Back to Archive
Sent to 17 readers

Race to AGI Daily Digest - Saturday, April 4, 2026

Share:

TLDR

DeepSeek V4 is training on Huawei chips instead of NVIDIA, accelerating China’s push for a self‑reliant AI hardware stack.

Read the DeepSeek–Huawei story ->

UnionPay’s APOP standard gives AI agents a formal way to initiate payments across China and beyond, moving agents into real financial rails.

How APOP powers agent payments ->

Anthropic is acquiring Coefficient Bio in a $400M AI drug discovery push while also tightening access to powerful tooling like OpenClaw.

Anthropic’s biotech play ->

Microsoft’s new multimodal MAI models and Galaxea’s robotics funding show agents spreading from chat windows into vision, tools, and the physical world.

Microsoft’s MAI models in Foundry ->

The Full Story

All week we’ve watched the ARM roller coaster and the Claude Code fake leak as stand‑ins for the AI race. Today, the story gets more concrete: other regions are quietly wiring up their own full stacks. First, chips. DeepSeek’s new V4 model is training on Huawei accelerators instead of NVIDIA, a direct response to US export limits. That’s not a lab toy; it’s a bet that China can stand up a competitive stack from silicon to serving. You can dive into the details in the DeepSeek V4 on Huawei story ->. It slots neatly into the broader AI chip export narrative -> we’ve been following since earlier in the week. Then come the agents. UnionPay’s new APOP standard gives AI agents a formal way to make payments across China and beyond. That turns our “Agentic AI Toolchains Leave the Lab” storyline into something that can actually move money at national scale. Check how they frame it in the APOP agent payments piece ->. Anthropic, meanwhile, is doing two things at once. On the upside, it’s buying Coefficient Bio in a $400M push into AI drug discovery, pulling Claude‑class models into wet‑lab science. On the downside, a top Hacker News post is calling out that Claude Code subs can no longer use OpenClaw. That’s exactly the kind of access tightening we expected after the fake toolkit saga, and it keeps our agent‑safety storyline very real. You can track Anthropic’s moves on their company page -> and the Coefficient Bio deal ->. Layer in Microsoft adding three multimodal MAI models to Foundry, covered in the MAI models update ->, and Galaxea’s funding for embodied robots, and you see the pattern: agents now have eyes, hands, and bank accounts. On the tape, the market keeps backing our Thursday thesis. Intel and AMD are green again, while ARM and ASML are down and Tesla slumps. Hardware stories with a credible software and agent angle are still getting the nod.

Get This Delivered Daily

Join thousands of AI professionals who start their day with Race to AGI.

No spam, ever. Unsubscribe anytime.