DeepSeek: AI’s new challenger disrupting Big Tech

While competition fuels innovation, concerns over privacy, security, and intellectual property remain critical issues. Can open-source AI level the playing field, or will tech giants continue to control the future of artificial intelligence? 

Dr Jeff Nijsse and Dr James Kang, senior lecturers at RMIT Vietnam, have weighed in on these issues, highlighting the opportunities and challenges posed by DeepSeek and the broader AI landscape.

A new wave of AI competition

AI is evolving at an unprecedented pace, with major advancements coming from both Big Tech giants and emerging startups. The release of DeepSeek R1, an open-source AI model, has introduced a new layer of competition, allowing smaller players to challenge the traditional powerhouses.

However, the industry is still largely controlled by companies with access to cutting-edge GPUs from NVIDIA and vast amounts of proprietary data. Dr James Kang, Senior Lecturer in Computer Science at RMIT Vietnam, highlights this imbalance.

“Big Tech has a significant advantage because of their financial resources and access to high-performance computing hardware. But the emergence of open-source models like DeepSeek R1 proves that smaller AI firms can still drive innovation by focusing on niche applications and efficiency,” Dr Kang said.

Meanwhile, Dr Jeff Nijsse, Senior Lecturer in Software Engineering at RMIT Vietnam, believes that open-source AI will ultimately shape the future.

“All it takes is one powerful open-source model to inspire others. Startups can build on existing work and innovate, even with limited resources. The dominance of expensive, high-end GPUs won’t last forever, and as older hardware becomes available, AI development will become more accessible,” Dr Nijsse said. 

DeepSeek is not alone in this open-source revolution. Dr Nijsse points out that other open-source models like Llama 3 and Qwen2.5-72B are also large-scale AI models trained using reinforcement learning. However, unlike DeepSeek R1, these models do not use the mixture-of-experts (MoE) architecture, a key feature that contributes to DeepSeek’s efficiency. Both models, along with DeepSeek, can be accessed on Hugging Face, a platform that enables developers to download and self-host AI models.

Privacy concerns and the challenge of trust

Despite its potential, DeepSeek faces hurdles in global adoption, particularly due to privacy concerns surrounding Chinese AI models. Many users fear that sensitive data could be exposed, leading to a reluctance in adopting these technologies.

“Shortly after its launch, privacy concerns pushed many users to self-host DeepSeek models,” Dr Nijsse said. “This is why platforms like Hugging Face are so popular – they allow developers to run AI locally, avoiding potential security risks.” 

This concern is not unfounded. Perplexity, a San Francisco-based AI startup, responded quickly by ensuring that its use of DeepSeek is restricted to data centres in the US and Europe, with CEO Aravind Srinivas publicly stating: “All DeepSeek usage in Perplexity is through models hosted in data centres in the USA and Europe. DeepSeek is open-source. None of your data goes to China.”

Additionally, Dr Nijsse raises another key concern regarding DeepSeek’s origins. 

“OpenAI claims it has evidence that DeepSeek distilled OpenAI’s models for their benefit. Distillation reduces the overall size of a model while increasing efficiency, but it could be a violation of OpenAI’s terms of service,” he said.

This accusation, combined with questions over how DeepSeek acquired high-performance GPUs despite NVIDIA’s strict export controls, paint a complex picture of AI’s competitive landscape. 

The future: Balancing innovation and risk

While AI promises increased productivity and automation, it also presents challenges in regulation, ethical considerations, and workforce transformation. Dr Kang compares this shift to previous technological revolutions.

“AI will inevitably replace certain jobs, but it will also create new opportunities. The key is to focus on reskilling and upskilling the workforce so that people can transition into new roles that require critical thinking and decision-making,” he said.

At the same time, Dr Nijsse emphasises the long-term affordability of AI models. 

“The cost of AI models in terms of capital expenditure is rising as more GPUs are on order, but at the same time, the cost per token is decreasing. This means that users are getting better value over time,” he said.

Despite these challenges, both experts agree that competition in AI is ultimately beneficial for innovation and accessibility. As open-source AI continues to evolve, the future may see a more democratised AI landscape, where technology is shaped not just by a few tech giants but by a broader, more diverse community of innovators.

This shift means that start-ups in Vietnam now have the opportunity to leverage open-source technologies to develop specialised AI applications tailored to niche markets with a significantly lower investment compared to larger tech companies. Such an approach can unlock new avenues for small enterprises. With the backing of the Vietnamese government, the AI industry in Vietnam is poised for significant growth. By homing in on specialised markets, Vietnam has the potential to emerge as a leading player in the region.

Story: Dr James Kang and Dr Jeff Nijsse 

Masthead image: Putilov Denis - stock.adobe.com

Thumbnail image: PixieMe - stock.adobe.com