Cuộc chạy đua AI giữa Mỹ và Trung Quốc đang bước vào giai đoạn quyết liệt
-
Sau gần 3 năm kể từ khi Mỹ khởi động làn sóng AI tạo sinh, Trung Quốc là đối thủ duy nhất tiệm cận trình độ công nghệ này.
-
Mỹ dẫn đầu nhờ các bước đột phá trong chip AI, mô hình ngôn ngữ lớn và agent AI, với các công ty chủ lực như OpenAI, Google, Microsoft và Anthropic.
-
Trung Quốc phản công với DeepSeek, Alibaba, Moonshot, phát triển mô hình AI tiết kiệm tài nguyên và theo hướng nguồn mở — chiến lược giúp nhanh chóng lan rộng toàn cầu.
-
DeepSeek ra mắt mô hình R1 vào tháng 1/2025 với chi phí thấp nhưng hiệu suất cạnh tranh, khiến OpenAI phải xem xét lại chiến lược nguồn mở.
-
Về sáng chế, cả Mỹ và Trung Quốc chiếm phần lớn bằng sáng chế AI tạo sinh, với Google, Baidu, Zhejiang University, Microsoft đứng đầu.
-
Về chính sách, Mỹ ưu tiên hạ tầng và năng lượng; Trung Quốc thúc đẩy AI như lợi ích công, mở lớp AI từ tiểu học, đồng thời kiểm duyệt nội dung nhạy cảm.
-
Mỹ chi hơn 344 tỷ USD trong năm 2025 cho hạ tầng AI, với các “đại gia” như Amazon, Google, Meta đẩy mạnh xây dựng trung tâm dữ liệu.
-
Trung Quốc đầu tư 98 tỷ USD vào AI trong năm 2025, 56 tỷ từ chính phủ, còn lại từ các tập đoàn như Tencent, Alibaba.
-
Mỹ có lợi thế vượt trội về tài năng AI nhập cư (70% sinh viên sau đại học là người quốc tế), nhưng đang bị ảnh hưởng bởi chính sách visa.
-
Trung Quốc thúc đẩy chương trình "Qiming" để thu hút hơn 7.000 nhà khoa học về nước, chú trọng tinh thần dân tộc thay vì lương cao.
-
Hạn chế lớn của Trung Quốc là thiếu chip AI cao cấp do lệnh cấm từ Mỹ; trong khi Mỹ tập trung vào dữ liệu và trung tâm điện lực.
-
Dù Trung Quốc đã thu hẹp khoảng cách, các bài kiểm tra như Humanity’s Last Exam cho thấy mô hình của OpenAI, Google, xAI vẫn vượt trội với độ chính xác trên 20%; DeepSeek mới đạt 14%.
📌 Mỹ vẫn giữ vị trí dẫn đầu AI với đầu tư 344 tỷ USD trong năm 2025 và vượt trội về độ chính xác mô hình (trên 20% trong Humanity’s Last Exam). Trung Quốc phản công bằng AI nguồn mở và đầu tư 98 tỷ USD, nhưng bị giới hạn bởi chip và kiểm duyệt. Cuộc đua công nghệ lớn nhất thế kỷ đang ngày càng nóng lên giữa hai siêu cường.
https://www.bloomberg.com/news/articles/2025-08-06/who-is-winning-the-artificial-intelligence-race-the-us-or-china
The AI Showdown: How the US and China Stack Up
The technology, the state, the strategy and the money: Here’s how the two economic superpowers are vying for AI domination.
By Saritha Rai and Seth Fiegerman
It’s been almost three years since the US kicked off the artificial intelligence boom, and most of the world is still struggling to catch up. Only one country is close to matching the US in AI development: China.
Chinese firms such as DeepSeek, Alibaba Group Holding Ltd. and Moonshot have released AI models that approach the capabilities of leading systems from US companies, spurred on by a government that has made AI leadership a national priority. To that end, it’s rolling out a package of generous state support and a national network of interlinked data processing hubs.
Chinese firms such as DeepSeek, Alibaba Group Holding Ltd. and Moonshot have released AI models that approach the capabilities of leading systems from US companies, spurred on by a government that has made AI leadership a national priority. To that end, it’s rolling out a package of generous state support and a national network of interlinked data processing hubs.
