By Lin Wanwan | At the end of 2025, Bitmain, a Chinese crypto equipment company, was placed on a U.S. national security review list. On Nov. 21, the U.S. Department of Homeland Security launched “Operation Red Sunset,” putting Bitmain under scrutiny on national security grounds. The allegations cut deep: investigators are examining whether its equipment contains remote backdoors and whether, in an extreme scenario, it could deliver a fatal blow to the U.S. power grid.

Why would a Chinese mining company be accused of potentially endangering the U.S. power grid? The answer is Washington’s extreme anxiety over core resources. Silicon Valley is now witnessing one of the most expensive “silences” in the history of technology. Inside AI data centers, tens of thousands of NVIDIA H100 GPUs are sitting idle on the floor, collecting dust. These $30,000 chips, which Jensen Huang has called “industrial gold,” should be running at full tilt, breathing life into GPT-5 or Sora. Instead, they have no power.

Humanity’s most advanced assets are now trapped by the most primitive physical bottleneck. The United States is short of electricity to an almost incomprehensible degree. The gap is 44 GW, roughly equal to the total power generation capacity of a mid-sized developed country such as Switzerland. Yet in the world’s self-proclaimed technology leader, the average wait to power up a new AI data center has stretched beyond 48 months.

The U.S. power grid looks like an old man nearing the end of his strength. At the very moment AI giants were sitting on hundreds of billions of dollars but could not find a socket, they discovered that their lifeline lay in a place they had long looked down on: Bitcoin mining farms. Wall Street suddenly realized that what these miners held was the scarcest asset of the AI era: massive amounts of power already contracted with energy companies.

But they are also realizing that this survival rule, “computing power is electricity,” had already been played out in full a decade earlier on the other side of the Pacific by a group of Chinese engineers. The first ring of “power training grounds” now being repaired for America’s AI era had already been completed in China 10 years ago, only to migrate to the United States three years ago after a regulatory ban. In the contest across the Pacific, accident concealed inevitability.

Just as the tide of history cannot be diverted, each generation has its own fate, and every footnote tells us the same thing: greatness cannot be planned. U.S. power inherited a “Chinese legacy.” History often writes the answer first, then waits for the person who will ask the question. In June 2024, U.S. Bitcoin miner Core Scientific announced news that shocked Wall Street: it had signed a $3.5 billion deal with CoreWeave, often described as NVIDIA’s favorite child, to lease power infrastructure originally used for Bitcoin mining to train AI models.

The news set off a stir in Silicon Valley and was called a “computing power marriage.” But across the Pacific in China, for miners and officials who lived through the “May 19” storm, it carried a very different flavor. Much of the infrastructure that miners such as Core Scientific, IREN and Cipher are using to house NVIDIA H100s, in fact, carries Chinese DNA.

In a sense, the first ring of “power fortifications” for America’s AI era was built by fully absorbing the industrial legacy left behind after China’s great computing power outflow. The person who unintentionally drew up the blueprint was Micree Zhan. Zhan, a classic engineering-minded graduate of the Institute of Microelectronics of the Chinese Academy of Sciences, seemed destined for a life of writing code, drawing circuit diagrams and quietly becoming a technical heavyweight in some technology park.

Then in 2013, Zhan and Jihan Wu founded Bitmain. Zhan reportedly spent only two hours reading the Bitcoin white paper. He may not have understood the future of money, but he understood the essence behind the mathematics: this was an arithmetic game of hash collisions. In 2016, Bitmain made a decision that stunned the industry: it placed enormous wafer orders with TSMC.

The Antminer S9, built on TSMC’s most advanced 16 nm FinFET process, arrived like a bolt from the blue. It was not only a manufacturing miracle in chip history; it also created a kind of “thermodynamic furnace” the world had never seen before. In Zhan’s eyes, the S9 was a chip. In the eyes of the State Grid, it was pure industrial load. It did not cycle day and night like a factory, nor did it fluctuate with the weather.

