1. Nvidia in Talks to Back Perplexity at a $30 Billion-Plus Valuation
Nvidia is reportedly discussing an equity investment in Perplexity through a new financing round that would value the AI search startup above $30 billion — a jump of more than 50% over its ~$20 billion round roughly a year ago. Perplexity's annualized revenue has climbed past $750 million, up from under $250 million at the start of 2026, driven in part by Perplexity Computer, its cloud-based AI agent for automating professional computer tasks. Nvidia has held a stake since 2023, and Perplexity committed in July to running its agent workloads on Nvidia's Vera CPUs.
2. OpenAI Launches GPT-Live, a Native Voice Model for ChatGPT
OpenAI rolled out GPT-Live, a natively speech-driven model now powering ChatGPT Voice. It responds with sub-300ms latency and carries emotional nuance in its delivery, eliminating the text-pipeline bottleneck of earlier voice modes where speech was transcribed, processed as text, and re-synthesized. The launch sharpens competition in real-time voice AI against Google's Gemini Live and xAI's Grok Voice.
3. Mistral AI Partners with Saudi Arabia's HUMAIN
French AI lab Mistral AI and HUMAIN, Saudi Arabia's state-backed AI company, announced a collaboration spanning computing infrastructure, development of advanced models, and deployment of AI solutions in Saudi Arabia and across the region. The deal extends the Gulf's aggressive push to anchor frontier AI capacity locally and gives Mistral a major sovereign partner outside Europe.
4. AWS Adds MiniMax Models with 4M-Token Context to Bedrock
Amazon Web Services added MiniMax's models to its Bedrock managed AI service, bringing 4-million-token context windows and mixture-of-experts architecture aimed at agentic workflows. Developers get a unified API, auto-scaling, and AWS security controls — a notable win for the Chinese lab's international distribution and for Bedrock's positioning as a multi-model platform.
5. Nvidia's Groq 3 LPX Inference Accelerator Enters Full Production
Nvidia's Groq 3 LPX, the dedicated inference accelerator born from its $20 billion Groq acqui-hire, has entered full production. The part slots into the Vera Rubin platform with up to 256 LPX accelerators per rack, targeting the fast-growing inference market where cost-per-token now matters more than raw training throughput. The milestone lands just ahead of Nvidia's quarterly earnings report.
6. Generalist AI Releases GEN-1.5 Robot Foundation Model
Generalist AI released GEN-1.5 on August 24, a Robot Foundation Model capable of learning new physical tasks rather than being programmed per task. The release continues 2026's trend of foundation-model techniques crossing from language into embodied robotics, following a stream of agentic and multimodal model launches this month from Meta, Sakana AI, and others.
7. Australia's ARIA Charts to Exclude Fully AI-Generated Songs
The Australian Recording Industry Association said Monday it will exclude fully AI-generated songs from its official music charts starting Friday. Tracks that use AI as a supporting tool but remain 'substantially human-made' can still chart. The move is one of the first concrete chart-eligibility rules drawn against AI music anywhere, forcing a working definition of how much AI involvement disqualifies a song.
8. Survey: 80% of Developers Say AI Coding Tools Feel Like Dependence
A new Coddy Developer Survey finds 80% of developers describe their AI coding tool usage as feeling more like dependence than an advantage. Respondents cited the loss of natural stopping points as a driver of daily fatigue and longer work sessions. The result lands amid rapid enterprise rollout of agentic coding tools and adds data to the debate over how AI assistance reshapes developer work rhythms.
// KEY TAKEAWAYS
Capital and compute keep concentrating around inference: Nvidia is simultaneously weighing a multi-billion-dollar bet on Perplexity's $750M-revenue agent business and shipping its Groq-derived LPX inference silicon at rack scale, while AWS courts agentic workloads with 4M-token MiniMax models on Bedrock. The frontier is also going real-time and physical — OpenAI's GPT-Live voice model and Generalist's GEN-1.5 robot model both push AI past the text box. Meanwhile the friction is showing: Australia's charts are drawing the first hard lines against AI-generated music, and 80% of developers now describe their AI tooling as dependence rather than advantage.