In earlier publications, EMIR Research warned of two main risks: excluding major AI players like China would fragment the AI landscape, and America’s AI dominance was far from assured. Both are now unfolding. DeepSeek’s ascent is forcing U.S. tech giants and investors to rethink how they allocate billions toward AI infrastructure—particularly GPUs and dedicated power. If DeepSeek can match GPT-4 performance with older GPUs and minimal overhead, the entire AI supply chain—especially Nvidia—faces a seismic shift.
DeepSeek: A Disruptor in AI Efficiency
China’s DeepSeek is upending the AGI race with minimal infrastructure and cost. While Google, Microsoft, and Meta invest billions in advanced GPUs, DeepSeek allegedly achieved similar results for only $6 million. According to its R1 white paper, DeepSeek repurposes older Nvidia GPUs—no longer freely exportable to China—and employs novel training methods that bypass resource-heavy U.S. approaches.
Among their most notable innovations:
1. Large-Scale Reinforcement Learning (RL): Traditional LLMs rely on heavily curated, supervised datasets. DeepSeek pioneered an almost “reinforcement learning from scratch” approach, allowing DeepSeek-R1-Zero to develop advanced reasoning skills—like self-verification and error correction—without extensive supervised fine-tuning.
2. Resource-Efficient Distillation: DeepSeek’s R1-Distill-Qwen-7B shows advanced reasoning can be compressed into smaller models without losing performance. This approach reduces training costs and runtime overhead, outperforming OpenAI’s o1-mini on MATH-500 and AIME benchmarks.
3. Reduced GPU Dependence: DeepSeek’s success jolts a market reliant on premium GPUs. By achieving top performance using older GPUs and novel optimizations, DeepSeek challenges the belief that cutting-edge AI demands costly infrastructure.
The implications have reverberated across global markets, prompting a sharp selloff in AI-related equities. Nvidia’s valuation plunged by over 16% in a single day; other tech companies involved in power generation and data-center hardware, like YTL Power in Malaysia, also faced investor panic. The question is no longer merely about performance or brand reputation—it’s about whether vast capital expenditure in advanced GPUs is justified if equally powerful models can be trained with far fewer resources.
China’s Centralized AI Strategy
DeepSeek’s achievements highlight China’s broader goal to lead AI by 2030, under the “New Generation Artificial Intelligence Development Plan.” Unlike the U.S. decentralized model, China’s centralized approach combines state funding and private innovation, enabling rapid progress in cybersecurity, aerospace, and beyond. Although export restrictions limit Nvidia GPU access, they have spurred creative engineering—proving ER’s earlier warning that exclusion will certainly fuel profound innovation. Scarcity pushed Chinese engineers to reinvent training methods and hardware stacks, while U.S. labs—flush with top-tier GPUs—lacked similar constraints.
Revisiting Old Warnings and New Realities
In “A Critical Perspective on the UN’s AI Governance Plan,” ER warned that excluding major AI players like China could foster parallel innovation ecosystems—a reality now unfolding as DeepSeek bypasses Western hardware pipelines. ER also cautioned that labelling certain nations’ AI as “rogue” or “unethical” might undercut global standards. If DeepSeek’s open-source platform gains traction, AI governance could fracture further, with nations aligning with either the American or Chinese ecosystem—or, more prudently for Malaysia, adopting “active neutrality,” as proposed in “Active Neutrality in the Pursuit of AI Innovation.”
In “Bridging East and West: How Malaysia Can Seize the AI and Climate Initiative,” ER urged emerging markets to remain flexible—investing in data centers without relying solely on Western or Chinese tech. Yet the global AI stock meltdown sparked by DeepSeek’s breakthroughs highlights the urgent need for “active neutrality” across both geopolitics and technology. Aspiring AI hubs may lean toward open-source solutions that cut upfront costs—precisely what DeepSeek now provides.
Paradigm Shift: Beyond the GPU Arms Race
DeepSeek’s success challenges the “bigger is always better” mindset long championed by U.S. firms. While OpenAI and others rely on massive models running in energy-intensive GPU farms, China’s leaner, more agile paradigm may redefine the AI landscape.
Training advanced models with fewer GPUs could mitigate the environmental impact of AI. If less computing power is truly needed, it might align more naturally with sustainability goals—an angle that ties into ER’s earlier call for inclusive AI and climate tech synergies.
With Nvidia’s market value plummeting, the future demand for expensive, high-end GPUs is being called into question. Even large-scale data-center projects—like those planned in Johor, Malaysia—might face scrutiny if more efficient models proliferate. This opens opportunities for new hardware players specializing in low-cost, high-efficiency AI chips.
Nations once deterred by hardware constraints now glimpse new possibilities. Smaller economies or startups with modest budgets can tap open-source models like DeepSeek’s, building robust AI ecosystems without top-tier GPUs or Silicon Valley support. The playing field just got significantly flatter!
U.S. Responses and Global Governance
U.S. policymakers have long tried to restrain China’s AI progress through hardware embargoes, but DeepSeek’s breakthroughs show such tactics can spur even more ingenuity. Now the U.S. stands at a crossroads: maintain restrictive measures or collaborate while protecting intellectual property. The proposed $500 billion “Stargate” AI project might still be overshadowed if leaner Chinese models flourish.
Policymakers and global bodies—from the U.N. to regional alliances—face a dilemma. Labeling non-Western AI as “rogue” can backfire, especially as Western firms now feel threatened by cost-effective innovations. A truly inclusive framework must acknowledge that transformative ideas can emerge anywhere. As ER has been emphasising, “active neutrality” may allow multiple AI ecosystems to coexist, emphasizing interoperability and ethics rather than one-size-fits-all rules.
The New AI Order
DeepSeek’s rise signals a new era of lean, agile AI, challenging long-held assumptions. Like a modern Sputnik moment, it compels the West to acknowledge that groundbreaking innovation can thrive under tight resource constraints. For nations hedging between the U.S. and China, for investors rethinking GPU-heavy strategies, and for global AI governance bodies, the message is clear: ignoring China’s contributions to AI is no longer an option!
In our increasingly multipolar world, AI leadership will not be dictated solely by who controls the best hardware or invests the most capital. Instead, we are poised to see deeper competition—and perhaps collaboration—around ideas, efficiency, and innovative strategies. If we can foster a genuinely inclusive, globally balanced framework, this upheaval might catalyze more sustainable AI development. If not, expect further fragmentation and the rise of parallel AI orders.
Either way, the era of expensive, compute-hungry AI might soon give way to a new paradigm—one in which agility, open-source collaboration, and resourcefulness are the defining traits of true leadership. DeepSeek has shown what is possible; it is now up to the world to adapt, evolve, or risk being left behind. We have predicted this earlier … and it is just the beginning
Dr Rais Hussin is the Founder of EMIR Research, a think tank focused on strategic policy recommendations based on rigorous research.
** The views and opinions expressed in this article are those of the author(s) and do not necessarily reflect the position of Astro AWANI.
