Chinese artificial intelligence lab DeepSeek shocked the world on Jan. 20 with the release of its product “R1,” an AI model on par with global leaders in performance but trained at a much lower cost.

Former Intel CEO Pat Gelsinger referred to the new DeepSeek R1’s breakthrough in a LinkedIn post as a “world class solution.” Artificial Analysis’s AI Model Quality Index now lists two DeepSeek models in its ranking of the top 10 models, with DeepSeek’s R1 ranking second only to OpenAI’s o1 model.

It’s not the first time that this Hangzhou-based AI lab has impressed the industry. Funded by parent company High-Flyer—once among China’s top four quantitative hedge funds—the lab has consistently pushed boundaries in AI innovation with its open-source models. In May 2024, DeepSeek’s V2 model sent shock waves through the Chinese AI industry—not just for its performance, but also for its disruptive pricing, offering performance comparable to its competitors at a much lower cost.

The launch of the open-source V2 model disrupted the market by offering API pricing at only 2 RMB (about 25 cents) per million tokens—about 1 percent of ChatGPT-4 Turbo’s pricing, significantly undercutting almost all Chinese competitors. For context, API pricing refers to the cost that companies charge users to access their AI services over the internet, measured by how much text (or “tokens”) the AI processes. A token can be as small as a word or a part of a word.

In December 2024, DeepSeek gained even more attention in the worldwide AI industry with its then-new V3 model. The V3 model was already better than Meta’s latest open-source model, Llama 3.3-70B in all metrics commonly used to evaluate a model’s performance—such as reasoning, coding, and quantitative reasoning—and on par with Anthropic’s Claude 3.5 Sonnet. Even more impressively, this model was released just two months after Anthropic’s latest model launched and in the same month that Meta unveiled Llama 3.3. Once again, DeepSeek’s latest R1 model came only four months after Open AI released a preview version of its o1 model in September 2024.

The gap between Chinese AI labs and their U.S. competitors has rapidly narrowed, and the industry has continued to move at an astonishing pace since the late 2022 release of GPT-3—the first large language model (LLM) that ignited the global AI frenzy, Previously, many U.S. policymakers and business leaders (including former Google CEO Eric Schmidt) believed that the United States held a few years’ lead over China in AI—a belief that appears to be clearly inaccurate now.

U.S.-China AI competition is becoming ever more heated on the industry side, and both governments are taking a strong interest. DeepSeek’s founder and CEO Liang Wenfeng was spotted in a recent meeting with Chinese Premier Li Qiang as the only representative of the AI industry in the room. The week after DeepSeek’s R1 release, the Bank of China announced its “AI Industry Development Action Plan,” aiming to provide at least 1 trillion yuan ($137 billion) over the next five years to support Chinese AI infrastructure build-outs and the development of applications ranging from robotics to the low-earth orbit economy.

The Bank of China’s latest AI initiative is merely one of the many projects that Beijing has pushed in the industry over the years. Back in 2017, the Chinese State Council announced the “New Generation AI Development Plan”—a grand set of strategic guidelines aiming to make China a global leader in AI by 2030, with intermediate milestones to enhance AI infrastructure, research, and broader industry integration by 2025. Since 2017, more than 40 policy and regulatory initiatives have been introduced—with goals ranging from enhancing AI infrastructure to ensuring AI safety and governance.

This includes recent initiatives such as the AI Capacity-Building Action Plan in September, the AI Safety and Governance Framework 1.0, introduced the same month, and the AI Industry Standards System Guidance, published in July 2024.

While these initiatives demonstrate some commitment, the Chinese government has so far played more of a guiding and regulatory role than an investment role in shaping the sector. This contrasts with industries such as semiconductors, electric vehicles (EVs), and solar panels, where the government plays a more pivotal role in development.

Meanwhile, Chinese firms are pursuing AI projects on their own initiative—though sometimes with financing opportunities from state-led banks—in the hopes of capitalizing on perceived market potential. This became particularly evident after ChatGPT-3 showcased breakthroughs in AI technology, which then prompted major technology giants such as Baidu, Alibaba, Tencent, and ByteDance to dive into LLM development.

