“Any American policy toward Asia must come urgently to grips with the reality of China.” – Richard Nixon

Since Nixon’s historic opening to China, the U.S. has watched a slow erosion of its industrial base. Telecommunications, steel, shipbuilding, solar panels, batteries, and genome sequencing have all been gradually absorbed into Beijing’s orbit, ceded, not conquered, through a mixture of Western complacency and China’s relentless state-backed industrial ascent through cost-cutting and unfair trade practices.

Artificial intelligence and semiconductors must not be next.

This is not just a rivalry over code or chips. This is a war of operating systems, a full-spectrum contest between competing visions of technological sovereignty and global order. And unless America gets serious, fast, we risk watching the foundation of 21st-century power, compute, slip into the hands of our greatest strategic adversary.

China has a playbook, and it’s working:

1. Top-down coordination to dominate AI, big data, quantum, and biotech
2. Massive investment in energy, manufacturing, foundries, and rare earths
3. Global saturation through open-source AI and aggressive infrastructure exports
4. A full-stack plan to own sovereign data, data center real estate, and inference globally

And for China, so far so good: Chinese LLMs are already trailing American LLMs in the open source market. GPU export controls were meant to slow them down. Instead, they’ve accelerated Beijing’s push for self-reliance and pushed some of our own allies toward uncertainty.

Here’s the truth: technical excellence isn’t enough. Global adoption is the real battleground. And the only way to win is to ensure U.S. AI platforms become the global default, from chips to inference, from open models to trusted infrastructure.

As Nixon said, “We must come urgently to grips with the reality of China.” That reality today is AI. This new Cold War will define the next half-century.

Total AI = Total War

China has driven its ascent through innovation, reverse engineering, subsidies, and scaling, initially dominating its domestic market and then expanding across key global supply chains deemed vital by the CCP. This strategic approach has been applied in industries such as batteries, drones, telecom, genome sequencing, solar, steel, and other strategic sectors. Now, this state-backed engineering effort is shifting towards AI, as part of a wider plan to weaken U.S. influence and reshape international trade routes, with the goal of ensuring the Chinese government's stability and independence.

Historically, China achieves what it prioritizes, and its AI ambitions are a central part of its strategic goals. Following Nixon’s 1972 visit that paved the way, Deng Xiaoping initiated China’s “four modernizations” in 1978: agriculture, industry, defense, and science technology.^[i]

This plan established the basis for China’s industrial growth. Xi Jinping has adapted this strategy to AI, infusing it with nationalist urgency. He has stated that “the next ten years will be a key decade," emphasizing technologies like AI, big data, quantum computing, and biotech. Central to these is AI and the semiconductor supply chain. For Xi, AI is more than just a sector; it is a transformative force of stability. China needs to surpass Western countries, enhance its economy, and expand its digital infrastructure worldwide as a means of securing the Party’s stability and the country’s self-reliance.

In April, the Politburo, China’s top 24 leaders, held a comprehensive study session on AI.^[ii] Xi didn’t just discuss LLMs and GPUs. For the first time, he highlighted NVIDIA’s core software-hardware advantage and called for the development of a comprehensive domestic AI ecosystem. In CCP politics, study sessions reveal what’s coming next. They indicate what the Communist Party considers crucial, and in this case, it’s about closing the AI gap with the U.S. across all levels of the AI stack, from semiconductors to advanced models.

Beijing is actively promoting its national champions, like Huawei, to “go global,” using its extensive toolkit of state-backed partnerships. Through the Digital Silk Road, the Belt and Road Initiative, AI governance pacts, and targeted industrial subsidies, China is exporting its computing infrastructure and standards.^{[iii] [iv]} Major Chinese tech firms have leveraged these collaborations to expand their reach in telecommunications, energy, and computing sectors.

Chinese tech giants have leveraged these tools to bolster their influence across the global telecom, energy, and cloud industries. Domestically, Beijing is increasing its investment in AI hardware, allocating billions to semiconductor development. Despite economic difficulties, Chinese alternatives to NVIDIA and AMD are gaining ground, with non-GPU AI servers increasing their market share by 49% year-over-year.^[v]

The Resource War Behind the Compute War

The AI era is exposing a harsh truth about the US’s failed industrial policy since Nixon's "Great Reset" of Chinese relations and China’s entrance into the WTO: America outsourced its will to build. The next global Cold War won’t be won with just software or semiconductors. It will be won with raw horsepower: power, minerals, and industrial capacity.

In response to our GPU export controls, Beijing introduced new export controls on critical minerals and magnets, inputs essential to everything from missiles and EVs to the GPUs powering the AI boom. China controls nearly 50% of global rare earth reserves and over 70% of global production. The U.S., by comparison, holds just 1.9 million tonnes and remains highly dependent on Chinese supply chains. China wants chips, but it knows its advantage lies in a resource war.

