- Cua ships a Metal shim that accelerates LLMs inside macOS VMs
The process-scoped layer unlocked llama.cpp GPU paths and delivered up to 16.36x faster generation in Cua's M1 Ultra tests.
- Open-weight models surge past closed rivals in Vercel token traffic
DeepSeek V4 Flash and Step 3.7 Flash lead the shift, though Vercel's data covers its own gateway rather than the entire AI market.
- Silicon Data raises $30.5 million as CME readies GPU futures
Carmen Li's benchmark startup will expand the pricing and performance data behind CME Group's planned cash-settled contracts.
- Former Qwen lead Junyang Lin launches Pragmatik Labs for digital and physical agents
HSG and Gaorong co-led the Shanghai lab's financing, with Tencent joining a founder-driven bet reportedly valued at $2 billion.
- Spotify launches Xirp to orchestrate coding agents without model lock-in
The beta preserves context across Claude, Gemini and Codex sessions, then feeds the work into Spotify's Portal developer platform.
- Mistral launches compute coalition to steer Europe's next AI data centers
Arthur Mensch's Mistral will issue European Compute Units against multi-year commitments, with pricing and contracted capacity still undisclosed.
- ngrok publishes an interactive guide linking LLM prediction to compression
Annie Sexton's interactive guide explains how token probabilities become compression costs; ngrok separately operates an AI Gateway for routing model traffic.
- Scoop: ChampionHQ discloses $4.8M sold toward a planned $6M offering
ChampionHQ filed the Form D on August 11th, 2026, nearly two years after founders Jeff Reekers, Gianna Scorsone and Courtney Crispin announced a $3.3 million seed. The filing does not identify the new offering's investors, lead investor, valuation or closing date.
- SpaceX frames Mars and AI as its next challenge in employee update
Elon Musk's address ties Starship, Starlink and xAI into the broader corporate mission SpaceX presented to public investors.
- Quantinuum reports $8 million Q2 revenue, partners with Oracle on Helios cloud service
CEO Rajeeb Hazra says Quantinuum's $2.1 billion in cash, cash equivalents and short-term investments gives it room to accelerate its business plans while maintaining disciplined capital allocation.
- CoreWeave reports $2.58 billion Q2 revenue with $35 billion in debt
Michael Intrator's AI cloud reported $2.58 billion in quarterly revenue and a $626 million loss while its backlog reached $104 billion.
- Kikitora discloses $19.7M round for offline AI motion capture
Founder Astrid Wilde is betting ordinary video can replace specialized capture rigs, with local processing as Kikitora's privacy edge.
- Robot.com unveils legged delivery robot to push beyond sidewalk limits
Felipe Chavez's R-dog pairs FieldAI autonomy with delivery and mobile ads, with commercial deployments planned for 2027.
- Firefly Aerospace hits record $117.7M Q2 revenue, beating estimates by 34%, NASA awards mount
Revenue rose 45.5% sequentially as NASA, defense and lunar awards expanded Firefly's backlog; the company maintained 2026 guidance of $420 million to $450 million.
- OpenAI launches ChatGPT for Linux, bringing Codex to major desktop distributions
The preview supports Ubuntu, Debian and Fedora, closing a platform gap after Anthropic shipped Claude for Linux in July.
- Bloomberg says Phia founders pushed affiliate features that could claim unearned sales credit
Bloomberg says Phia founders Phoebe Gates and Sophia Kianni pushed affiliate-attribution features that could claim credit for sales Phia did not drive, extending July reporting on the software.
- Neros raises $250 million to expand autonomous drones and interceptor production
Drone-racing founders Soren Monroe-Anderson and Olaf Hichwa plan to use the Bloomberg-reported round to expand autonomous and interceptor programs and manufacturing capacity.
- Microsoft upgrades its Copilot coding model 10 weeks after launch
The 138 billion-parameter mixture-of-experts model adds image input and arrives 10 weeks after Microsoft's first in-house Copilot coding model.
- Four Grok 4.5 checkpoints appear in Arena near xAI's 4.6 window
The roster entries point to multimodal and web-enabled variants, but their labels do not establish that Grok 4.6 is behind them.
- RuntimeWire launches Investigations for original startup and AI reporting
The new section gives document-driven scoops, reverse engineering and accountability reporting a home outside The Wire's daily news stream.
- OpenAI, Anthropic and Google blocked a cross-model reasoning attack
The attack used weaker sibling models as decoders and exposed credentials in public agent logs before providers blocked it.
- Unsloth Desktop connects locally run models to Claude Code and Codex
The open-source Mac, Windows and Linux app combines local inference, training and connections for Claude Code and Codex.
- Nvidia ships Nemotron 3.5 Lightning for single-GPU AI agents
The 30-billion-parameter open-weight model activates 3 billion parameters and permits commercial use under Nvidia's OpenMDW license.
- Cursor and SpaceXAI launch Grok Bot for work beyond coding
The beta gives paid users cloud agents that sign into websites, learn routines and keep running after a laptop closes.