China’s ascendance has set off alarm bells in Silicon Valley and Washington. To stay ahead, the Trump administration put out an AI Action Plan in July calling for cutting red tape to build more AI data centers and tap into enough electricity to power them. The US will do “whatever it takes to lead the world in artificial intelligence,” President Donald Trump said in a speech.
The outcome of this contest may determine which country is the true technological superpower of the 21st century.
The Technology
The US led the way with the key breakthroughs that have defined the current AI era. American companies pioneered the advanced computing chips and large language models that underpin today’s “generative” AI chatbots. US companies such as OpenAI Inc. and Alphabet Inc.'s Google also debuted AI systems that mimic the human process of reasoning and spit out videos, images and audio. They’ve also released so-called agents that are designed to field increasingly complex tasks on a user’s behalf.
China’s tech companies have quickly followed in their footsteps, signaling how fast these technological advances become commoditized. Faced with constraints around access to US chips, Chinese companies have built AI models that are able to do more work with less computing power. Perhaps most significantly, they have embraced open-source standards, making their AI products freely available for anyone to use and adapt for their purposes, unlike the proprietary systems favored by some top US firms. China is flooding the global market with these platforms, while OpenAI, Anthropic and others are charging hundreds of dollars a month for access to some of their top-tier, closed models, with the goal of offsetting the immense cost of AI development.
Two-Horse Race
US and Chinese companies and institutions dominate in generative AI patent applications
0100200300400500600GoogleZhejiang University, ChinaMicrosoftUniversity of Electronic Science &Technology of ChinaBeijing Baidu Netcom Science &Technology Co.NvidiaIBMTsinghua UniversityHangzhou Dianzi UniversityState Grid Corp. of China
China’s push into open-source reflects a strategic choice to sacrifice some profits in the short term to ensure that Chinese AI is adopted globally. The 14th five-year blueprint for the country's development, published in 2020, promoted open-source technologies. There’s been no explicit directive from Beijing obliging AI developers to embrace this approach. Rather, the strategy is the result of a complex interplay of market-driven incentives and state priorities. Managers at some Chinese AI startups have said they see it as the quickest way to enter new markets and compete with US models there.
In its Action Plan, the White House encourages the development of open models, citing their potential to “become global standards” in business and academic research. “For that reason,” the plan says, “they also have geostrategic value.” After DeepSeek stunned the global market with the release of its R1 model in January, purportedly built at a fraction of the cost of leading US platforms, OpenAI’s leadership said the company needed to rethink its open source strategy. In August, OpenAI released a pair of open and freely available AI models.
The State
In both countries, AI is an economic, political and national defense imperative — and a tool for exerting soft power abroad.
“We want the world to be built on an American technology stack, not on Chinese or some other country’s technology stack,” US Vice President JD Vance said at an AI event in Washington in late July.
Chinese President Xi Jinping has said AI should not be a “game of rich countries,” in a not-so-subtle dig at America’s profit-driven approach to AI. On July 26, Chinese Premier Li Qiang announced the establishment of an international body to ensure that AI doesn’t become the preserve of just a few nations. He made the announcement at the World AI Conference in Shanghai, where AI was presented as a public good and several sessions focused on setting global standards and governance for the industry.
Beijing has moved with breathtaking speed, introducing AI classes in elementary schools, launching a venture fund to invest in the technology, and allowing state-backed companies to deploy AI services for a variety of functions including surveillance of the public space and digital finance. China’s centralized political system requires AI developers to prioritize societal stability and censor certain results. DeepSeek, for example, restricts answers to questions about issues that are sensitive in China, such as the suppression of pro-democracy protests in Beijing's Tiananmen Square in 1989 and the independence movement in Taiwan, which China regards as lost territory.
At the same time, the US is also pushing to rethink certain speech in AI systems. Trump issued an executive order in July mandating that companies with government contracts — pretty much every major AI company — demonstrate that they “do not manipulate responses in favor of ideological dogmas such as DEI,” among other provisions. It remains unclear how developers will prove their systems are free from terminology and viewpoints that Trump finds objectionable.