It took major Chinese tech firm Baidu just four months after the release of ChatGPT-3 to launch its first LLM, Ernie Bot, in March 2023. In a little more than two years since the release of ChatGPT-3, China has developed at least 240 LLMs, according to one Chinese LLM researcher’s data at Github. These range from models created by the aforementioned leading tech giants Tas well as start-ups—such as MiniMax, Zhipu AI, Moonshot AI, and 01.AI—to those developed by prestigious academic institutions, including Peking University and Tsinghua University.

This rapid development underscores the significant progress and focus on AI in China, with industry insiders now remarking that it would be strange not to have an in-house AI model today.

A compelling example of this trend is Xiaomi, a company traditionally focused on consumer electronics and—more recently—the EV sector.

Even Xiaomi is now increasingly venturing into the AI space, developing its own LLM, which highlights the widespread integration of AI development across various sectors in China. Another example is Meituan, a company traditionally focused on delivery services, which has also developed its own LLM and deployed AI assistants on its platform.

Moving forward, DeepSeek’s success is poised to significantly reshape the Chinese AI sector. DeepSeek’s open-source model offers invaluable technical guidance, enabling local tech giants to quickly adopt and build upon its cutting-edge approach with their extensive resources. It will also provide a viable road map for medium- or small-size LLM developers to compete with tech giants in spite of limited resources.

Chinese LLM developers are likely to rapidly optimize DeepSeek’s innovations and deploy them at a pace that poses a serious challenge to U.S. companies. DeepSeek’s reasoning model—an advanced model that can, as OpenAI describes its own creations, “think before they answer, producing a long internal chain of thought before responding to the user”—is now just one of many in China, and other players—such as ByteDance, iFlytek, and MoonShot AI—also released their new reasoning models in the same month.

Moreover, DeepSeek’s success could inject fresh confidence into investors and local policymakers to double down on industry support. Confidence is key—over the past two years, China has faced record-low funding from the private equity and venture capital industry due to concerns about the rapidly shifting regulatory and unfavorable macroeconomic environment.

But DeepSeek’s impact will not be limited to the Chinese AI industry. It will further pervade Silicon Valley beyond its V2 and V3 models. Indeed, a report published in the Information in late January suggested that the biggest U.S. open-sourced player, Meta, is “scrambling” to catch up with the “know-how” from DeepSeek’s V3 and R1 models. DeepSeek’s R1 is MIT-licensed, which allows for commercial use globally. In a recent interview with CNBC, Perplexity CEO Aravind Srinivas shared a similar view.

In Washington, there is an increasingly heated debate over whether the United States’ export control-driven containment strategy needs an overhaul.

Analysts such as Paul Triolo, Lennart Heim, Sihao Huang, economist Lizzi C. Lee, Jordan Schneider, Miles Brundage, and Angela Zhang have already weighed in on the policy implications of DeepSeek’s success. Analysts generally agree on two points: one, that DeepSeek’s model is the real deal, and two, that China’s AI industry is rapidly narrowing the gap with the United States.

Yet it would be unfair to label U.S. export controls on high-end AI chips and semiconductors, introduced in batches in October 2022 and October 2023, as entirely ineffective. Chinese firms such as SMIC have clearly faced challenges, such as low yield rates for advanced 7 nanometer (7 nm) chips and limited progress in advancing beyond the 7 nm node as demonstrated by Huawei’s latest 7 nm smartphone processors and Ascend 910B graphics processing units (GPUs)—critical chips to power AI—manufactured by SMIC’s 7 nm process node.

This shows that export control does impact China’s ability to obtain or produce AI accelerators and smartphone processors—or at least, its ability to produce those chips manufactured with advanced nodes 7 nm and below.

Chinese firms also stockpiled GPUs before the United States announced its October 2023 restrictions and acquired them via third-party countries or gray markets after the restrictions were put in place. These loopholes should be limited by former President Joe Biden’s recent AI diffusion rule—which has proved to be a very controversial regulation in the industry as industry believe the regulations could undermine U.S. AI firm’s global competitiveness by limiting their chip sales abroad, but will take some time and strong enforcement to be effective, given that it has a 120-day comment period and complicated enforcement.

