Anushruti Singh
It’s late January, and it’s one disruption everyone’s talking about. In an industry where billion-dollar budgets and corporate giants dominate, DeepSeek, a Chinese AI startup, has achieved the unthinkable—launching a high-performing AI model with just $5.6 million. For comparison, tech titans like OpenAI and Google typically spend over $100 million to develop similarly advanced systems.
And this has rocked the global tech stocks. DeepSeek, founded by Liang Wengfeng has challenged the existing dynamics of AI innovation and investment. It’s not just another AI startup; it’s a signal that the rules of the game are changing.
Innovation without breaking the bank
DeepSeek’s standout achievement is that its cost-efficient. Its architecture employs innovative training techniques and optimised resource management, demonstrating that you don’t need billions to create cutting-edge AI.
Furthermore, it’s available free for download on the apple app store.
What do we know about DeepSeek
DeepSeek has developed two main families of AI models: ‘DeepSeek-V3’ and ‘DeepSeek R1’. The V3 model is a large language model that uses a Mixture-of-Experts (MOE) architecture.
This approach combines multiple smaller models to work together, resulting in high performance while using significantly fewer computing resources compared to other large models. The V3 model has 671 billion parameters in total, with 37 billion active at any given time.
It also incorporates innovative techniques like Multi-Head Latent Attention (MHLA), which reduces memory usage, and mixed-precision training using FP8 computation, which improves efficiency.
The second model, ‘DeepSeek R1’, builds on the V3 foundation but uses Reinforcement Learning (RL) and other techniques to significantly improve reasoning capabilities. The R1 model has been particularly impressive, performing competitively against OpenAI’s models in reasoning tasks.
Bruce Keith, Co-Founder and CEO of InvestorAi, says, “DeepSeek R1 has definitely challenged the dominance of a few players in the models and data ecosystem—OpenAI, Google, and Meta will feel it the most. The fact that this platform was created with under $6 million investments has shaken tech CEOs globally, highlighting that game-changing innovations don’t necessarily need billion-dollar budgets.”
Lessons for Indian startups
DeepSeek’s breakthrough has sparked conversations about its implications for emerging markets like India. Narendra Bhandari, General Partner at Seafund, believes this moment underscores an opportunity for Indian startups.
“Indian startups can aspire to build relevant models at much lower costs. This is the right time to focus on the India opportunity in building AI models because DeepSeek has shown that innovation can be done at a much lower cost. The government should fund capital buildouts, attract global talent to build new models, and balance research funding across universities and labs. And focus on DeepTech investors to drive and manage the innovation pace,” he says.
The Indian government, with its focus on AI adoption and initiatives like the Digital India program, is already laying the groundwork. The question is whether this momentum can be accelerated.
Ripple effect on AI hardware
Despite DeepSeek’s cost efficiency, the demand for high-powered GPUs—dominated by Nvidia and AMD—remains a critical factor in AI innovation. Experts argue that while DeepSeek democratises AI, the hardware dependency keeps costs high.
As Bhaskar Majumdar, Managing Partner at Unicorn India Ventures, points out, “DeepSeek reminds us that accepting the status quo is the wrong strategy. All of us will witness significantly power-efficient, high-performance, solution-centric chips in the next few years that will challenge the likes of Nvidia. This disruption will benefit India’s solutions community as well.”
“This was the logic of us investing in Netrasemi as the company believes in this disruption and betting on this change,” he further adds.
Geopolitical underpinning
DeepSeek’s rise is not just an economic story but a geopolitical one. The model emerged partly in response to U.S. export restrictions on advanced chips to China. Ryan Cox, Global Head of Artificial Intelligence at Synechron, notes that this reflects China’s strategic positioning in AI.
“DeepSeek has disrupted our understanding of AI economics by achieving what seemed impossible. However, this disruption is also a geopolitical response, raising questions about its long-term AI leadership versus strategic positioning,” Cox explains.
According to him, this disruption highlights the need for smarter, not bigger AI investments.
“What sets DeepSeek apart is its efficient architecture, innovative training techniques, and optimised resource management. This opens the door for businesses of all sizes to adopt advanced AI without the exorbitant costs. 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,” he adds.
Challenge of governance
While DeepSeek’s open-weight models democratise AI, they also introduce governance challenges. Open-weight systems lack built-in security certifications, placing the compliance burden on deploying organisations.
“The real challenge for enterprises isn’t just cost optimisation—it’s governance. Open-weight models demand robust validation frameworks to ensure performance, security, and ethical benchmarks,” adds Cox.
Contradiction over DeepSake’s $5 million claim
While DeepSake has taken the tech and AI world by storm, not everyone is convinced that it has achieved the impossible at such a low cost. An analysis by Bernstein challenges the circulated claim, that the Chinese company has built the model for just $5 million.
According to the report, this claim is misleading and doesn’t reflect the full picture.
The report states, “We believe that DeepSeek DID NOT “build OpenAI for USD 5M”; the models look fantastic, but we don’t think they are miracles; and the resulting Twitter-verse panic over the weekend seems overblown”.
The real cost of Chinese AI model, according to Bernstein
DeepSeek’s V3 model was trained using a cluster of 2,048 NVIDIA H800 GPUs over a period of two months, totalling approximately 2.7 million GPU hours for pre-training and 2.8 million GPU hours including post-training. While some estimates suggest that renting this computing power at $2 per GPU hour would amount to roughly $5 million, the Bernstein report highlights that this figure only covers GPU rental costs—not the research, experimentation, and infrastructure required to develop the model.
Additionally, the report suggests that DeepSeek’s R1 model likely required substantial additional resources, though these costs remain unquantified in the company’s official research paper.
Commenting on this, from a supply and demand perspective Keith of investor.ai opines that the GPU market which Nvidia dominates is still far away from hitting peak demand.
What’s next for AI?
As we head into 2025, the DeepSeek phenomenon serves as a reminder that AI innovation doesn’t require extravagant budgets—it requires smarter investments and governance. For India, this is a golden opportunity to lead the charge by focusing on efficient AI solutions and leveraging global talent.
On the other hand, technology mavens believe “DeepSeek’s ‘open source’ nature opens it up for exploration – by both adversaries and enthusiasts.
Chester Wisniewski, director and global field CTO, Sophos concern is that Like llama, it can be played with and largely have the guardrails removed. “This could lead to abuse by cybercriminals, although it’s important to note that running DeepSeek still requires far more resources than the average cybercriminal has,” he says.
“More pressing for companies, however, is that, due to its cost effectiveness, we are likely to see various products and companies adopt DeepSeek, which potentially carries significant privacy risks,” he adds.
Overall, as with any other AI model, it will be critical for companies to make a thorough risk assessment, which extends to any products and suppliers that may incorporate DeepSeek or any future LLM. With this, tech world needs to be certain they have the right expertise to make an informed decision.
