Algo4hi Race: Chasing AI Supremacy

DeepSeek R1: A Silent Surge from China

The AI world buzzed recently, not with a grand announcement, but with the quiet, almost stealthy, launch of DeepSeek R1, an upgraded version of Chinese AI startup DeepSeek's reasoning model. Released without a formal fanfare, DeepSeek R1 has nonetheless sent ripples across the globe, intensifying its direct competition with established giants like OpenAI and Meta.

What truly surprised the industry was DeepSeek R1's rapid development cycle and its remarkable cost-effectiveness. Reports indicate that DeepSeek has managed to achieve top-tier performance in complex reasoning tasks, including mathematics, programming, and general logic, with overall capabilities approaching those of leading international models like OpenAI's O3 and Gemini 2.5 Pro. Crucially, it has done so at a fraction of the operating cost. This efficiency, reportedly achieved through innovative architectures like a Mixture-of-Experts (MoE) design, demonstrates China's formidable capability to innovate under constraints.

This move by DeepSeek is a testament to the strategic shifts underway within China's tech giants. Facing tightening U.S. curbs on key semiconductors, companies like Alibaba and Baidu have publicly outlined their multi-pronged approach to maintaining their competitive edge. These strategies include:

• Chip Stockpiling: Building reserves of essential semiconductors to cushion against supply disruptions.

• Enhanced AI Model Efficiency: Investing heavily in optimizing AI model architectures and training methodologies to achieve more with less computational power. DeepSeek R1's cost-efficiency is a prime example of this.

• Homegrown Semiconductors: A long-term, ambitious push to develop and utilize domestically produced AI chips, reducing reliance on foreign imports and fostering greater technological self-sufficiency. This national initiative, backed by significant government investment, aims to close the technological gap with Western counterparts.

These proactive measures highlight China's determination to not only participate but to lead in the global AI race, even in the face of significant geopolitical headwinds.

India's Sarvam AI: Navigating a Distinct Path

Closer to home, India's homegrown AI narrative saw a significant development with the recent release of Sarvam-M by Sarvam AI. Positioned as a versatile, locally relevant large language model, Sarvam-M is a 24-billion-parameter hybrid open-weights model built on Mistral Small, designed to excel in Indian languages, mathematics, and programming tasks.

However, the initial response to Sarvam-M has been notably mixed, prompting crucial discussions within the Indian AI community. While praised for its focus on Indian languages and its potential for local use cases, the model received a comparatively lukewarm response in terms of downloads – less than 720 within three days on Hugging Face, drawing criticism from some investors who questioned its immediate relevance compared to globally viral open-source models.

Despite the initial skepticism, Sarvam AI remains committed to its vision of building foundational models for India, emphasizing that Sarvam-M represents an important stepping stone. The company has highlighted the model's performance in specific Indian language benchmarks and its ability to handle complex problems like JEE Advanced questions in Hindi. Other Indian efforts, like the government-backed BharatGen's Param-1 model, have also faced similar challenges in gaining widespread traction.

1. Deep Investment in Foundational Research and Compute Infrastructure: China's aggressive, state-backed investments in AI research and securing high-end GPUs have been instrumental. While India has recently scaled up its IndiaAI Mission, increasing allocations and securing more GPUs (e.g., 15,000 high-end GPUs now available, up from 10,000 previously, and a commitment to secure over 18,000), sustained and substantial investment in compute capacity remains paramount. Subsidized computing power, as planned by the government, is a positive step.

2. Focus on Domestic Chip Capabilities (Long-Term): While challenging, China's push for homegrown semiconductors demonstrates a commitment to technological sovereignty. India's current focus on utilizing available GPUs is crucial, but a long-term vision for fostering domestic chip design and manufacturing capabilities, even for specific AI accelerators, could be a game-changer.

3. Efficiency and Optimization: DeepSeek R1's cost-effectiveness highlights that sheer scale isn't the only metric. India's AI startups must prioritize efficient model architectures and training techniques to maximize impact from available resources.

4. Strategic Data Collection: Developing large volumes of high-quality, India-specific datasets is critical for training truly effective Indic language models and addressing unique Indian challenges. Initiatives like the IndiaAI Dataset Platform (AIKosha) are essential but need rapid scaling and widespread adoption.

5. Cultivating Top-Tier Talent: While India has a vast talent pool, attracting, nurturing, and retaining cutting-edge AI research talent is crucial. This involves fostering a vibrant research ecosystem in universities and providing opportunities for fundamental research.

Start learning AI in 2025

Keeping up with AI is hard – we get it!

That’s why over 1M professionals read Superhuman AI to stay ahead.

• Get daily AI news, tools, and tutorials

• Learn new AI skills you can use at work in 3 mins a day

• Become 10X more productive

Sign up and start mastering AI