Realtime AI News
Huawei Pangu Pro Trains 505 Billion Parameters Without Nvidia: Supply Chain Tells Different Story
Huawei's Pangu Pro model has reportedly been trained at a 505-billion-parameter scale without using Nvidia hardware, according to Tech Times. Supply chain sources, however, tell a different story, casting doubt on whether the training infrastructure is truly Nvidia-free.
Huawei's Pangu Pro large language model has reportedly been trained at a scale of 505 billion parameters without using Nvidia hardware, according to Tech Times.
If accurate, the claim would mark a milestone in Huawei's push for self-sufficient AI training, coming against the backdrop of export controls that have pushed major Chinese AI labs to reduce their reliance on Nvidia GPUs.
The report notes, however, that supply chain sources tell a different story, suggesting Huawei's training infrastructure may not be as Nvidia-free as the headline claim implies.
Such discrepancies between official claims and supply chain accounts are common in coverage of Chinese AI compute: public narratives and supplier-level intelligence frequently diverge, and the Pangu Pro report is no exception.
Why it matters: if Huawei can genuinely train a model of this scale without Nvidia, it would demonstrate that domestic compute can support frontier-scale training, reshaping expectations for both Nvidia's China business and the pace of domestic chip substitution.
If the supply chain account is closer to the truth, it would suggest Nvidia chips are still reaching Chinese compute markets through various channels, putting the real progress of domestic substitution into question.
What to watch next: whether Huawei discloses training details for Pangu Pro, whether supply chain data further verifies which chips were actually used, and how domestic AI silicon performs in ultra-large-scale training.
Why it matters
The two competing narratives around Pangu Pro's training directly test the credibility of both domestic compute substitution and Nvidia's China business, so supply chain evidence deserves close tracking.
Nearby Updates
All08/02, 01:33
Open-Source 700B Models Released Two Days Apart as Korea Nears Sovereign AI Elimination Round
Tech Times reports that two open-source 700B-parameter models were released within two days of each other, just as Korea's sovereign AI push enters what the report calls an "elimination round." The back-to-back releases raise the bar for open-weight models and sharpen the competitive stakes for national AI programs racing to reduce dependence on US and Chinese technology.
08/02, 01:57
RufRoot: Patching Doesn't Undo Poisoning — The MCP Flaw That Persists Inside AI Memory
A newly disclosed flaw named RufRoot targets the Model Context Protocol and persists inside AI memory, meaning patching the issue does not undo the poisoning it enables. The finding highlights how memory-based attacks on AI agents can outlive code-level fixes.
08/02, 01:57
Okta Bets $200M That AI Agents Need Their Own Identity Threat Detection
Okta is betting $200 million that AI agents need their own identity threat detection, treating agent security as a category distinct from human identity security. The move comes as autonomous agents increasingly hold credentials and act on enterprise systems, creating a new attack surface that existing identity tools were not built to cover.
08/01, 22:46
Google launches Gemini Spark AI agent in Malaysia for AI Pro/Ultra subscribers
Google has launched its Gemini Spark AI agent in Malaysia, available to Google AI Pro and Ultra subscribers. The rollout marks a significant step for Google's agent products in Southeast Asia, with access initially limited to paying subscribers.