DeepSeek, a Chinese AI startup, has shaken up the tech world and has proved that building a high-quality AI model doesn’t have to cost hundreds of millions of dollars.
While companies like OpenAI invest over $100 million in developing AI models, DeepSeek has reportedly built a GPT-4-like model for just $5.6 million. This remarkable achievement is attributed to its efficient design, innovative training methods, and strategic resource management.
So far, most Global Capability Centres (GCCs) in India have relied on OpenAI for AI-driven innovation. For instance, Lowe’s partnered with OpenAI on its AI initiatives even before the launch of ChatGPT. However, DeepSeek’s cost-efficient approach raises an important question: how will it impact GCCs in India?
As hubs of cost arbitrage, GCCs may find DeepSeek an attractive alternative for AI innovation, offering potential savings and greater flexibility. Yet, despite the promising opportunities, many GCCs are adopting a cautious approach, preferring to observe how DeepSeek evolves before making significant commitments.
What GCCs Think about DeepSeek
In a conversation with AIM, Ryan Cox, Synechron’s global head of artificial intelligence, said that he sees DeepSeek’s feat as a major shift in how AI is developed and deployed.
“DeepSeek has disrupted our understanding of AI economics by achieving what seemed impossible: creating a high-performing AI model for $5.6 million, compared to the $100 million plus budgets required by industry leaders like OpenAI.”
For Cox, this achievement partly reflects a geopolitical response to US export restrictions, raising questions about its long-term AI leadership versus strategic positioning.
He explained that as AI continues to grow, the demand for powerful computing resources will continue to increase. Although DeepSeek’s model is cost-effective, it still runs on GPUs from big suppliers like NVIDIA and AMD.
According to Cox, efficiency makes DeepSeek’s approach special. It shows that businesses of all sizes can now access advanced AI without spending a fortune.
“Beyond cost savings, this democratisation of AI introduces new competition, spurring better results and innovation for end users worldwide. It also underscores the importance of leveraging global AI talent,” Cox added.
At Synechron, their testing of DeepSeek’s 32-billion-parameter model showed promising results but also highlighted the need for careful bias detection and compliance measures before widespread use.
Looking ahead, Cox believes that successful AI adoption in 2025 will require balancing innovation and responsible management. He advises companies to focus on three key areas: clear validation processes, strong compliance systems, and flexible technology that can keep up with rapid changes.
Meanwhile, Raghavendra Vaidya, managing director and CEO of Daimler Truck Innovation Center India (DTICI), takes a more cautious stance. “I think it’s too early to comment on DeepSeek. I think it has created havoc in the capital markets,” he said.
Vaidya pointed out that while AI models are incredibly complex and expensive to build, the core algorithms have been publicly known for years, owing to companies like Google.
However, creating an AI model is just one part of the equation; it doesn’t automatically bring business value. “I don’t think everybody should start building their LLMs. It doesn’t make any sense. There will be a few big tech players who will build those LLMs.”
“Building an LLM is like building a capability, and a capability doesn’t deliver business value. You must actualise that capability,” Vaidya further said.
Why is Governance Required?
DeepSeek’s breakthrough may change the AI landscape, but its real impact will depend on how businesses use and govern these models in the coming years.
However, there are challenges beyond cost. Open-weight AI models like DeepSeek’s allow for customisation but also require strong governance. “The real challenge for enterprises isn’t just cost optimisation, it’s governance,” Cox warned.
Since some AI models are developed with certain restrictions, especially in domains like politics and culture, businesses using them must carefully validate their accuracy, fairness, and security.
“CIOs and technology leaders must establish rigorous governance and validation frameworks to qualitatively ensure their GenAI solutions meet performance, security, and ethical benchmarks,” he added.
As per Salman Waris, managing partner at TechLegis Advocates and Solicitors, the Indian government should launch a high-level investigation into the extent of data mining and web scraping carried out by DeepSeek AI. He stressed that its model raises serious security concerns, particularly since India recorded the highest number of app downloads in the first 24 hours of its launch.
Moreover, from a regulatory standpoint, the government should consider adding specific provisions in the proposed Digital Personal Data Protection (DPDP) Rules to safeguard data privacy and security, especially concerning the use of AI bots like DeepSeek.
Waris also pointed out the relevance of these concerns for Global Capability Centres (GCCs). GCCs handle sensitive data, making them potentially vulnerable to major data breaches if they implement DeepSeek AI. He stressed the importance of investing in stronger data protection, IT security, and regulatory compliance to prevent such risks.
“It might only be a matter of time before GCCs using DeepSeek AI face serious data breach incidents. They need to prioritise data security and compliance to safeguard sensitive information,” Waris concluded.