At the same time, the foundation of the AI lifecycle, electricity, is flashing red when compared to China. The largest planned AI data centers already require more than 1 GW of power. By 2030, U.S. data center demand alone is projected to require an additional 18 GW, about three times the current demand of New York City. AI is moving real-world work online, and that shift is triggering the most aggressive power demand growth the West has ever seen. In other words, there is a need to add three NYCs to the US power grid by 2030.

While Western policies have centered on productivity offsets and environmental compliance for decades, China has adopted a “whatever it takes” approach. Their capacity to produce electricity is 2.5 times greater than that of the US and continues to grow linearly year after year.

On a per-capita basis, China still trails the U.S. and the rest of the G20 in electricity production, but it is on a rapid path to surpass the U.S. within the next decade. Meanwhile, U.S. per-capita energy output has actually decreased over the past 20 years. As AI is bringing real-world operations online and demand for computing power is skyrocketing, this trend should be raising alarms in D.C. However, they prefer to focus on controlling China’s Best Buy’s purchases of consumer GPUs rather than expanding our grid.

China is not just thinking about chips. They’re securing every part of the supply chain: grid capacity, rare earths, silicon, packaging, cooling, and fabs for all AI semiconductor production. It’s working. China now holds 21% of global foundry capacity, second only to Taiwan, and is on track to take the lead 2030.^[vi]

“Since the US started the semiconductor trade war, China has been accelerating the development of an independent ecosystem. China is rapidly emerging as a key player in the foundry market,” said the Yole group, which monitors the semiconductor industry.

They predicted that China’s foundry production capacity will continue to grow, dominating the global market by 2030 with 30%. If China successfully controls Taiwan by the end of the decade (one of their publicly stated goals), they will hold half of the world's chip production.

To understand the scale of production in Taiwan and China, the “Stargate in the Gulf” is expected to be among the largest data center complexes in the world, surpassing the total capacity of Virginia’s data center corridor. According to industry analysis by semiconductor research firm SemiAnalysis, TSMC’s manufacturing capabilities are expected to double almost every year. Impressively, securing deals like those under discussion with the UAE and Saudi Arabia would only use about 3.4% of TSMC’s anticipated 2027 capacity output.

Note: (1) Middle East deals refers to 500k advanced chips shipped annually to the UAE and 500MW deals with NVIDIA and AMD for the KSA

We can’t keep just defending ourselves while Beijing advances. The future of AI won’t depend only on model weights or benchmark ratings. Instead, it will depend on who controls the industrial foundation. Those who control essential inputs such as power, minerals, and manufacturing will influence the final results. As the Communists like to say, “We will seize the means of production.” When it comes to AI, they are heading towards victory.

Beijing’s Open Source Bet

China is doubling down on its open-source AI strategy, and it’s paying off. With an ecosystem expanding at an estimated 26% annually and projected to reach $194 billion by 2031, China is establishing itself as the global alternative to Western closed systems frameworks.^[vii] Even before international recognition of viral firms like DeepSeek and Manus, the number of specialized AI companies grew from about 4,000 in 2023 to over 4,500 by mid-2024, representing 15% of all AI firms worldwide time.^[viii]

China’s AI Market Growth

Notes: Data was converted from local currencies using average exchange rates de respective year.
Most recent update: March 2025
Source: Statista Market insights

Since Chinese AI firms gain more ground, they will have an advantage that no other nation can match: full support from Beijing’s industrial policy backing their global expansion. Zhipu AI, spun out of Tsinghua University, is already trying to leverage the Belt and Road Initiative to build AI partnerships abroad.^[ix] China’s large language models are not only catching up but are already surpassing or matching Western counterparts on key benchmarks. Beijing's strategy is effective. According to Stanford’s 2025 Artificial Intelligence Index, PRC LLMs are now only slightly behind US models' capabilities.^[x]

Source: LMSYS,2025

Chart: 2025 AI Index report

However, focusing only on models and Chinese internal software adoption, which is a symptom of an advantage rather than an advantage in itself, overlooks the real threat: Chinese hardware's looming dominance abroad.

The Huawei Problem

Huawei isn't just the most powerful tech company in China. It is one of the most powerful in the world. What started in the 1980s as a small producer of telecom switches has grown into a national champion competing in or leading all areas from undersea cables and 5G to electric vehicles and operating systems. Huawei now belongs to an elite group of companies, along with Google, Microsoft, and Apple, that operate their own full-stack operating systems. It develops its own chips that meet or even surpass the capabilities of U.S. tech giants. It has full support from Xi Jinping, and more importantly, it now produces its own GPU.^[xi]

Huawei’s Ascend series has quickly become a major competitor to U.S. dominance in AI hardware. The Ascend 910C, along with DeepSeek R1, provides about 60% of the inference capacity of NVIDIA’s top offerings H100.^[xii] That’s not full parity, but it’s close, more affordable, and getting better each year. The 910C is scheduled for mass production this year through Huawei’s strategic partnership with SMIC, Beijing’s state-supported semiconductor giant.^[xiii] The full range of Ascend products is expected to ship nearly a million units this year, excluding the 1.5 million units apparently illicitly produced by TSMC.^[xiv]