It ran 24 hours a day with a power curve as smooth as a straight line, indifferent to voltage and origin. From that moment, a new system was born: electricity changed from a public service into a “B2B raw material” that could be priced, traded and monetized instantly. Electricity, a form of energy that is hard to store cheaply once generated, found another way to lodge its value inside strings of numbers.

Bitcoin mining began to become an industry. From hydropower in the mountains of Sichuan to wind power on the grasslands of Inner Mongolia, Bitcoin mining rigs ran wherever China had surplus power. Zhan may not have realized at the time that the industrial standards he defined for Bitcoin mining rigs were unintentionally rehearsing a perfect energy supply model for a desperately power-hungry U.S. AI industry 10 years later. At the frenzy’s peak in 2018, Bitmain alone swallowed 74.5% of the global market.

But that was not the most frightening part. The most frightening part was that the remaining share was also entirely in Chinese hands. Whether it was MicroBT, founded by Bitmain’s former chief chip designer Yang Zuoxing, or Canaan, the pioneer of ASIC mining, the faces were overwhelmingly Chinese. This was not global competition at all, but a “civil war among Chinese engineers” stretching more than 2,000 kilometers: from Aobei Technology Park in Beijing’s Haidian district to Smart Park in Shenzhen’s Nanshan district, 99% of the world’s computing power hearts beat with a Chinese pulse.

It was an absolute closed loop locked down by China’s supply chain, one Silicon Valley had no choice but to look up to. Then, in May 2021, with a single regulatory ban, the roar that had lasted for years along the Dadu River suddenly stopped. For the state, this marked the end of a power-hungry industry. For the industry, it was the beginning of an epic “technology migration.” Tens of thousands of containers were loaded onto ships and sent across the ocean. They carried not only the latest generation of Antminer rigs designed by Zhan, but also a unique “power survival philosophy” honed in China.

One destination was Texas. The state has its own ERCOT grid and the freest, wildest power trading market in the United States. For this group of “computing power refugees” from the East, it looked like an enlarged version of “Sichuan plus Inner Mongolia.” Yet once these Chinese operators actually landed, the U.S. energy sector was surprised to discover that these were not refugees at all. They were a well-equipped “energy special forces” unit.

Back in Sichuan, mining farm owners had secured cheap electricity by drinking heavily with power station chiefs, building relationships and signing relationship-based “understandings.” In Texas, that logic was quickly upgraded into high-frequency trading algorithms. Texas power prices fluctuate in real time, changing every 15 minutes, and in extreme cases can surge from 2 cents to $9.

Traditional Silicon Valley data centers, such as those run by Google and Meta, avoid that volatility at all costs. They are used to living like greenhouse flowers on fixed rates. How did Zhan’s “disciples” respond? With excitement. They turned the experience of manually switching machines on and off in China into automated demand-response programs. When electricity prices went negative, which happens in Texas when there is too much wind power, they ran at full power, devouring electrons, with the grid effectively paying them to consume electricity.

When heat waves hit and power prices soared, they could cut hundreds of megawatts of load within seconds, “sell” the power back to the grid and earn spreads far higher than what they could make from mining. This kind of “energy arbitrage” stunned even veteran U.S. power traders. U.S. mining giants such as Riot Platforms and Marathon have been able to thrive, and to pivot toward AI data centers, precisely because of this power algorithm brought over from China.

Another major legacy of the Zhan era was an extreme obsession with speed in physical infrastructure. A traditional U.S. data center takes two to three years to build, the product of meticulous work by elite engineers. The mining world had no patience for that. Its logic was simple: every second of downtime is a crime against profit. So on the plains of Texas, a kind of “China speed” emerged that left local builders stunned: no elegant glass curtain walls, no complex central air conditioning, only massive industrial fans roaring away.

This “modular, containerized, minimalist cooling” infrastructure model compressed construction cycles to three to six months. Silicon Valley initially mocked this rough but highly efficient engineering capability as an “electronic junkyard.” Today it has become a coveted asset, because AI computing power is exploding too quickly. OpenAI and its peers cannot wait three years. They need this kind of plug-and-play infrastructure now.