These stockpiled chips have enabled Chinese AI firms to train models on GPUs (e.g. H100, H800, and A100) not too inferior to the ones that U.S. labs are using while advancing domestic alternatives such as Huawei’s Ascend 910B and upcoming 910C GPUs.

That said, export controls have pressured Chinese firms by limiting access to next-generation chips, such as Nvidia’s latest Blackwell GPUs—which began shipping globally in the fourth quarter of 2024 but remain out of reach for China—as well as Nvidia’s next-gen Rubin-series GPU. As these latest generation GPUs have better overall performance and latency than previous generations, they will give U.S. AI labs a hardware and computing edge over Chinese firms, though DeepSeek’s success proves that hardware is not the only deciding factor for a model’s success—for now.

Nonetheless, there is little doubt that U.S. export controls over the past two years have acted as a significant catalyst for Chinese innovation and investment, particularly in sectors such as AI and semiconductors that are directly impacted by these regulatory restrictions. In response, the Chinese government has ramped up its support for key industries, viewing them as crucial for national competitiveness.

This policy shift, coupled with the increasing market potential driven by AI as well as additional market opportunities created by the absence of U.S. companies in China, has attracted a growing number of domestic players. This includes companies such as Huawei, Biren, and Moore Threads in the GPU space, along with semiconductor manufacturing and equipment firms such as SMIC, AMEC, and Naura, which are eager to secure government backing or capitalize the market.

Pressure on hardware resources, stemming from the aforementioned export restrictions, has spurred Chinese engineers to adopt more creative approaches, particularly in optimizing software to overcome hardware limitations—an innovation that is visible in models such as DeepSeek.

The United States’ increasing restrictions have also fostered increased collaboration across the domestic AI value chain, from upstream to downstream, enabling closer partnerships between Chinese companies and in many cases facilitating growing ties between the Chinese government and private sectors. These developments significantly accelerate the pace of domestic innovation, further strengthen local supply chains, and undermine foreign firms’ ability to gain a foothold in China. As a result, China’s technological advancements are increasingly notable in the space of semiconductor and AI, as some experts have already pointed out.

Together, these developments really call into question about the U.S. export control-driven approach and its ability to strike a balance between limiting China’s progress in critical technologies and inadvertently accelerating advancements in these areas. This is a vital question that remains largely unaddressed in Washington.

The rationale behind the U.S. strategy is that by restricting China’s access to advanced AI hardware and limiting its capacity to produce such hardware, the United States can maintain and expand its technological edge in AI, solidifying its global leadership and strengthening its position in the broader strategic competition with China.

However, while the administration of former President Joe Biden has introduced general guidelines on AI governance and infrastructure, there have been few major and concrete initiatives specifically aimed at enhancing U.S. AI competitiveness.

The flaw in this strategy is the focus solely on slowing down competitors without prioritizing the acceleration of domestic innovation and development. As highlighted by Lee, the aforementioned economist, key measures to boost the country’s AI competitiveness must be pursued. Additionally, considering the potential downside of an export control-led strategy laid out earlier, future export controls should be implemented with greater caution, supported by thorough cost-benefit analyses

The U.S. strategy cannot rely on the assumption that China will fail to overcome restrictions. Instead, it must be grounded in a proactive and measured policy framework that ensures that the U.S. outpaces China in AI development if the goal is to prevail in this competition.

In the coming weeks and months, several key developments are likely.

Chinese AI companies, including DeepSeek, will face increased scrutiny from the United States. The Trump administration may also lay out more detailed plan to bolster AI competitiveness in the United States, potentially through new initiatives aimed at supporting the domestic AI industry and easing regulatory constraints to accelerate innovation.

Finally, both the public and private sectors are likely to intensify efforts to address what some are calling a “Sputnik moment” in AI. These developments all point to an increasingly intense U.S.-China technological competition.

As Trump said on Jan. 27, “The release of DeepSeek AI from a Chinese company should be a wake-up call for our industries that we need to be laser-focused on competing to win.” While Trump’s Stargate project is a step toward enhancing U.S. AI competitiveness, there’s a long road ahead.

Source: Foreignpolicy.com

We give you energy news and help invest in energy projects too, click here to learn more

Crude Oil, LNG, Jet Fuel price quote

ENB Top News 
ENB
Energy Dashboard
ENB Podcast
ENB Substack