For context, NVIDIA is expected to ship 5.2 million units of Blackwell this year, but that number will decrease to 1.8 million next year. In 2026, NVIDIA is projected to ship 5.7 million units of the Rubin GPU and 1.5 million units of the Vera CPU.^[xv]

SMIC does not have access to the most advanced lithography tools, but they are pursuing a different strategy that leverages their strengths. SMIC has shown the ability to work with existing technology to make incremental improvements, positioning themselves to build enough scale to compete with U.S. GPU companies globally. If SMIC continues to increase yields and manufacturing capacity at the same pace since 2024, Huawei could sell millions of Ascend chips as early as 2026.^[xvi]

Huawei’s worldwide distribution network for AI hardware is already established and expanding. In 2024, Huawei reported a remarkable 69.4% year-over-year growth in international markets, mainly fueled by its extensive global telecommunications presence.^[xvii] The same sales team that contributed to building much of the world’s 5G infrastructure is now moving to sell AI infrastructure, such as locally produced GPUs. This shift is a deliberate component of their wider industrial strategy. Keep in mind, one of Silicon Valley’s principles of innovation is that distribution can often outperform a better product.

China’s top-level industrial blueprint, the 14th Five-Year Plan for Digital Economy Development, explicitly calls for overseas expansion in digital infrastructure, including data centers and AI compute.^[xviii] The directive is clear: support Chinese digital enterprises in “going global” and integrate them into international technology cooperation. In 2023, Beijing executed this strategy through a multi-agency action plan that tasked various state ministries and regulators with helping Chinese computing firms expand into Belt and Road countries. The aim: deploy Chinese-built computing capacity across emerging markets and build worldwide service capabilities in line with Beijing’s strategic goals interests.^[xix]

Adding momentum, China’s Ministry of Foreign Affairs revealed its AI Capacity Building Action Plan late last year.^[xx] Presented as a generous effort to bridge the global digital divide, the initiative seeks to establish AI infrastructure partnerships worldwide, leveraging China’s significant influence in multilateral organizations like the United Nations. The message to developing countries is clear: avoid U.S. export controls and partner with a willing, well-funded China. Huawei now serves not only as China’s domestic leader but also as the global leader in Beijing’s AI infrastructure ambitions.

Becoming the Export Nation

We need to stop thinking that this ends once someone creates a model like “the bomb,” because then the contest is over and everyone can go home. This will be an ongoing struggle, not just for hardware dominance but also for winning the hearts and minds of the global developer community.

The biggest strategic advantage U.S. companies have isn’t just the chip itself, but the software platforms and libraries that developers worldwide prefer to build on. It is distribution and utilization. Today, more than five million developers across the globe write code for U.S. GPU architectures, boosting America’s lead in artificial intelligence every time they write a line of code.^[xxi]

The Trump administration showed foresight by removing the AI Diffusion Rule, which could have alienated allies and ultimately encouraged the use of Chinese hardware abroad. NVIDIA CEO Jensen Huang is also right to stress the importance of keeping access to the China market, not just for immediate revenue but because “50% of the world’s AI developers” are there.^[xxii] Every Chinese coder working on an Nvidia GPU, even one restricted by export controls, unintentionally supports America’s ecosystem. Conversely, each US GPU used by a Chinese researcher is a Chinese GPU losing a customer. This imbalance offers strategic leverage. It would be unwise to relinquish it.

But this lead is shrinking. Huawei’s developer ecosystem is expanding rapidly. In 2021, its Ascend platform supported 400,000 developers. By 2023, that number had quadrupled to 1.5 million. As of March 2025, Huawei has 3.3 million developers on Ascend, with over 50,000 core contributors to its CANN software stack and more than 6,000 deep ecosystem engineers. This is creating a new ecosystem and an alternative AI foundation that opposes the West. The software-hardware advantage that once protected U.S. leadership is now under attack. Our export controls, meant to maintain that advantage, are speeding up its decline under the guise of patriotism.^{[xxiii] [xxiv] [xxv] [xxvi]}

The U.S. is at a crucial turning point, where it can shape the next wave of global infrastructure, just as China did with 5G. In the age of “America First” AI diplomacy, compute isn’t optional. It’s a matter of sovereignty.

Exporting American tech isn’t charity; it’s strategy. Total export dominance isn’t about market share; it’s about shaping the world’s defaults. China understood this with 5G. Whether the U.S. learns from that playbook or waits until it’s too late will determine whether we lead the stack in the AI and 6G era or get locked out.

Winning this new Cold War is similar to how we won the last one: reaching escape velocity first.

Aaron Ginn is the CEO of Hydra Host, a horizontal data center operating system and multi-channel compute monetization platform powering the next generation of American AI. Aaron is a co-founder of the Foundation for American Innovation (FAI) and Fabius Labs.