In Silicon Valley, the logic is clear: money can buy GPUs, but it cannot buy time. That “time” is the legacy of the frenzy a decade ago. To mine Bitcoin, Chinese miners and their successors frantically bought land, built substations and hoarded what is now priceless “grid interconnection capacity” in the United States. Power quotas have become the new hard currency of U.S. capital. The so-called inheritance was not a pile of scrap silicon, but the right of way into the power grid.

Miners have been able to win massive contracts because, at a time when the United States is short of power, they have their hands firmly on the switch that starts the AI era. The migration night of the “hidden champions.” These violent delights will have violent ends. The year 2018 was a hidden dividing line in business history. That year, ChatGPT founder Sam Altman was still worrying about the survival of a nonprofit.

Elon Musk had just emerged from the edge of near bankruptcy. In their eyes, computing power was still obedient servers in a machine room. But across the Pacific, Zhan and Bitmain had already turned computing power into an industrial beast. They did not understand the future of AI, but that did not stop them from already holding the key to it: how to tame greedy silicon chips at gigawatt scale.

This is a story about rough-edged heroes, state will and history’s jokes. Over seven years, China tolerated and raised a power-devouring beast amid the torrents and coalfields of its western regions. Then, on a summer night in 2021, in pursuit of greater financial security and dual-carbon goals, it uprooted the beast with its own hands. To understand why U.S. AI companies today can bow to miners and absorb an explosion in power demand, one must understand the “energy field exercise” that took place a decade ago on the banks of Sichuan’s Dadu River.

Roll the camera back to August 2019. It was Bitmain’s most glorious moment, and a brief window when China’s mining industry shifted “from gray to white.” At the time, the Sichuan provincial government introduced a policy called “hydropower consumption demonstration zones” to solve the long-standing problem of “abandoned water” during the flood season, when hydropower was generated but could not be transmitted and had to be wasted. This was a real government document in places such as Ganzi and Aba in Sichuan.

According to Caixin’s reporting at the time, under this policy, Zhan’s mining rigs were no longer “unregistered residents” hiding deep in the mountains. They became honored guests helping local grids with “peak shaving and valley filling.” Bitmain, in effect, served as a “supercapacitor” for western China’s energy network. What Zhan was proud of was not only 7 nm chips, but also the ability to convert surplus electricity into digital assets instantly.

At that time, China controlled 75% of global Bitcoin computing power. From Wall Street to the City of London, anyone who wanted to take part in the game had to read Zhan’s mood and rely on the power load of Sichuan and Xinjiang. Yet behind this “gray prosperity,” two swords of Damocles always hung overhead. The first was “financial security.” Regulators had long realized that this was not merely technological innovation, but a huge capital channel operating outside foreign exchange controls.

The second was “dual control of energy consumption.” After China proposed its “3060 dual-carbon” goals in 2020, the destination of every kilowatt-hour became a political account. An industry such as mining, with “high energy consumption, low employment and no physical output,” was destined to be sacrificed on the scale of macro strategy. The turning point in history was fixed precisely on May 21, 2021.

That evening, the State Council’s Financial Stability and Development Committee held its 51st meeting. The meeting readout contained a very short but extremely weighty line: “crack down on Bitcoin mining and trading.” This was no longer the previous language of “risk reminders” or “restricting development.” It was a top-level “clearance order.” The month that followed was the most dramatic 30 days in the history of China’s computing power industry.

Inner Mongolia responded first, cutting power directly to thermal-power mining farms. Xinjiang followed with dragnet inspections. The climax came late on June 19, 2021. That day, Sichuan’s provincial development and reform commission and energy bureau issued a notice ordering the cleanup and shutdown of virtual currency “mining” projects. This became known in the industry as the “Sichuan shutdown night.” Real videos from that night still circulate online: at a super mining farm in Aba Prefecture, as the clock struck midnight, staff on duty pulled down the knife switches on rows of high-voltage distribution cabinets, tears in their eyes.

