Font size:
Print
DeepSeek and the New AI Race
Global AI’s ‘DeepSeek Moment’: Impact and Implications
Context:
The Global Technology Summit in New Delhi (April 2025) focused on Sambhavna (possibility), reflecting a cautiously optimistic stance on how global disruption in tech, especially AI, can present exponential opportunities if political will is aligned.
More on News
- The event, hosted by India’s Ministry of External Affairs, reflected a cautious optimism about the exponential opportunities that technological disruption can bring-provided there is political will to harness them.
- This sentiment set the stage for a seismic shift in the global technology landscape, as China’s DeepSeek AI model emerged as a formidable challenger to US dominance in artificial intelligence.
The DeepSeek Disruption
- US President Donald Trump announced the $500 billion “Stargate project” to secure America’s AI future, China responded with DeepSeek-a startup with a “hacker spirit” that upended prevailing norms in AI compute capacity and algorithmic output.
- DeepSeek’s approach promised more efficient, affordable, and adaptable AI development, directly challenging the US’s competitive “moat” built on export controls and the CHIPS and Science Act.
- The launch of DeepSeek’s R1 model on Trump’s inauguration day was quickly dubbed AI’s “Sputnik moment,” drawing parallels to the Soviet Union’s 1957 satellite launch that jolted the US into the space race.
- The impact was immediate: US tech stocks, including Nvidia, OpenAI, Meta, Microsoft, and Google, saw the Nasdaq Composite lose over $1 trillion in a single day. DeepSeek’s open-source, cost-effective model stood in stark contrast to OpenAI’s resource-intensive, closed development strategy.
Technical Innovation Born from Constraint
- DeepSeek’s rise was not due to breakthroughs in China’s semiconductor sector but rather innovation under duress. US export controls on Nvidia’s H100 chips forced the startup to optimise algorithms and model architectures.
- Reports suggest DeepSeek either stockpiled chips before the ban or acquired them via intermediaries, but its real edge lay in ingenuity.
- The R1 model, trained using large-scale reinforcement learning and drawing on Meta’s open-source Llama, reportedly cost less than $6 million to train-compared to over $100 million for OpenAI’s GPT-4.
- DeepSeek’s “mixture-of-experts” architecture mimics the human brain by activating specialised neural networks for specific tasks, drastically reducing memory and compute requirements.
- This efficiency enabled DeepSeek to offer output at $2.19 per million tokens, compared to $60 for OpenAI’s o1 model-a nearly 30-fold difference.
Geopolitical and Economic Implications
- DeepSeek’s breakthrough is a direct response to US trade restrictions, which inadvertently spurred Chinese innovation. The model’s open-source nature has global implications:
- AI Democratisation: DeepSeek’s efficiency could make advanced AI accessible to more countries, especially in the Global South. Smaller models can run on everyday devices without reliance on cloud providers, reducing demand for high-end GPUs.
- US Policy Dilemma: The US faces a choice between tightening export controls (risking further Chinese innovation) or relaxing them to meet global demand for AI hardware. The chips ban may have backfired, prompting China to accelerate its own semiconductor R&D.
- Competition and Collaboration: Open-source models commoditise foundational AI layers, shifting value to application builders and reducing barriers for new entrants. This could erode the dominance of US Big Tech and foster a more multipolar AI ecosystem.
Risks and Regulatory Challenges
- The open-source revolution in AI brings both promise and peril:
- Security Concerns: Open-sourcing advanced models increase transparency and community oversight but also expose vulnerabilities and enable misuse. Several countries have already restricted DeepSeek’s application layer in sensitive sectors.
- National Interests: The Chinese government is likely to exert control over DeepSeek’s international release, balancing openness with security and strategic interests.
- Global Standards: The US must craft nuanced policies that balance economic interests, security, and the need for international cooperation-moving beyond a zero-sum approach.
Rethinking the “AI Race”
- The AI race must shift from a “nuclear arms race” mindset to a “global public good” approach.
- Unlike stockpiled nuclear weapons, AI actively shapes societies and economies.
- Success will belong to those who:
- Invest in skills and process innovation
- Embrace a new “technological quotient” that balances technology, capability-building, and human capital.
The Requirements for India: Strong Foundations and Strategic Partnerships in AI
- India’s Digital and Economic Foundations: As of 2024, India had:
- 863 million Internet users
- US$568 billion in monthly UPI transactions
- Over 100 unicorn startups
- 49% share in global real-time payments
- India launched the IndiaAI Mission in March 2024, aiming to create a comprehensive AI ecosystem built on seven pillars: computing infrastructure, R&D, data access, impactful applications, skills development, financing, and trust and safety.
- India was excluded from the list of 18 countries granted unfettered access to US high-end GPU chips under the ‘Framework for Artificial Intelligence Diffusion’ by the Biden administration, effective 15 May 2025.
- In response, India: Empanelled 18,693 GPUs under a Common Computing Facility, including 12,896 Nvidia H100 chips. Made 14,000 GPUs accessible by early 2025. Plans to acquire 15,000 more chips to expand computing capacity.
- Despite robust digital infrastructure, Indian AI stakeholders face challenges. Private sector R&D investment in India lags behind peers in the US, China, Japan, and South Korea.
- India’s GERD is below 1% of GDP, far behind countries like China (2.68% of GDP in 2024).
- India must deepen cooperation via frameworks like:
- TRUST (Transforming Relations Utilising Strategic Technologies)
- Indo-Pacific Economic Framework
- Key objectives:
- Co-develop frontier tech
- Secure inclusion in the Tier-1 countries in the AI diffusion framework
- De-risk potential misalignments with US investments and operations in India
