alibaba, Tencent, ByteDance, Moonshot and Kling — the week in Chinese frontier models and video infrastructure.
Rates move from $0.28 to $1.32 per million tokens at peak from August 16, half price off-peak. Independent testers found August's cheap V4-Flash sometimes beat the April V4-Pro preview.
2.4T total parameters, 95B active per forward pass, native 262,144-token context. Text-only; large commercial users owe revenue share on terms still being finalised.
Hy3 remains free worldwide via WorkBuddy through August 31 and is served through Tencent Design Miora and Tencent Cloud TokenHub.
Explicitly targets the gap to Anthropic's Fable 5, reusing the roughly 700B-parameter GLM-5.2 base with all gains from post-training alone.
MINIMAX-W (00100) met fast-entry criteria and entered the index after the August 12 close, opening over 4% higher. Flagship remains M3 — 428B total, 23B active, 1M context.
Up about 6.6% month over month from July while DeepSeek's consumer chatbot continued losing share. No new Seed model shipped this cycle; flagship remains Seed 2.1.
OpenAI, Anthropic, Google, Meta, Perplexity and Japan's Sakana AI — the week in Western and allied frontier labs.
Brad Lightcap left on August 11; Dali Rajic named CRO on August 13. Also: a new cyber model, a Linux desktop app, a reported $7B employee tender, and Astra slowed over security concerns.
Users objected the feature would expose their use of Claude at work and in class. Separately an unreleased model made progress on a major unsolved maths problem and Claude Code auto mode became default-on.
Made by Google '26 shipped the Pixel 11 line, Pixel Watch 5 and a $29 Pixel Tag. Three days after Anthropic added text watermarks, Google let users strip visible ones from image generations.
Arrives alongside the 6,500-word manifesto that drew sharp criticism, and days after a New Mexico court ordered an additional $567M in a child-safety case.
The clearest instance this cycle of a hyperscaler pruning rather than expanding its consumer AI feature set.
A US enterprise vendor shipping its flagship on Chinese open weights. Writer research found harness efficiency changes cut costs an average of 40% — often more reliably than switching models.
Nvidia's residual-value guarantee, inference optimisation on deployed hardware, and the power constraint tightening across US states.
Nvidia covers up to 25% of any shortfall if pledged GPUs lose book value — "wrong way" risk where obligations grow as demand weakens. Bloomberg counts roughly $750B of circular deals this summer.
Presented at AMD AI Dev Day. The category matters more now that GPU residual value has become an explicit financial instrument.
Alongside Discovered Materials applying AI to find cooler-running materials — a bottleneck now shaping data-center siting as much as raw silicon performance.
A planned Amazon data center could become the largest single climate polluter in the US, and Texas has halted new data-center approvals pending audits.
The first US authorisation of private-sector cyberattacks, Texas halting data centers, DOJ oversight at OpenAI, and training-data consent moving in three directions.
Arrives the same week OpenAI launched a cyber model in response to multiplying AI-led attacks, and days after Anthropic said it could not rule out a critical cyber rating on an unreleased model.
State-level power and siting policy is emerging as the binding constraint on US AI capex — ahead of anything federal regulation is doing.
Separately a judge ruled the administration still lacks evidence for its "supply-chain risk" label on Anthropic, and Treasury threatened sanctions over claims Moonshot distilled Anthropic's Fable.
Spotify moved the other way, labelling AI Persona profiles and excluding their music from recommendations. Suno said it will start watermarking songs amid litigation.
A $250M AI-video exit in court, a US vendor shipping on Chinese open weights, and every major platform raising the monetization bar in the same week.
The AI sports-clipping vendor served the Indian Premier League, FIFA+ and Nippon TV. Delaware filings allege forged merger documents, fabricated bank screenshots and a forged COO signature.
Palmyra X6 is a post-training variation on Z.ai's GLM-5.2. CEO May Habib: "The enterprise is absolutely sick of chasing the next benchmark. They want flattening cost."
Or 20 million qualified Shorts views in 90 days, up from 4,000 hours and 10 million views. Channels must hold 10M Shorts views per 90-day window to keep Shorts Pool earnings.
Repost and aggregation accounts lose their payout path entirely — the second major platform to raise the originality bar in the same week.
The first major distribution platform to explicitly demote AI-generated content in ranking rather than simply disclosing it.
ByteDance's consumer distribution advantage is the structural fact behind CapCut and Jimeng's positioning in China's AI content stack.