Debates about intellectual property, algorithmic bias and the safety risks posed by AI play out in US courts, boardrooms and public forums. A “fair use” principle enshrined in US law makes it possible — in certain circumstances — to use books, news stories, song lyrics and other copyrighted material without paying their creators. Whether that principle applies to AI development, however, remains a legal gray area. Leading developers including OpenAI and Anthropic are embroiled in a growing list of copyright lawsuits from media companies, authors and artists whose material they ingest to train AI models.
China's legal landscape around AI and copyright is also evolving. There have been legal battles over the use of content scraped from streaming services, online video sharing platforms and other sources. But there have been few binding decisions penalizing China’s AI community for copyright infringement. The country’s courts have ruled that ingesting copyrighted material for AI model training is a type of temporary reproduction that doesn’t break the country’s copyright laws as long as it isn’t reused in its original form.
Several leading US developers, including OpenAI, have called on the Trump administration to free them from state-level laws that make it riskier for them to access copyrighted works. If China’s “developers have unfettered access to data and American companies are left without fair use access,” OpenAI said, “the race for AI is effectively over.”
US states have implemented a patchwork of AI regulations to make the technology safer — but which technology executives argue risks slowing down development. Trump’s AI Action Plan comes with a threat to withhold certain funds from states that put burdensome rules on the emerging technology.
The Money
DeepSeek has demonstrated a more cost-efficient approach to AI development. Yet building and supporting AI services remains an expensive endeavor.
In the US, AI startups raised more than $100 billion in funding in the first half of 2025 alone, according to data from PitchBook. Meanwhile, four large tech firms – Google, Meta Platforms Inc., Microsoft Corp. and Amazon.com Inc. – are expected to spend more than $344 billion in 2025, with much of it going to build the data centers necessary to run AI models.
Big Tech Boosts Capex to $89 Billion in Second Quarter
050$ 100 BQ22024Q3Q4Q12025Q2
Source: Company Filings, Bloomberg
In the decade leading up to the AI boom, Chinese government-backed venture capital plowed an estimated $912 billion into sectors seen as critical to the country’s development, with almost a quarter of that going toward 1.4 million AI-related firms. This year, China’s AI capital expenditure is projected to reach as much as $98 billion, marking a 48% increase from 2024 levels, according to a Bank of America report. The majority of that — $56 billion — is expected to come from the government, and another $24 billion from big tech firms including Alibaba and Tencent Holdings Ltd. A new AI Industry Investment Fund is injecting $8.2 billion into early-stage projects. Provincial governments are in a race to attract AI startups, offering subsidies, grants and housing to lure entrepreneurs and AI experts.
The Talent
For decades, the US has dominated the field for AI talent by attracting researchers, coders and academics from overseas. The landmark 2017 Google paper Attention Is All You Need, which gave rise to today’s generative AI products, had eight authors, six of whom were born outside the US. The two others have immigrant backgrounds.
That’s not an anomaly. In the US, 60% of top AI companies count at least one immigrant among their founders, and 70% of graduate students in specialisms related to AI are international. That talent pipeline risks being upended by tighter visa regimes, shifting political winds and fluctuating federal research funding.
China, historically plagued by brain drain, is working to reverse the tide. The country’s Thousand Talents Plan, launched in 2008, has been quietly rebranded in recent years into successors such as the Qiming Programme. It has lured more than 7,000 scientists and entrepreneurs — many of them Chinese nationals educated abroad — back to domestic laboratories and universities.
While some American companies have begun offering pay packages of $100 million or more to top AI researchers, China’s pitch is less about money and more about patriotism.
The Infrastructure
China has constructed a formidable AI ecosystem rooted in vast, state-curated pools of data drawn from sources including social media, surveillance footage and financial transactions. It has also tapped into a sprawling network of data centers powered increasingly by renewable energy.
China’s closed and censored internet represents a potential disadvantage, as the datasets drawn upon to train Chinese AI models are more likely to be incomplete or biased, so they risk generating inaccurate responses. That could be one further incentive for prioritizing open-source models: They can be more easily engineered by customers worldwide to complement and enrich their datasets so they’re better at specific tasks.
A more serious obstacle for China is its lack of cutting-edge AI chips. Export controls spearheaded by the US have throttled China’s access to Santa Clara, California-based Nvidia Corp.’s top-of-the-line AI processing units, prompting Beijing to double down on domestic production. Chips from Chinese powerhouses Huawei Technologies Co. and Semiconductor Manufacturing International Corp. are not as effective as Nvidia’s Hopper and Blackwell when it comes to training more sophisticated models, but they are good for many use cases and deployed increasingly for operating AI systems and applications. Huawei is also constrained in how many chips it can offer compared with Nvidia, but the firm is rapidly scaling up.
