ByteDance’s AI Ambitions: Huawei Chips and U.S. Curbs in the AI Race

As ByteDance, the Chinese parent company of TikTok, ventures deeper into artificial intelligence (AI), it faces a world reshaped by geopolitical tensions and shifting market dynamics. A new report suggests that ByteDance is planning to train its next large-language AI model using Huawei’s Ascend 910B chips. This move, driven largely by U.S. export restrictions, marks another instance of China’s tech giants turning inward for technological solutions as they navigate increasingly complex supply chains and regulatory environments.

In this article, we will explore the motivations behind ByteDance’s decision to rely on domestic suppliers like Huawei, the impact of U.S. chip restrictions on the global AI race, the role of AI in ByteDance’s future strategies, and the broader implications for the technology industry.

ByteDance and Huawei: The Partnership in Focus

The collaboration between ByteDance and Huawei signals a significant shift in China’s tech landscape. ByteDance, known globally for its social media platform TikTok, has diversified its portfolio significantly over the past few years, with AI playing an increasingly central role. Meanwhile, Huawei, once a telecommunications giant known for its smartphones and networking equipment, has repositioned itself as a key player in the AI hardware market following its own struggles with U.S. sanctions.

ByteDance’s decision to adopt Huawei’s Ascend 910B chip for its upcoming AI model stems primarily from the U.S. government’s escalating restrictions on AI-related technology exports. In 2022, the U.S. imposed curbs on the export of advanced chips, such as those produced by Nvidia, to China. This move hampered the ability of Chinese tech companies to access cutting-edge technology essential for developing large-language models and other AI applications.

As a result, ByteDance and other Chinese companies have increasingly turned to domestic alternatives like Huawei. The Ascend 910B chip, though not as powerful as Nvidia’s premium graphics processing units (GPUs), is seen as a viable option for developing AI models in China. According to insiders familiar with the matter, ByteDance has already placed orders for over 100,000 Ascend 910B chips in 2023, with fewer than 30,000 delivered by mid-year. While this represents a significant investment in domestic AI capabilities, the supply of Huawei chips has been slower than anticipated, potentially delaying ByteDance’s AI development plans.

The Impact of U.S. Chip Restrictions

The U.S. government’s export restrictions on advanced AI chips have profoundly impacted China’s tech industry. The move has disrupted supply chains and forced companies like ByteDance to accelerate their reliance on domestic suppliers. Nvidia’s chips, in particular, are considered the gold standard in AI development, offering unmatched performance in the computationally intensive task of training AI models.

Large-language models, such as those developed by ByteDance, require immense computing power to process vast amounts of data and perform tasks like natural language understanding and generation. Nvidia’s GPUs have been the cornerstone of many AI innovations, powering everything from autonomous vehicles to language models like OpenAI’s GPT.

However, with U.S. restrictions in place, Chinese companies have faced difficulties accessing these critical components. As a result, they have had to seek alternatives within China’s own semiconductor industry. Huawei’s Ascend chips, though less powerful than their Nvidia counterparts, have emerged as a key option for companies looking to continue their AI development efforts without relying on U.S. technology.

Huawei’s Ascend 910B: Strengths and Limitations

The Ascend 910B chip is part of Huawei’s broader strategy to establish itself as a major player in the AI hardware market. Huawei has poured significant resources into the development of AI chips, positioning them as a solution for China’s growing demand for AI technology.

While the Ascend 910B has proven effective for certain AI tasks, such as inference (where pre-trained models make predictions), it lacks the raw computational power needed for large-scale training of AI models. Training AI models, especially large-language models, requires massive amounts of data processing and computation, often necessitating the use of ultra-high-performance chips like Nvidia’s GPUs.

According to sources familiar with ByteDance’s AI efforts, the new model it plans to develop with Huawei’s chips will be less powerful than Doubao, ByteDance’s flagship large-language model launched in August 2023. This suggests that while ByteDance can continue making progress in AI development using domestic chips, there will be limitations in terms of performance and capabilities.

ByteDance’s AI Strategy: From Doubao to Future Models

AI is at the heart of ByteDance’s long-term strategy. The company’s first major foray into large-language models came with the launch of Doubao, an AI chatbot that was introduced in August 2023. Doubao competes directly with models like OpenAI’s GPT and Google’s Bard, offering capabilities in natural language processing, content generation, and other AI-driven applications.

Doubao has already found applications within ByteDance’s ecosystem, powering tools like Jimeng, a text-to-video tool, and two video-focused Doubao models designed to compete with OpenAI. ByteDance’s integration of AI into its platforms represents a broader trend within the tech industry, where AI is becoming a key differentiator for products and services across sectors such as e-commerce, gaming, and social media.

However, ByteDance’s reliance on domestic AI chips like Huawei’s Ascend 910B could place limitations on the company’s ability to compete globally in terms of AI capabilities. As the competition in AI heats up, companies with access to the most advanced hardware, like Nvidia’s chips, are likely to maintain a significant advantage in terms of performance and innovation.

Supply Chain Constraints and Delays

One of the biggest challenges facing ByteDance in its efforts to develop a new AI model using Huawei’s chips is the constrained supply of hardware. Despite ordering over 100,000 Ascend 910B chips in 2023, ByteDance has only received fewer than 30,000 as of July, according to sources familiar with the situation. This slower-than-expected pace of chip deliveries has delayed the company’s AI development efforts and prevented it from setting a concrete timeline for the release of its next large-language model.

The tight supply of Huawei chips underscores the broader challenges facing China’s semiconductor industry. Although the Chinese government has made significant investments in the domestic chip sector, building a self-sufficient supply chain for advanced AI chips is a complex and time-consuming process. For companies like ByteDance, this means that while they can continue to innovate using domestic technology, they may face delays and limitations in terms of performance compared to their international competitors.

The Broader Implications for the Technology Industry

ByteDance’s efforts to develop a new AI model using Huawei’s chips are emblematic of the broader shifts taking place in the global technology landscape. As geopolitical tensions between the U.S. and China continue to rise, technology companies are increasingly caught in the crossfire. The U.S. government’s restrictions on AI chip exports to China have forced Chinese companies to accelerate their efforts to develop domestic alternatives, with mixed results.

While companies like ByteDance and Huawei have made impressive strides in developing their own AI capabilities, the gap between domestic and international technology remains significant. Nvidia’s GPUs, for example, are still considered far superior to any chips currently available in China, making it difficult for Chinese companies to compete on a level playing field when it comes to large-scale AI development.

At the same time, the growing importance of AI in industries ranging from gaming to e-commerce means that companies without access to the most advanced hardware could find themselves at a competitive disadvantage. For ByteDance, the decision to rely on Huawei’s chips represents a pragmatic solution to the immediate challenges posed by U.S. sanctions, but it also highlights the longer-term difficulties that Chinese tech companies will face in maintaining their edge in the global AI race.

Conclusion

ByteDance’s partnership with Huawei to develop its next AI model using the Ascend 910B chip underscores the profound impact of U.S. technology export restrictions on China’s tech industry. As companies like ByteDance look to navigate these challenges, they are turning inward, relying on domestic suppliers like Huawei to meet their AI development needs. While this strategy has allowed ByteDance to continue making progress, the limitations of Huawei’s chips and the constrained supply of hardware may ultimately slow the company’s AI ambitions.

The broader implications for the technology industry are clear: as AI becomes increasingly central to product differentiation and innovation, companies with access to the most advanced hardware will have a significant competitive advantage. For Chinese tech giants like ByteDance, finding ways to overcome these challenges will be crucial in the years ahead.