!["Actually, I think [China is] closer to the US frontier models than maybe we thought one or two years ago. Maybe they’re only a matter of months behind at this point." - Demis Hassabis - DeepMind co-founder, CEO](https://i0.wp.com/globaladvisors.biz/wp-content/uploads/2026/01/20260119_05h01_GlobalAdvisors_Marketing_Quote_DemisHassabis_MW.png?fit=600%2C600&ssl=1 "20260119_05h01_GlobalAdvisors_Marketing_Quote_DemisHassabis_MW")
“Actually, I think [China is] closer to the US frontier models than maybe we thought one or two years ago. Maybe they’re only a matter of months behind at this point.” – Demis Hassabis – DeepMind co-founder, CEO
Context of the Quote
In a CNBC Original podcast, The Tech Download, aired on 6 January 2026, Demis Hassabis, co-founder and CEO of Google DeepMind, offered a candid assessment of China’s AI capabilities. He stated that Chinese AI models are now just a matter of months behind leading US frontier models, a significant narrowing from perceptions one or two years prior1,3,5. Hassabis highlighted models from Chinese firms like DeepSeek, Alibaba, and Zhipu AI, which have delivered strong benchmark performances despite US chip export restrictions1,3,5.
However, he tempered optimism by questioning China’s capacity for true innovation, noting they have yet to produce breakthroughs like the transformer architecture that powers modern generative AI. ‘Inventing something is 100 times harder than replicating it,’ he emphasised, pointing to cultural and mindset challenges in fostering exploratory research1,4,5. This interview underscores ongoing US-China AI competition amid geopolitical tensions, including bans on advanced Nvidia chips, though approvals for models like the H200 offer limited relief2,5.
Who is Demis Hassabis?
Demis Hassabis is a British AI researcher, entrepreneur, and neuroscientist whose career bridges neuroscience, gaming, and artificial intelligence. Born in 1976 in London to a Greek Cypriot father and Chinese Singaporean mother, he displayed prodigious talent early, winning the Eurovision Young Musicians contest at age 13 and becoming a chess master by 131,4.
Hassabis co-founded DeepMind in 2010 with the audacious goal of achieving artificial general intelligence (AGI). His breakthrough came with AlphaGo in 2016, which defeated world Go champion Lee Sedol, demonstrating deep reinforcement learning’s power1,4. Google acquired DeepMind in 2014 for £400 million, and Hassabis now leads as CEO, overseeing models like Gemini, which recently topped AI benchmarks3,4.
In 2024, he shared the Nobel Prize in Chemistry with John Jumper and David Baker for AlphaFold2, which predicts protein structures with unprecedented accuracy, revolutionising biology1,4. Hassabis predicts AGI within 5-10 years, down from his initial 20-year estimate, and regrets Google’s slower commercialisation of innovations like the transformer and AlphaGo despite inventing ‘90% of the technology everyone uses today’1,4. DeepMind operates like a ‘modern-day Bell Labs,’ prioritising fundamental research5.
Leading Theorists and the Subject Matter: The AI Frontier and Innovation Race
The quote touches on frontier AI models – state-of-the-art large language models (LLMs) pushing performance limits – and the distinction between replication and invention. Key theorists shaping this field include:
- Geoffrey Hinton, Yann LeCun, and Yoshua Bengio (‘Godfathers of AI’): Pioneered deep learning. Hinton, at Google (emeritus), advanced backpropagation and neural networks. LeCun (Meta) developed convolutional networks for vision. Bengio (Mila) focused on sequence modelling. Their work underpins transformers1,5.
- Ilya Sutskever: OpenAI co-founder, key in GPT series and reinforcement learning from human feedback (RLHF). Left to found Safe Superintelligence Inc., emphasising AGI safety3.
- Andrej Karpathy: Ex-OpenAI/Tesla, popularised transformers via tutorials; now at his own venture5.
- The Transformer Architects: Vaswani et al. (Google, 2017) introduced the transformer in ‘Attention is All You Need,’ enabling parallel training and scaling laws that birthed ChatGPT and Gemini. Hassabis notes China’s lack of equivalents1,4,5.
China’s progress, via firms like DeepSeek (cost-efficient models on lesser chips) and giants Alibaba/Baidu/Tencent, shows engineering prowess but lags in paradigm shifts2,3,5. US leads in compute (Nvidia GPUs) and innovation ecosystems, though restrictions may spur domestic chips like Huawei’s2,3. Hassabis’ view challenges US underestimation, aligning with Nvidia’s Jensen Huang: America is ‘not far ahead’5.
This backdrop highlights AI’s dual nature: rapid catch-up via scaling compute/data, versus elusive invention requiring bold theory1,2.
References
2. https://intellectia.ai/news/stock/google-deepmind-ceo-claims-chinas-ai-is-just-months-behind
4. https://biz.chosun.com/en/en-it/2026/01/16/IQH4RV54VVGJVGTSYHWSARHOEU/
6. https://brief.bismarckanalysis.com/s/ai-2026
