There is No Wall for China

The rise of the AI dragon signals not just a shift in power but a transformation in how artificial intelligence is developed and deployed globally. The message is clear for a world watching with bated breath: the race for AGI is heating up, and for China, there is no (great) wall.

Look at DeepSeek’s R1-Lite Preview, for instance. Hailed as a reasoning model rivalling OpenAI’s o1, it has achieved real-time chain-of-thought transparency.?

And here’s the kicker: Within just two months of o1’s release, Chinese researchers not only caught up but surpassed it, dominating benchmarks like AIME and MATH. DeepSeek’s rise as a leader in reasoning isn’t just impressive—it’s a wake-up call you can’t afford to ignore.

Moreover, StepFun’s Step-2, a trillion-parameter mixture of experts (MoE) model ranking 5th globally on LiveBench, showcased innovations in efficiency and long-context tasks. Not to be left behind, Yi-Lightning, developed by the startup, 01.AI, demonstrated how careful engineering can replace brute computational force, achieving competitive results with only $3 million in training costs—a fraction of what Western labs spend. (OpenAI’s GPT-4 costs about $80-100 million to develop).

“It is not that you must spend a billion dollars to train a great model. Detailed engineering and resource prioritisation is key,” said Kai-Fu Lee, founder of 01.AI.?

“The top open source models are Chinese, and they are ahead because they focus on building, not debating AI risks,” said Daniel Jeffries, chief technology evangelist at Pachyderm.?

China’s success is underpinned by its focus on efficiency and resource optimisation. With limited access to advanced GPUs due to export restrictions, Chinese researchers have innovated ways to reduce computational demands and prioritise resource allocation.?

“When we only have 2,000 GPUs, the team figures out how to use it,” said Lee. “Necessity is the mother of innovation.”?

He further highlighted how his company transformed computational bottlenecks into memory-driven tasks, achieving inference costs as low as 10 cents per million tokens, a mere fraction of their competitors. “Our inference cost is one-thirtieth of what comparable models charge,” said Lee.?

The rapid rise of Chinese AI extends beyond its borders, with companies like MiniMax, ByteDance, Tencent, Alibaba, and Huawei targeting global markets.?

MiniMax’s Talkie AI app, for instance, boasts 11 million active users, half of whom are in the US.?

At the Wuzhen Summit 2024, analysts noted that as many as 103 Chinese AI companies were expanding internationally, focusing on Southeast Asia, the Middle East, and Africa, where the barriers to entry were lower than the Western markets.?

ByteDance has launched consumer-focused AI tools like Gauth for education and Coze for interactive bot platforms, while Huawei’s Galaxy AI initiative supports digital transformation in North Africa.?

AI Video Models?

Models like Kling and Hailuo have outpaced Western competitors like Runway in speed and sophistication, marking a shift in leadership within this emerging domain. This success is mirrored in advancements in multimodal AI, where models like LLaVA-o1 rival OpenAI’s vision-language models by implementing structured reasoning techniques that break down tasks into manageable stages.

The Silent AI Leader?

Meanwhile, Baidu is quietly outpacing the West with groundbreaking innovations like the hallucination-mitigating iRAG, no-code multi-agent tool Miaoda, and Xiaodu AI Glasses, solidifying its leadership in AI applications despite geopolitical sanctions.

Yet, this progress is not without challenges. The US has tightened restrictions on the export of high-performance GPUs, limiting China’s access to essential hardware. Plus, Chinese AI developers must navigate stringent government regulations mandating that models embody “core socialist values” and avoid politically sensitive topics. These obstacles, however, have only spurred further innovation, with researchers designing new architectures and cost-efficient training methods to maintain their competitive edge.

China’s strategy of combining cost efficiency with high performance is reshaping the global AI landscape. While US companies remain dominant in foundational AI research, China’s focus on application-driven innovation is enabling it to create scalable and accessible AI solutions. Open source initiatives like Alibaba’s Qwen 2.5 and DeepSeek’s R1 are democratising access to advanced AI, ensuring that innovation rises above the borders.?

Meanwhile, Shenzhen’s rapid hardware iteration cycles are giving Chinese firms a unique advantage in robotics and AI hardware development, driving advancements in physical intelligence and autonomous systems.

“Shenzhen iterates hardware at the speed of software,” said Jensen Huang, CEO of NVIDIA.?

For more insights on AI, stay tuned to AIM.

Top Stories of the Week >>?

AI Scaling Laws are Crumbling Under Token Pressure?

AI scaling laws are crumbling—or are they? At Microsoft Ignite 2024, CEO Satya Nadella and industry leaders spoke about the future of AI computation, from tokens per watt efficiency to test-time scaling innovations. With giants like NVIDIA and Microsoft reshaping data centres into “intelligence factories,” the race to scale AI is on, and it’s redefining how we think about compute as the new currency.

Dive deeper here.

