AI Infrastructure news and analysis on Web Pulse: 34069 curated articles tracking the latest AI Infrastructure developments, tools, and research, updated continuously from vetted sources.
The Trump family's affiliated ventures have secured $620 million in Pentagon loans for AI infrastructure and announced a $6 billion merger between Trump Media & Technology Group and TAE Technologies in December 2025, acc…
Typesafe AI, founded by former OpenAI researcher Diogo Almeida, emerged from stealth Tuesday with $40 million in funding and introduced Jev, a model it says is hundreds of times cheaper and faster than frontier models fr…
A September 2026 survey of self-hosted inference orchestrators compared LocalAI, exo, GPUStack, vLLM, Ollama, llama.cpp, LiteLLM, Xinference, NVIDIA Dynamo/llm-d, SkyPilot and dstack across modalities, multi-machine supp…
Jeff Dean, former Chief Scientist at Google, said on the Dwarkesh Podcast in February 2025 that reinforcement learning and improved electronic design automation could cut cutting-edge chip design time from 18–30 months t…
Ds4 now supports directional steering, a runtime activation edit that applies a normalized f32 direction per transformer layer during inference, with steering files shaped 43x4096 for DeepSeek V4 Flash, 45x4096 for GLM 5…
A developer has published a CI-runnable script that fetches a gateway's pricing.usd.json file and flags near-duplicate model identifiers whose cache-read prices diverge, after finding that strings like claude-fable-5 and…
A GPU cloud startup tested its own pricing calculator's claim of supporting up to 512 GPUs and found real-world caps of 8 GPUs per pod on RunPod and 16 GPUs on workstation cards (4 on B200) via Vast.ai, according to the …
President Donald Trump announced on Truth Social that he plans to establish an "AI Force" modeled after the Space Force and appoint an AI "czar," saying the technology could eventually represent as much as 25% of U.S. GD…
A developer identified and fixed a KV-cache offloading bug in vLLM that caused incorrect chunking for models with mixed KV-cache groups. The existing implementation assumed a single KV-cache group layout and relied solel…
Huawei unveiled the Atlas 960E SuperPoD at HUAWEI CONNECT 2026 in Shanghai on September 17, an AI cluster that links 4,096 Ascend neural processing units into a single logical machine with unified memory addressing, targ…
A developer published a demo showing how the OWASP Core Rule Set (CRS), running via Coraza inside Solo Enterprise for agentgateway, can inspect LLM prompts and MCP tool calls rather than just URLs and headers. The setup …
Zizka-ai released ZizkaDB, an open-source, self-hosted audit trail for AI agents that logs each agent step with a parent_id so teams can trace a tool call or response back to its root cause. The stack ships via a Docker-…
Aptos is building infrastructure for autonomous AI agent commerce anchored by its x402 payment protocol and a partnership with io.net that lets AI agents rent idle GPU capacity on demand, with settlements in USDC finaliz…
Post-deployment operations consume roughly 84% of total enterprise AI budgets over a three-year total cost of ownership, while visible acquisition costs such as initial API consumption and baseline cloud compute account …
Huawei unveiled the Atlas 960E SuperPoD at HUAWEI CONNECT 2026 in Shanghai on September 17, a computing cluster packing up to 4,096 Ascend Neural Processing Units under a single unified memory-addressing scheme. Huawei c…
RuntimeWire's static analysis of Claude Desktop 2.2553.1 found an unreleased "Self-hosted execution" configuration that can redirect Code session traffic to an organization-specified HTTPS session host, with the app's ow…
Pirate Face launched a decentralized peer-to-peer mirroring service that converts open models from Hugging Face — including LLMs, image, audio, and datasets — into checksum-verified torrents, with each file carrying its …
Immanuel Gabriel, founder of FreshContext, has built an evaluation engine that scores AI agent context on freshness, attribution, and coherence before it reaches the model, addressing a gap that retrieval scoring alone d…
Ternary Bonsai 2 27B, a ternary-weight multimodal model built on a Qwen3.8 27B base, ships at roughly 6GB with weights restricted to {-1, 0, +1} at 1.76 bits per weight with FP16 group scales, according to the model's pr…
The aamio rendezvous MCP server, version 0.7.3, was published to the official MCP registry on 2026-09-20 as a hosted remote over streamable-http, according to the mcpindex registry index. mcpindex's semantic screen found…