1. Nvidia in Talks to Back Perplexity at a $30B+ Valuation
The Information reported on August 23 that Nvidia is in discussions to join an equity round valuing AI search startup Perplexity at more than $30 billion — over 50% above its roughly $20 billion mark from about a year ago. Perplexity's annualized revenue has reportedly climbed past $750 million, up from under $250 million at the start of 2026. Nvidia has held a stake since 2023; neither the round nor the valuation is confirmed as closed.
2. Nvidia Q2 Earnings Land Today — The AI Trade's Stress Test
Nvidia reports Q2 FY27 results after the close on August 26. Consensus expects roughly $92 billion in revenue (up ~96% year over year) and adjusted EPS near $2.09, with Data Center alone projected above $85 billion. Nvidia now splits Data Center reporting between hyperscalers/AI clouds and its ACIE segment (industrial and enterprise), so the mix will be read as a signal on how broadly AI spending has spread beyond the big clouds.
3. OpenAI's Custom Inference Chip Beats Nvidia on Perf-per-Watt
SemiAnalysis published a teardown of OpenAI's in-house inference accelerator, codenamed Jalapeño, built with Broadcom on TSMC's N3P process in about 16 months. The part delivers 13.4 PFLOPs at 700W against Nvidia Rubin's 900–1,150W draw, which the analysis puts at 1.5–1.9x more work per watt across several model benchmarks. It is the clearest sign yet that frontier labs can build competitive silicon for their own serving workloads.
4. Skild AI's S1 Learns 10-Minute Robot Tasks From a Single Video
Skild AI released S1, a robotics foundation model that executes long-horizon tasks after watching one human video demonstration, with no fine-tuning. The company reports 66% success on unseen tasks versus 9% for comparable language-prompted systems trained at the same scale. Generalist AI shipped GEN-1.5 on August 24 in the same vein — robot foundation models learning from demonstrations are having a moment.
5. Apple Ships First 2nm Chip and a Quad-Die M5 Ultra
Apple unveiled the M6, its first 2-nanometer processor, alongside the M5 Ultra — a quad-die design offering up to 4.5x the AI compute of the M3 Ultra. The M5 Ultra tops out at 512GB of unified memory at 1.2TB/s of bandwidth, enough to hold very large models entirely in memory on a desktop. That capacity puts serious local inference within reach of individual developers rather than data centers.
6. $8B Order for 9,000 Vera Rubin Systems From India's Greenko
Greenko Group's AI arm, AM Intelligence, placed binding orders for roughly 9,000 Nvidia Vera Rubin NVL72 systems for Q1 2027 delivery, committing about $8 billion toward 200MW of compute and a stated path to 1GW across multiple countries. Deals of this size from non-hyperscaler buyers are the newest feature of this build-out cycle — sovereign and energy-linked players are now ordering at cloud scale.
7. EU AI Act Transparency Rules Are Now Being Enforced
As of August 2, the European Commission's AI Office and national authorities began enforcing the AI Act, and new transparency obligations took effect. Chatbots must disclose that users are talking to a machine, and synthetic or altered content — including deepfakes — must be labelled, with penalties reaching €15 million or 3% of worldwide turnover. The heavier high-risk obligations covering biometrics, employment, education and migration were pushed back to December 2, 2027.
8. Cognition in Funding Talks at a $40B Valuation
Bloomberg reported that coding-agent startup Cognition is in early talks with investors on a round that would lift its valuation more than 50% to at least $40 billion. Coding remains the single clearest commercial application of agentic AI, and capital is concentrating there accordingly — Meta's Muse Code assistant and Alibaba's Qwen3.8-Max are both pitched at the same workload.
9. Open-Weight Releases Keep Landing: DeepSeek, Zhipu, Qwen
The August release tracker shows DeepSeek shipping V4-Flash-Vision-Exp on August 21 and V4-Pro-0813 the week prior, Zhipu's GLM-5.3 and Alibaba's open-source Qwen3.8-27B on August 14, and Google's Gemini 3.7 Flash on August 13. The cadence — several credible releases a week, most of them open-weight and Chinese-origin — is now the baseline rather than the exception.
// KEY TAKEAWAYS
Money and silicon dominated this week: Nvidia is simultaneously the buyer (Perplexity), the seller ($8B in Vera Rubin orders), and the benchmark everyone is measured against — while OpenAI's own chip beating it on perf-per-watt shows the largest customers are quietly becoming competitors. Agentic coding is where private capital is concentrating, with Cognition alone in talks at $40 billion. On the regulatory side, the EU has moved from writing rules to enforcing them, though only the transparency layer — the hard high-risk obligations slipped to late 2027.