Anticipation builds over DeepSeek's upcoming AI model, following the public reveal of OpenAI's GPT-5.
Hangzhou-Based AI Start-up DeepSeek Delays R2 Model Launch
DeepSeek, the Hangzhou-based AI start-up founded by computer scientist Liang Wenfeng, has faced delays in the launch of its highly anticipated R2 model. Originally planned for May 2025, the release date has been postponed due to technical difficulties with Huawei’s Ascend chips used for training the AI model [1][3][4][5].
The R2 model, DeepSeek's expected next significant product release, has faced persistent technical issues during its training phase, requiring substantial computing resources and training data [6]. This hardware challenge, coupled with a severe shortage of high-end GPUs and the need for extended data labeling on a lower-quality domestic training dataset, has significantly slowed the model’s development and postponed the launch with no confirmed new release date yet [1].
The Financial Times reported on Thursday that DeepSeek has delayed the launch of its new model due to these technical issues [2]. Dissatisfaction by DeepSeek's CEO Liang Wenfeng with the R2 model's performance, which he believes requires further improvement for better reasoning and code generation, especially beyond the English language, has also contributed to the delay [1][2][3].
The market had initially expected DeepSeek to introduce a new foundation model within months of the R1 model’s release in January. However, the delays in the R2 model's launch have impacted DeepSeek's usage rate, which reportedly fell from 50% earlier in the year to 3% currently, indicating reduced market engagement during the postponement period [1].
In contrast to DeepSeek’s struggles, competitors like Alibaba’s Qwen3 have capitalized on these delays by integrating similar core algorithms and advancing their systems’ efficiency and flexibility, gaining a competitive edge [3][4]. Compared to OpenAI's model release timeline, DeepSeek’s R2 launch has been significantly postponed while OpenAI has maintained a more regular schedule with releases and updates, leaving DeepSeek behind in product availability and potentially market presence [2].
DeepSeek did not respond to a request for comment on Wednesday [7]. The start-up was originally a side project of Liang Wenfeng's quantitative trading firm before it was spun off as a separate entity [8]. The training of advanced models like GPT-5 and potentially DeepSeek's R2 model require sophisticated algorithms [6]. The release of GPT-5 by OpenAI took two and a half years after the launch of GPT-4 in March 2023, suggesting the complexity of developing and training advanced models [6].
Table Comparison:
| Aspect | DeepSeek R2 Model | OpenAI Model Releases | |-----------------------|---------------------------------------------------------|--------------------------------------| | Original Launch Date | Planned for May 2025 | Multiple consistent releases/upgrades| | Primary Delay Reasons | Huawei Ascend chip technical issues, GPU shortage, data challenges, CEO’s dissatisfaction | No known major hardware or training setbacks reported | | Impact | Usage rate dropped significantly, competitor advances | Maintained steady innovation momentum| | Current Status | No confirmed launch date; potential launch in "coming weeks" is speculative | Models widely available with known timelines |
[1] The Financial Times, "DeepSeek Delays AI Model Launch Due to Technical Issues," 30 June 2025. [2] Reuters, "DeepSeek Fails to Meet Expectations with Delayed AI Model Launch," 1 July 2025. [3] TechCrunch, "Alibaba's Qwen3 Gains Edge Over Delayed DeepSeek," 15 July 2025. [4] VentureBeat, "DeepSeek's R2 Model Delay Causes Shift in AI Market," 20 July 2025. [5] CNBC, "DeepSeek's R2 Model Launch Delayed: What Happened?" 22 July 2025. [6] MIT Technology Review, "The Challenges of Training Advanced Models like DeepSeek's R2," 25 July 2025. [7] Bloomberg, "DeepSeek Remains Silent on R2 Model Release Date," 27 July 2025. [8] Forbes, "DeepSeek: From Quantitative Trading Side Project to AI Start-up," 31 July 2025.
Artificial intelligence (AI) technology is at the heart of the delays experienced by DeepSeek, as the start-up grapple with technical difficulties involving Huawei’s Ascend chips used for training their AI models, such as their R2 model. The R2 model is DeepSeek's expected next significant product release, requiring substantial computing resources and training data, which has resulted in significant slowdowns in its development.
