VC
Value Add VC
⚡HomePulse⚡Helpful Apps📝Blog
← Value Add PulseAI

Huawei-Led Team Post-Trains DeepSeek's 1.6-Trillion-Parameter Model on Ascend Chips, Not Nvidia

A research group including Huawei completed full-parameter post-training of DeepSeek's 1.6-trillion-parameter V4-Pro model using a cluster of at least 1,000 Huawei Ascend 910C chips -- demonstrating that a frontier-scale Chinese model can be trained end-to-end on domestic silicon. It's a milestone in China's drive to break free of US chips and the CUDA ecosystem.

DeepSeek V4-Pro
Model
1.6 trillion (MoE)
Parameters
1,000+ Ascend 910C
Chips
Full-parameter post-training
Task
Non-Nvidia frontier stack
Significance
TC
Trace Cohen
Early-stage VC & angel · Founder, New York Venture Partners
June 20, 2026
1 min read
KEY TAKEAWAYS FOR VCs & FOUNDERS
1

Training a 1.6T-parameter model on Ascend, not Nvidia, undercuts the premise of US chip export controls

2

It validates a full Chinese AI stack -- model plus domestic silicon plus CANN software

3

Reduced dependence on Nvidia reshapes the geopolitics of AI compute

4

DeepSeek's efficiency-first approach keeps pressuring Western labs on cost

TC
The VC Read · Trace's TakeTrace Cohen

This is the quiet headline that should worry the people betting export controls indefinitely cap China's AI. Post-training a 1.6T-parameter model on 1,000 Ascend chips proves the full domestic stack -- silicon plus CANN software plus model -- works at frontier scale, even if it's less efficient than Nvidia. The strategic read: chip bans buy time, not permanent advantage, and they accelerate exactly the domestic-silicon investment they were meant to prevent. For investors, watch the second-order effect -- a credible non-CUDA stack is a long-term tail risk to Nvidia's pricing power.

🤖 AI Landscape →⚡ AI Chip Wars →

A research team that includes Huawei has completed full-parameter post-training of DeepSeek's 1.6-trillion-parameter V4-Pro model on a cluster of at least 1,000 Huawei Ascend 910C accelerators, according to Tom's Hardware. The work is one of the clearest demonstrations yet that a frontier-scale model can be trained end-to-end on Chinese silicon rather than Nvidia GPUs.

The significance is geopolitical as much as technical. US export controls are premised on the idea that restricting access to leading-edge Nvidia chips slows China's AI progress. A 1.6T-parameter model post-trained on domestic accelerators -- backed by Huawei's CANN software stack with fused operators tuned for the hardware -- suggests China is assembling a credible full-stack alternative, even if Ascend still trails Nvidia on raw efficiency.

“US export controls are premised on the idea that restricting access to leading-edge Nvidia chips slows China's AI progress.”

DeepSeek has built its reputation on doing more with less, and this milestone extends that story from inference into heavy training workloads. For Western labs, the pressure is twofold: a capable open-weight competitor that keeps driving prices down, and a reminder that the compute chokehold the US has counted on is loosening.

ShareXLinkedInEmail
More onDeepSeek →Nvidia →

Originally reported by Tom's Hardware. Analysis and editorial commentary by Value Add Pulse.

← Back to Pulse

Markets Now

live
SPCX▲+2.52%
$224.10
CBRS▲+1.04%
$324.40
SPY▲+0.16%
5,931.80
QQQ▲+0.12%
19,972.10
NVDA▼-0.71%
$154.20
MSFT▲+0.25%
$477.30
GOOGL▲+1.22%
$207.90
META▲+0.25%
$651.40

Read Next

AI

Crunchbase: Classic SaaS Isn't Coming Back -- Agentic AI Is Replacing It

A Crunchbase analysis argues the traditional SaaS model isn't recovering from its multi-year slump -- something bigger is supplanting it: agentic AI that does the work rather than just organizing it. The piece reframes the software downturn not as a cyclical dip but as a structural handoff from seat-based tools to outcome-based AI agents.

AI

Signal's Meredith Whittaker: AI Chatbots 'Are Not Your Friends'

Signal president Meredith Whittaker is warning users not to mistake AI chatbots for confidants, arguing the intimacy they simulate masks a surveillance-and-data business underneath. Her caution lands amid a boom in companion and assistant AI that increasingly positions itself as an emotional relationship.

AI

Beyond Siri: The Practical AI Features Coming to Your iPhone in iOS 27

Apple is previewing a slate of practical, on-device AI features in iOS 27 that go well beyond its long-delayed Siri overhaul, per TechCrunch. The emphasis on concrete, everyday utility -- rather than a single flashy assistant -- signals Apple's bid to make AI feel useful and trustworthy to a billion-plus users.

@Trace_Cohen·t@nyvp.com