There may be other workarounds for Chinese developers. US officials are probing whether DeepSeek circumvented export restrictions by purchasing prohibited Nvidia chips through third parties in Singapore.
In the US, top tech firms are investing in ever-larger data centers stocked with thousands of Nvidia chips to train and operate more sophisticated AI models. In his first days after returning to the White House, Trump championed a joint venture from OpenAI, Oracle Corp. and Japan’s SoftBank Group Corp. to invest up to $500 billion in physical infrastructure to support AI. The AI Action Plan also calls for expediting permits for data centers and embracing new sources of energy to power them.
By continuing, I agree to the Privacy Policy and Terms of Service.
The US has been constrained by aging power grids. China added 429 gigawatts of new generation capacity in 2024 alone — far more than the US. Some industry leaders, including OpenAI, have called for the US to do more to supercharge the amount of power available for AI data centers, including moving more aggressively to boost nuclear capacity. “There’s just not enough energy right now,” Amazon CEO Andy Jassy said in 2024.
The Players
OpenAI kicked off the generative AI boom with the launch of ChatGPT and remains the best-known AI lab in the US, if not the world. Other leading AI developers in the country include AI pioneer Google as well as Anthropic and xAI, Elon Musk’s startup. Meta has slipped behind in recent months, but it’s working to catch up by poaching top talent from competitors with unprecedented pay packages.
China’s AI market is dominated by a mix of large internet firms — Baidu, Alibaba and TikTok owner Bytedance Ltd. — and rising stars such as Zhipu and Moonshot. Baidu’s Ernie and Alibaba’s Qwen have closed the gap with US platforms in areas such as reasoning and coding proficiency. DeepSeek can tackle complex word problems, while Zhipu can be asked to generate a short video, and users get results that feel comparable to any service in the US. Chinese startup Manus received global attention by launching a general purpose AI agent that can create spreadsheets and conduct research for users, much like tools from OpenAI and others.
US tech executives say their Chinese counterparts are still playing catch-up, at least for now. On various leaderboards, including Chatbot Arena and Humanity’s Last Exam, models from US firms still rank at the top. Humanity’s Last Exam, in particular, was designed by AI safety researchers with thousands of questions to test how the most advanced AI systems perform across math, science and other subjects. As of mid-2025, only three companies had surpassed 20% accuracy on the test: OpenAI, Google and xAI. DeepSeek came in at 14% and a version of Qwen was at 11%.
“It is very hard to say how far ahead we are,” OpenAI Chief Executive Officer Sam Altman said during a May hearing in Washington. “But I would say not a huge amount of time.”
Thảo luận
Follow Us
Tin phổ biến
TAG
AI giáo dục
AI sinh-y-duoc
AI nghệ thuật
AI pháp lý-quản trị-chủ quyền
AI models
AI xã hội
AI prompts
AI kiến thức-khóa học
AI công nghiệp-lĩnh vực
AI edge
AI viễn thông
AI tools
AI chính phủ
AI cybersecurity
AI so sánh
AI đạo đức
AI tips
AI market
AI quân sự
AI an toàn-an ninh-techwar
AI việc làm
AI doanh nghiệp
OpenAI ChatGPT
AI môi trường-năng lượng
AI skill-talent
AI & công nghệ khác
AI nghiên cứu
AI chips-hardware-compute
AI vs con người
AI coding assistant
AI mở-nguồn mở
AI năng suất
AI startup-M&A
AI tương lai
AI báo chí
AI data
AI bản quyền
AI PC
AI riêng tư
AI deepfake-ảo giác-ANTT
AI ảnh-video-music-âm thanh
AI minh bạch
AI nhỏ
AI nông nghiệp-thực phẩm
AI ngân hàng-tài chính
AI giao thông
AI smartphone
AI robotics-auto-agents
AI consumer devices
AI manufacturing
AI benchmark
Telecom
AI thành công-thất bại
Digital
Semi-Cloud-DC-Green
HTS
STI
FAQ