Pulse
What's shipping in AI right now — curated, summarised, updated through the day.
- Socket·INFRA·3w ago
Socket raises $60M at $1B for AI-era software supply chain security
Socket closed a $60M Series C at a $1B valuation on 21 May 2026, led by Thrive Capital with Andreessen Horowitz participating. The product detects malicious open-source packages before they reach production; the company claims it caught the compromised Axios package six minutes after publication. The customer list is the AI dev stack itself: Anthropic, xAI, Replit, Cursor, Figma, Vercel. As agents pull more dependencies into more codebases, automated supply-chain checks stop being defensive hygiene and start being a default-install line in any agent harness.
- TechCrunch·INFRA·0mo ago
Osaurus launches Mac-only LLM server bridging local and cloud models
Osaurus, an open-source Apple-only LLM server from ex-Tesla/Netflix engineer Terence Pae and Sam Yoo, hit TechCrunch on 15 May 2026 after a year of public building and over 112,000 downloads. The pitch is a single Mac-native interface that can route between local models (DeepSeek v4 and similar, requires ~128GB RAM for the big ones, 64GB for smaller) and cloud providers like OpenAI and Anthropic, while keeping memory, files, and tools on your own hardware in a hardware-isolated sandbox. Differentiated from Ollama and LM Studio by going for a consumer-friendly UI rather than a developer terminal.
- Webmatrices·INFRA·1mo ago·+1 source
Roblox cheat and AI tool caused Vercel platform outage
Vercel's hosting platform suffered a major outage triggered by an unusual combination: a Roblox cheat tool and an AI application. Details of the incident remain sparse, but the pairing suggests an exploit or resource exhaustion scenario where automated AI requests interacted with game modification software in a way that overwhelmed Vercel's infrastructure. The incident highlights the fragility of shared hosting platforms when faced with unexpected load patterns from AI tooling. Hacker News readers noted the increasingly common pattern of AI-driven traffic causing infrastructure failures, with several pointing to the challenge of rate-limiting and abuse detection for legitimate-seeming AI requests. This matters because it exposes a growing attack surface as AI tools proliferate across developer platforms.
- Hacker News·INFRA·1mo ago·+1 source
Zero-copy GPU inference from WebAssembly on Apple Silicon
A developer has demonstrated that WebAssembly modules can share memory directly with the GPU on Apple Silicon, eliminating the usual serialization overhead. The technique chains three components: mmap for page-aligned memory, Metal's bytesNoCopy API to wrap that pointer without copying, and Wasmtime's custom memory allocator to use the same region as Wasm linear memory. The result is that a Wasm guest and the GPU read and write the same physical bytes, with no intermediate buffers. The developer measured zero RSS delta compared to 16.78 MB for the copy path. This matters because it turns Wasm into a viable control plane for stateful AI inference on Apple hardware, with near-zero overhead between the sandbox and the accelerator.
- Anthropic·INFRA·1mo ago
Anthropic orders gigawatts of TPU capacity from Google and Broadcom
Anthropic has signed a deal with Google and Broadcom for multiple gigawatts of next-generation TPU capacity, expected online from 2027. The compute will support Claude's frontier models as the company's run-rate revenue hits $30 billion, up from $9 billion at end-2025. Over 1,000 business customers now spend more than $1 million annually, double the figure from February. Most capacity will be sited in the US. Anthropic trains Claude across AWS Trainium, Google TPUs, and NVIDIA GPUs, maintaining AWS as primary partner while expanding its multi-cloud strategy. The scale of this order reflects genuine demand, not speculation.