Commentary: As DeepSeek shakes up AI, what’s next for Malaysia’s data centre dreams?

Understanding the Costs

The initial cost of developing DeepSeek’s AI model is reported to be $6 million, significantly lower than that of its U.S. counterparts, which often run into tens of millions of dollars. However, this figure only represents the pre-training phase and not the total research and development (R&D) expenses. While $6 million is indeed a fraction of what U.S. companies typically spend, the overall R&D and hardware costs, estimated to be over $500 million, highlight that the actual investment remains substantial. This context is crucial because it shows that while DeepSeek has achieved cost efficiency, the overall expenditure in the AI race remains high.

The Nature of AI Advancement

Dario Amodei, CEO of Anthropic, offers a unique perspective on AI development: the traditional cycle of high initial costs followed by decreasing prices does not apply here. Instead, as innovations reduce costs, companies often reinvest these savings into developing more advanced models. This phenomenon, where the "scaling curve shifts," means that the quest for better AI is incessant. The value of superior performance in AI models, capable of solving complex tasks more effectively, often outweighs the desire to cut costs. As a result, the AI industry is characterized by continuous investment, with companies striving to push the boundaries of what their models can achieve.

Efficiency and Investment

The argument that cost efficiency in AI will lead to reduced infrastructure investments is flawed. According to Amodei, when training costs decrease due to technological advancements, companies typically choose to invest more in training better models rather than reducing their overall expenditure. This is because the benefits of higher-performance AI—from improved decision-making to enhanced user experiences—far outweigh any potential cost savings. Consequently, even as companies like DeepSeek achieve greater efficiency, the trend suggests that they will continue to invest heavily in their AI infrastructure to maintain a competitive edge.

Expanding Accessibility

The efficiency gains achieved by DeepSeek could have a profound impact on the accessibility of AI technology. By lowering the barrier to entry, more companies, especially smaller ones, might be able to develop or integrate AI applications into their products and services. This potential democratization of AI could lead to a proliferation of AI-driven innovations across various industries, ranging from healthcare to education. The success of this vision, however, largely depends on how effectively DeepSeek and similar models can be integrated into everyday devices and how consumers embrace these technologies.

The Great Catch-Up Begins

DeepSeek’s emergence marks a significant shift in the global AI landscape. By overtaking Alibaba and closing the gap with OpenAI, DeepSeek has demonstrated that the dynamics of the AI race are evolving. This "great catch-up" is not just about achieving parity; it signals the beginning of a new phase where competition is intensifying. As more companies achieve similar breakthroughs, the pace of innovation is likely to accelerate, leading to a scenario where the leader in AI technology could change rapidly.

The Future of AI and Infrastructure

The bets placed on AI-driven infrastructure, such as Malaysia’s ambitious investments, will soon face a critical test. The combination of falling costs, increased accessibility, and relentless innovation creates both opportunities and challenges. Policymakers and business leaders will need to balance the pursuit of technological advancement with considerations of equity, ethics, and sustainability. The coming years will reveal whether these investments pay off and which players in the AI race manage to maintain their momentum.

James Chai, author and political analyst, emphasizes the significance of this era in AI development, highlighting the potential for transformative change and the need for strategic foresight.