The roar of cooling fans that had lasted for years, sounding like aircraft taking off, disappeared in an instant. Indicator lights on millions of mining rigs went dark at the same time. The world suddenly became frighteningly quiet, with only the sound of the Dadu River still rushing on. At that moment, the global Bitcoin network’s computing power plunged by nearly 50%. With the determination of someone cutting off his own arm, China forcibly stripped this industry, which consumed more than 100 billion kWh a year, from the veins of the national grid.

China successfully defended its financial perimeter and freed up valuable energy space. But in the cracks of that grand narrative, an unexpected foreshadowing was buried: China kept the electricity, but expelled the people who best understood how to use it. The machines that had been cut off did not disappear. They began to wander. In the second half of 2021, Shenzhen’s Yantian Port saw unprecedented congestion.

According to freight forwarders at the time, tens of thousands of containers were piled up like mountains, all filled with S19 mining rigs dismantled from Sichuan and Xinjiang. It was a computing power version of the “Dunkirk evacuation.” The story returns to the beginning. In 2024, after ChatGPT electrified the world, AI giants suddenly found themselves short of power, substations and high-power server rooms that could be deployed quickly.

China had cleaned up “backward capacity,” but it also packaged up and sent into the world the complete capability to build and operate ultra-large-scale, high-energy-consuming computing power centers. This was a strategic tradeoff involving national financial sovereignty, a decisive abandonment of a high-risk digital high ground. From a macroprudential perspective, it was absolutely correct and necessary at the time. Yet the irony and paradox of history is that the enormous bubble and excess computing power that China deliberately squeezed and expelled eventually solidified across the Pacific into one of the most durable foundations of a rival’s power grid and energy system.

But if one believes the final outcome of this great computing power migration was simply “the East loses, the West gains,” then one has seen only the chips on the table, not the table itself. The AI arms race, stripped to its essence, is the endless consumption of energy by computing clusters, and it will ultimately become a war over electricity costs. In this war of attrition, no country has more strategic depth than China. The United States needs miners as “flexible load” to patch and prolong the life of its grid. It is using miners like a medicinal catalyst to treat the grid’s “diseases of old age.”

China is different. It has State Grid as a central brain. Using ultra-high-voltage transmission, or UHV, it can send the cheapest clean energy from the west to eastern data center clusters continuously and with low losses, like an arterial transfusion. In any case, swept along by history’s current, Bitmain, the master integrator of power management in China’s computing power era, unintentionally became a strategic force reshaping the global energy landscape.

The skills Bitmain and its peers honed on the banks of the Dadu River were unintentionally handed to the other side of the Pacific, helping build the first power wall for the coming U.S. AI era. The fate of the “recruited” miners. So have these co-opted “former Bitcoin miners” really risen overnight and taken a seat at the table of the AI era? The answer may lie in the calculations of the giants. Have you ever wondered why Microsoft and Google, sitting on hundreds of billions in cash flow, would really hand the lifeblood of electricity to miners? Is it only because building their own capacity takes too long? Of course not.

The fundamental reason is that they fear the lessons of history more than anyone. Looking back at business history, there is an invisible tombstone on the mahogany desks of Silicon Valley bosses, engraved with a name that once rang across the sky: Global Crossing. It was the infrastructure giant that died the most brutal death in the 2000 internet bubble. At the time, America’s elites firmly believed the whole world would enter the internet era within a few years, and people would need ever-faster connection speeds.

In that almost religious fervor, founder Gary Winnick borrowed tens of billions of dollars in just a few years and laid more than 100,000 kilometers of fiber optic cable through the deep sea like a madman, connecting the Americas, Europe and Asia. When the internet bubble burst, dot-com websites needed only to shut down servers and lay off employees to complete their bankruptcy cleanups.

Infrastructure suppliers faced huge asset burdens. Fiber optic cables buried under the Pacific, capable of transmitting trillions of bytes per second, became overnight the most terrifying “death assets” in shareholders’ eyes: impossible to sell, impossible to move, left to sit quietly in the dark seabed and rot slowly on the balance sheet. In 2002, Global Crossing collapsed under $12.4 billion in debt.