Perplexity’s Shopping Assistant is a Killer

Perplexity is killing the shopping experience—and not in the way you'd think. By ditching ads, spam, and the endless search chaos, it’s making shopping smarter, faster, and surprisingly enjoyable. Who knew AI could make e-commerce feel like a stroll in the park? Read more here.

People & Tech >>?

Tata Communications is Building a Self-Reliant AI Cloud in India

India is gearing up to make its mark on the global AI stage as Tata Communications unveils its sovereign AI Cloud, a game-changing platform designed to empower enterprises, startups, and the government with end-to-end AI capabilities. From its cutting-edge AI Studio to scalable GPU resources and a focus on foundational models, Tata is leading the charge to transform businesses into AI Super Factories while championing data sovereignty and privacy.?

Read the full story here.

GenAI Enterprise Revolution Slower Than it Looks: Lingaro CEO

Lingaro Group CEO Sam Mantle visited Bengaluru last week, and AIM had the opportunity to catch up with him, delving into his vision for Lingaro’s future in India. The company first set foot in India a few years ago. Now, it is here again with ambitious plans to increase its revenue by 30% year over year and double its workforce to 400 employees. Read more here.?

WTF is Nikhil Kamath Doing with Young Entrepreneurs?

What’s Nikhil Kamath up to with India’s brightest young minds? Through his ‘Innovators Under 25’ initiative, the Zerodha co-founder is backing ingenious startups with up to INR 20 lakh in non-dilutive grants—no equity, just a mission to empower! From AI-powered toys and DNA data storage to life-saving health tech, these startups are tackling real-world challenges head-on. But what’s driving his bold move?

Find out here.

AI Events >>?

[Webinar Alert] Building Blocks for Bharat: AI Drivers for Next-Gen Applications

Building for Bharat isn’t optional—it’s inevitable. Join the webinar ‘Building Blocks for Bharat: AI Drivers for Next-Gen Applications’ to uncover how AI is revolutionising accessibility and inclusivity in India’s digital landscape. From Indic language interfaces to voice-driven innovations, this session will equip you with the right tools and insights to lead the AI charge in Bharat.

Ready to shape the future? Register here.

?? Date: Thursday, December 12 ?? Time: 4:30 – 5:30 PM

The Rising 2025: India’s Premier Diversity & Inclusion Summit in Tech and AI

If you think diversity is just a checkbox, think again. The Rising 2025 is here to prove why it’s the driving force behind innovation in tech and AI. In its 7th edition, India’s premier DEI summit, The Rising, will go beyond surface-level talks to tackle the real challenges, strategies, and milestones shaping inclusivity in the industry.

With over 1,000 professionals and 50+ thought leaders converging at the J N Tata Auditorium on March 20-21, 2025, this isn’t just a conference—it’s a movement.

Don’t miss out. Secure your early-bird passes here.

?? Date: March 20-21, 2025 ?? Location: J N Tata Auditorium, Bengaluru

AIM Videos >>?

Has OpenAI Hit The Wall?

“There is no wall”—Sam Altman is convinced. But have LLMs really hit a dead-end? Or are we finally moving from “bigger is better” to “smarter is sharper”? From scaling debates to AGI deadlines, this reaction video spills it all—with a side of roast.?

AI Nuggets >>?

• NVIDIA has launched Hymba-1.5B-Base, a hybrid language model combining Transformer attention with state space models. It sets new benchmarks for efficiency and performance in small-scale LLMs under 2 billion parameters.
• NVIDIA’s Q3 revenue surged to a record $35.1 billion, fueled by soaring demand for Hopper and Blackwell GPUs, as the AI boom drives transformative adoption across industries, nations, and workflows.
• OpenAI has hired Darin Fisher, the creator of Google Chrome, to develop an AI browser that could revolutionise online interactions and challenge Chrome's dominance.?
• AWS has invested $4 billion in Anthropic, becoming its primary cloud partner to accelerate advanced AI systems development, optimise Trainium hardware, and deepen integration for secure and efficient AI deployments.
• Google DeepMind recently unveiled AlphaQubit, an AI-driven system that enhances quantum computing reliability by identifying and correcting errors, marking a breakthrough in scalable and practical quantum systems.
• Google’s Gemini-Exp-1121 has debuted at the top of LLM rankings alongside OpenAI’s GPT-4o, showcasing advancements in coding, reasoning, and visual processing.?
• Nokia has expanded its five-year deal with Microsoft Azure. It will provide advanced routers and switches to enhance data centre networks, supporting Azure’s shift to 400GE connectivity and scaling to meet surging global cloud demands.
• Snowflake has partnered with Anthropic to integrate the Claude 3.5 Sonnet model into its Cortex AI on AWS. This would enable secure, scalable AI solutions like chatbots and generative analytics, bridging the gap for enterprises to harness trusted AI capabilities.
• Runway has launched its new ‘Expand Video’ feature, enabling users to transform simple clips into cinematic masterpieces by extending video frames seamlessly.?
• Renault Nissan has set up a new Bengaluru centre under its RNTBCI 3.0 initiative, focusing on Software-Defined Vehicles and AI-driven innovations.