Company@X — 2026-07-04#

Signal of the Day#

Wafer AI published benchmarks demonstrating that engineers have successfully optimized GLM 5.2 inference on AMD MI355X GPUs. This enables roughly 80% of Nvidia Blackwell (B200) throughput at more than a 2x cost reduction, marking a highly consequential milestone in breaking Nvidia’s monopoly on high-performance AI inference infrastructure.

Key Announcements#

Wafer AI · Source Engineers have successfully served the GLM 5.2 model on AMD MI355X accelerators, achieving 2626 tokens/sec per node and 213 tokens/sec on a single stream. The benchmark proves AMD’s rapidly improving software stack, delivering ~80% of the throughput of Nvidia’s B200 but at less than half the cost.

Hugging Face · Source Hugging Face released a major update to its Diffusers library, bringing first-class support for highly anticipated new generation models including Ideogram4, MotifVideo, and DiffusionGemma. This rapid platform integration of new image and video pipelines cements Hugging Face’s position as the default development hub for open multimodal architecture.

Smol Cloud · Source Smol Cloud has officially launched its beta, offering instantaneous virtual computers specifically designed to host AI agents. The platform dynamically allocates up to 4 CPUs and 8GB of RAM per instance and features a CLI that allows developers to spin up hundreds of isolated cloud agents simultaneously at a highly competitive baseline cost.

Hugging Face · Source CEO Clément Delangue published a strategic defense of open-source AI alongside a curated list of 250 open AI milestones, framing the current industry trajectory as a critical battle between open participation and closed-door corporate gatekeeping. The messaging doubles down on Hugging Face’s positioning as the champion of open ecosystems against proprietary, highly capitalized frontier labs.

Also Noted#

• OpenClaw (Source): The volunteer-driven open-source project surpassed 100,000 issues and pull requests in just 222 days, highlighting massive developer momentum operating entirely without venture capital.
• Grok (Source): xAI deployed new America 250-themed generation templates live within its Grok Imagine suite.
• HyLight (Source): YC S23 alumni HyLight Aero is successfully deploying automated airships to inspect infrastructure like power lines and pipelines across France, replacing high-emission and high-risk helicopter flights at scale.