Sources: DeepSeek and Huawei are collaborating on future AI chip designs; DeepSeek plans to start using Ascend 950DT for pre-training as soon as 2026's end
THE SO WHAT
A model lab shaping Huawei’s Ascend 950DT roadmap is a direct attempt to close the loop between frontier training needs and domestic silicon—reducing dependence on US-controlled GPUs. Anyone exposed to China-facing AI demand should assume a more credible non-Nvidia path by 2027 and plan for a bifurcated tooling and optimization stack.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIPartnering with Blacksmith
If agents can write production-scale code, the new bottleneck is proving that code is safe, performant, and compliant at the same speed. Expect “agent-era CI/CD” vendors like Blacksmith to become a required layer in any serious agent deployment, not a nice-to-have devtool.
Applied AIHackers hit AI coding tool OpenCode, running up charges on users’ AI accounts
Agentic dev tools that proxy user API keys are now a direct fraud surface—OpenCode’s breach shows the blast radius is not data loss but runaway model spend. If you ship AI tools that touch customer keys, move key storage, rotation, and spend caps into your core product requirements, not “nice-to-have” security hardening.
Applied AIReplika’s founder built the AI assistant she pitched at Disrupt 12 years ago
Wabi’s launch from Replika’s founder shows the personal AI assistant thesis has real persistence—what changed is that infra and models finally caught up. The competitive bar for assistants is shifting from “can it answer” to “does it remember, adapt, and feel aligned with me over years.”
Applied AIThe Fear of a One-Man-Band Weaponizing AI Agents Is Becoming Reality
A lone operator compromising hundreds of sites with AI agents collapses the cost of coordinated attacks—defense now has to assume “small team, big footprint” as the default threat model. Security teams should be testing how their own systems fare against automated recon and exploitation, not just human red teams.