It seems that the good folks over at DeepSeek, a Chinese artificial intelligence (AI) firm, have found themselves in a bit of a pickle. Their latest AI creation, the R2, has hit the skids, all thanks to a spot of bother with Huawei’s chips. You see, the Chinese government, ever so keen on reducing reliance on American tech, urged DeepSeek to give Huawei’s Ascend processors a whirl. But, alas, it appears that these chips are more adept at causing headaches than solving them.
After the successful launch of R1 in January, the powers that be decided it would be a splendid idea to swap out those trusty Nvidia chips for a bit of homegrown tech. However, the best-laid plans of mice and men often go awry, and so it proved here. Despite the best efforts of Huawei’s finest engineers, who were dispatched to assist, DeepSeek found itself floundering in a sea of technical difficulties. The R2, originally slated to make its grand entrance in May, was forced to take a backseat, much to the delight of rival firms.
This little hiccup has not only delayed the R2 but has also cast a shadow over the capabilities of Chinese chips. It seems that when it comes to the heavy lifting of training AI models, Huawei’s Ascend just can’t hold a candle to the American competition. To add insult to injury, the Chinese government is now asking its tech companies to justify their continued reliance on Nvidia’s H20 chips, pointing out the usual suspects: instability, slow connections, and a general lack of oomph.
In a twist worthy of a farce, DeepSeek has resorted to a bit of a juggling act, using Nvidia chips for the training phase and Huawei’s for the inference part, where the model does its thinking and chatting. Meanwhile, the labeling of data for the R2 took longer than anticipated, adding another layer of complexity to the whole affair.
But fear not, for all is not lost. According to Ritwik Gupta, an AI researcher at the University of California, Berkeley, “Models are commodities that can be easily swapped out.” He also noted that developers are flocking to Alibaba’s Qwen3, a model that’s both powerful and flexible, and has even borrowed a few tricks from DeepSeek’s playbook, including a training method that sharpens the model’s logical thinking and boosts its efficiency.
Despite the current setbacks, Gupta remains optimistic, stating, “Just because we’re not seeing leading models trained on Huawei today doesn’t mean it won’t happen in the future. It’s a matter of time.” And speaking of time, Nvidia, ever the diplomat in the U.S.-China tech rivalry, has agreed to share some of its profits from China with the U.S. government, paving the way for the sale of its H20 chips back in the Middle Kingdom.
Chinese media, ever the optimists, suggest that the R2 might finally see the light of day in the coming weeks, as DeepSeek scrambles to catch up with the competition. One can only hope that the next time around, the chips will be a bit friendlier to the cause. 🍞🛠️
Read More
- Violence District Killer and Survivor Tier List
- All Data Pad Locations (Week 1) Destiny 2
- Top 8 UFC 5 Perks Every Fighter Should Use
- All Mafia The Old Country Returning Characters
- German Twitch streamer goes viral for playing Elden Ring & Street Fighter at a rave
- Unlock the Magic: New Arcane Blind Box Collection from POP MART and Riot Games!
- Unleash Devastation: Top Rupture Teams to Dominate in Limbus Company!
- USD ZAR PREDICTION
- Prestige Perks in Space Marine 2: A Grind That Could Backfire
- Dune: Awakening is Already Walking Back a Planned Change for the Next Update
2025-08-15 17:23