Is America Falling Behind China in the LLM Race? Yes… and No

The large language model (LLMs) race is heating up, and there’s good reason to suggest that the gap between the U.S. and other countries like China is closing.

Earlier this week, HuggingFace, a leading platform for AI research and benchmarking, released the Open LM Leaderboard v2 with new benchmarks, saying it was time to revisit how learning models were rated, with the gap closing between top LLMs.

The new leaderboard features Qwen-2-72B Instruct in the number one spot, a model developed in China by Alibaba Cloud. Another Chinese model featured in fourth place; Yi 1.5.

This highlights that China is gaining a deeper foothold in the world of open-source AI development and closing in on key performance benchmarks across the board.

While the U.S. still has a rich ecosystem of providers, including OpenAI, Anthropic, Google Claude, Microsoft, Meta, Amazon, and Nvidia, the gap in performance is still gradually closing. Here’s why.

Table of Contents Table of Contents

1. Key Takeaways
2. How China is Challenging U.S AI Development
3. Hugging Face's Top 10 AI LLMs & Country of Origin
4. Is the U.S. Falling Behind? Not Just Yet
5. So What's Next for the AI Market?
6. The Bottom Line
7. References

Show Full Guide

Table of Contents

1. Key Takeaways
2. How China is Challenging U.S AI Development
3. Hugging Face's Top 10 AI LLMs & Country of Origin
Show Full Guide
4. Is the U.S. Falling Behind? Not Just Yet
5. So What's Next for the AI Market?
6. The Bottom Line
7. References

How China is Challenging U.S AI Development

The Chinese AI ecosystem had another significant win just recently with the release of Kling AI, a text-to-video model produced by Kuaishou Technology.

Kling AI not only produces videos that match Sora‘s text-to-video content but also boasts the ability to create videos of up to two minutes in length (compared to Sora’s one-minute length).

We’ve also seen how Baidu’s Ernie bot, a Chinese alternative to ChatGPT has gone on to have over 200 million users, with its application programming interfaces (APIs) used 200 million times per day.

When we consider this alongside the success of Qwen-2-72B Instruct and Yi 1.5, the Chinese AI ecosystem is growing significantly.

At the same time, China is actively investing in innovation with a report produced by data analytics firm Govini finding that China has more patents than the U.S. in 13 of 15 critical technology areas.

In addition, China produces almost half of the world’s leading AI researchers, with the number of AI researchers developed in China increasing from 29% to 47% between 2019 to 2022.

China is not only a force to be reckoned with in the market but also has the potential to seriously challenge U.S. AI development over the long term.

Hugging Face’s Top 10 AI LLMs & Country of Origin

• Qwen/Qwen2-72B-Instruct – China (developed by Alibaba Cloud)
• meta-llama/Meta-Llama-3-70B-Instruct – United States (developed by Meta)
• microsoft/Phi-3-medium-4k-instruct – United States (developed by Microsoft)
• 01-ai/Yi-1.5-34B-Chat – Likely China, as the naming convention suggests a Chinese origin
• CohereForAI/c4ai-command-r-plus – Canada (developed by Cohere)
• abacusai/Smaug-72B-v0.1 – United States (developed by Abacus.AI)
• Qwen/Qwen1.5-110B – China (developed by Alibaba Cloud)
• Qwen/Qwen1.5-110B-Chat – China (developed by Alibaba Cloud)
• microsoft/Phi-3-small-128k-instruct – United States (developed by Microsoft)
• 01-ai/Yi-1.5-9B-Chat – Likely China, similar to other models from “01-ai”?

Via HuggingFace, June 2024

Is the U.S. Falling Behind? Not Just Yet

As of 2024, the U.S. is inarguably the most dominant country in the AI race. If we look at LLM providers, the U.S. is home to OpenAI and ChatGPT, Anthropic and Claude 3, Microsoft and Microsoft Copilot, Meta and Llama 3, X.Ai and Grok.

According to CB Insights, total AI funding reached $42.5 billion in 2023, $31.0 billion of which went to the U.S., compared to $6.4 billion for Europe, and $9.4 billion to Asia. This demonstrates that the U.S. is the undisputed hub of VC investment in the sector.

So is the U.S. falling behind? The general consensus among researchers we communicated with was ‘no’.

Nicholas Rioux, CTO of Labviva, told Techopedia:

“America is currently ahead of the curve in the development of LLM models and is easily three years ahead of major technological competitors like China. It could lose this position by bogging innovators down with needless regulations.”

Tom Burns, an AI researcher at Cornell University, shared similar thoughts with Techopedia.

“I do not think the U.S. is falling behind on LLM development, although the landscape is becoming more sophisticated and competitive.

“For instance, geopolitically-allied nations like Japan have seen the recent start-up Sakana AI emerge as a possible key player there, however they are strategically not focussing on training models from the ground up but rather merging and combining pre-trained models.”

That being said, Burns did note that China is the main threat to current U.S. dominance simply because it has the resources to invest in the training of large foundation models. Europe is seeing lots of investment activity but faces more potential regulation under EU law.

So What’s Next for the AI Market?

Based on China’s success in the realm of open-source LLMS, the development of solid proprietary products like Kling AI, and the large proportion of AI researchers from the region, it is likely that the gap between Chinese and American AI research will continue to close.

While U.S. sanctions could restrict industry growth, Rioux suggests that they could actually strengthen the Chinese technological ecosystem.

“Initial U.S. government restrictions on exports of open source LLM technologies and supportive hardware will slow China down a bit, but it will have the unintended consequence of driving China to develop other routes of importing technology and will make them faster over the long term as they build support technologies from the ground-up locally.”

Deborah Perry Piscione, co-founder and chief executive officer at Work3 Institute, also notes that China is making positive progress as competition ramps up.

“The rest of the world isn’t sitting still. China, for instance, is pouring seemingly unlimited resources into AI.

 

“Companies such as Baidu are making significant advancements, narrowing the gap with the U.S. and data access is a key consideration in the future of LLM success, as some countries, like China, may have advantages in data collection and usage due to different privacy laws, potentially providing more comprehensive training datasets for their LLMs.”

The Bottom Line

America may still be at the top of the AI development mountain, but the gap between it and other countries is thinning. Key competitors like China are well-positioned to eat away at that position and become a more potent force in the market.

It is also a reminder to Western markets that there is a wealth of AI tools out there that we are simply not exposed to.