The most ironic ending was that Li Ka-shing’s Hutchison Whampoa later tried to pick up these assets like scrap metal for less than 1% of their value. Global Crossing proved a brutal truth with its own corpse: in the early stage of technological change, whoever carries irreversible heavy assets becomes the first scapegoat when the cycle turns down. They thought they had seized the data arteries of the future world. Instead, they turned themselves into sacrifices to infrastructure.

Today, Microsoft CEO Satya Nadella and Google CEO Sundar Pichai surely remember that tombstone better than anyone. So when you open their financial reports from the past two years, you find their risk-control core can be summed up in four words: asset isolation. AI giants’ CapEx is soaring, but every dollar is calculated to the bone. On one side are GPUs and custom servers, relatively “general-purpose” assets that can be redirected quickly and, if things really go wrong, sold at a discount.

On the other side are data center buildings, cables and cooling systems, classic “dedicated heavy assets.” The goal is to spin out the hardest-to-exit assets as much as possible. The real calculation is here: they want others to share the “pit.” AI giants use long-term computing power contracts, power contracts and campus leases to stack up a chain that looks like OpEx but, in substance, pushes CapEx risk onto someone else.

For miners that have been recruited and infrastructure players eager to transform, the giants’ pitch is tempting: “You spend the money to build the facility. You handle the liquid cooling retrofit. I sign the contract and consume the power. If AI becomes the dividend of the era, you collect rent under the contract, and I take the business growth and stock-price rewards.” It sounds like risk sharing. Look closer, and it resembles the popular saying: “better you die than me.”

But what if AI ultimately proves to be another Global Crossing-style illusion? At most, the giants pay a termination fee, book an impairment and step off with dignity, ready to tell the next story. The ones who must truly face bank collection letters and explain to creditors what to do with facilities customized for high power density and useful for almost nothing except plugging in H100s will still be this batch of infrastructure bagholders who thought they had finally “gotten a seat at the table.”

Go one step further, and someone may ask: if the AI bubble bursts, can miners simply pull out the GPUs and plug mining rigs back in? The more realistic answer is that most “AI-converted” mining farms cannot swap hardware with one click. AI server rooms require GPUs and liquid cooling; Bitcoin wants ultra-low-cost ASIC containers. The two systems are almost incompatible. The capital market has already given you a valuation premium as an “AI infrastructure stock.” Announcing a return to mining would throw the valuation anchor from AI back to “high-energy-consuming miner.” The buildings would remain, but the story and market value would be liquidated first.

History does not repeat, but it always rhymes. Back then, the fiber was buried under the sea. Today, the server rooms stand on the plains. The people paying the bill have changed, but the roles have not. Greatness cannot be planned. In today’s China-U.S. AI competition, computing power and electricity are the two decisive moves. Although the United States has lost to China’s UHV buildout speed in grid construction efficiency, it has unexpectedly gained a huge “shadow inventory.”

When Silicon Valley’s data center construction is choked by environmental regulations and supply chains, these mining farms can step in immediately and power training for GPT-5 and GPT-6. The charm of the business world lies in its unknowability. All strategic planning is, in essence, trying to drive by looking in the rearview mirror. This was a strategic aid package no one predicted. It was not planned by White House policymakers or simulated by the Pentagon. It was unintentionally built through chaotic market games by wandering Chinese engineers and profit-chasing speculators.

The world is always full of “precise mistakes” and “vague correctness.” Perhaps this is the fable business history leaves behind: greatness can never be planned. - END - References: [1]. Core Scientific Announces CoreWeave Contract Expansion to $3.5 Billion, Bloomberg [2]. The Era of Flat Power Demand is Over, Grid Strategies [3]. Hutchison Whampoa Abandons Bid for Global Crossing, The New York Times [4]. Bitcoin Miners Are Getting Paid for Not Mining, The New York Times [5]. Bitmain’s Micree Zhan: The Man Behind the Mining Empire, Forbes / CoinDesk [6]. ERCOT 2024 State of the Grid Report, Electric Reliability Council of Texas [7]. 2024 Global Data Center Market Outlook, Cushman & Wakefield