
In my recent book, The Widening Turn, I observed that the United States and China are locked in a frantic race for technological supremacy that is sending the pursuit of innovation into overdrive.
And last week, the engine of the world’s collective intelligence was firing on all cylinders. One minute it was the United States breathlessly churning out an announcement about “Stargate,” a potential $500 billion dream that’s short on specifics but heavy on big name, heroic tech celebrities… and the next minute it was China quietly pumping out details about DeepSeek R1 and ByteDance’s Doubao-1.5-pro.
That’s when I had to wonder if our widening turn is spinning into a phase of strategic encirclement. Which is to say, China is innovating “open-source” AI models that by some measures surpass more expensive proprietary western models.
China's "Digital Silk Road" strategy is building on the Belt and Road's foundation at an ever-quickening pace.
How did we get here? Where might we be headed?
China’s Enormous Economic Experiment
Have you heard of the “China Playbook”?
The China Playbook represents the methods Chinese companies have employed to dominate a series of industries throughout the 21st century. Political centralization coupled with an extreme form of economic decentralization empowers private entrepreneurs to innovate and expand. China offers low-interest loans, free real estate, and other state assistance to strategic industries, helping to fundamentally transform the economics of those businesses while driving down costs (eg, solar panels, aluminum, steel, tires, textiles, shipbuilding). The process plays out in the following manner—
First, the Chinese state targets an industry that it regards as strategic, establishing incentives to attract entrepreneurs such as low-interest loans and free buildings. And then those entrepreneurs pile in, producing a rambunctious new domestic sector. This savage competition, including from foreign companies, results in massive innovation and a buildup of production capacity—so much that it spills over into exports. By this time, a few superstar Chinese companies emerge, and in the final stage the state tightens up rules and forces other, weak Chinese companies to die.
As Steve LeVine of The Information puts it, this is “a brutal, gladiatorial competition that leaves just a few ruthlessly efficient companies standing. The end result is businesses with huge factories and vertical supply chains that earn ultraslender profits but not only do not fail as predicted, but continue to beat virtually everyone on price. These survivors then push abroad for profits.”
That is, as the economics of products are fundamentally altered and collective capacity significantly exceeds Chinese domestic demand, Chinese companies push into the global market, and some Western firms die off as a result.
The War of 100 Models
Sometime around the late 2000s, China turned this playbook to the sphere of technology (think of the global EV and battery industries and the raw and processed materials that fuel them, such as nickel, manganese, lithium, and cobalt). By 2015, the Chinese government embedded its plans for the EV industry into a larger “Made in China 2025” program of industrial planning.
Today, the China Playbook is turning to all things AI with an intense techno nationalism. Chinese founders have launched dozens of startups in what’s been called the “hundred-model war.” As described in The Widening Turn, “It’s a heated competition between China’s biggest Internet companies—Baidu, Alibaba, Tencent, and Bytedance—and dozens of startups that are developing their own LLMs in what Chinese media refers to as ‘a war of 100 models.’”
China’s national innovation ecosystem comprises a set of institutions and policies aimed at not only avoiding, in the words of Xi Jinping, ‘technological vassalization’ to the West, but is also intended to astound the West… just as the West astounded China in the 19th century.
And that’s what seemed to be happening last week when ByteDance released Doubao-1.5-pro, “an upgrade to its flagship AI model, which it claims outperforms OpenAI's o1 in AIME, a benchmark test that measures how well AI models understand and respond to complex instructions…. Other Chinese firms that have unveiled their own reasoning models in the past weeks include Moonshot AI, Minimax, and iFlyTek.”
But it’s Chinese quant trading firm High-Flyer Capital Management’s startup DeepSeek and its “open-weight” large language model that has raised the most eyebrows across the West (as Stanford Associate Professor of Computer Science Percy Liang points out, “just a friendly reminder that they are not *open-source*; there’s no training/data processing code, and hardly any information about the data.”).
Either way, as The Wall Street Journal explains, “While DeepSeek’s flagship model is free, the company charges users who connect their own applications to DeepSeek’s model and computing infrastructure. An example is a business that wants to tap the technology to give AI answers to customers’ queries. Early last year,” in a page straight out of the China Playbook, “DeepSeek cut its prices for this service to a fraction of what other vendors charged, prompting the industry in China to start a price battle.”
All the while, in a feat of elegant engineering simplicity, DeepSeek was artfully designed such that the system’s algorithms, software, and hardware work together to make it easier to share data across chips. To create v3/R1, DeepSeek reworked its training process to reduce the strain on its GPUs, with the model using a similar “chain of thought” approach as ChatGPT o1.
This process utilized only 2,048 Nvidia H800 chips (a dumbed-down version of the H100 chip Nvidia is permitted to sell in China) and spent just $5.5 million to train the model. “The result was a model with scores on popular benchmarks that in some cases beat models from OpenAI, Anthropic, and Meta. By comparison, Meta said it trained its Llama 3 405B model, released in July, using 16,000 H100s, which are more expensive to run than the H800 chips.”
Alas, “DeepSeek has also released six smaller versions of R1 that are small enough to run locally on laptops. It claims that one of them even outperforms OpenAI’s o1-mini on certain benchmarks.” In effect, says Perplexity CEO Aravind Srinivas, “DeepSeek has largely replicated o1-mini and has open-sourced it.”
Might more AI projects start to use DeepSeek's model in open-source or via their API (app developers can freely download DeepSeek or buy access to it through a cloud-based application programming interface)… to power their AI endeavors, “resulting in an ecosystem effect and them becoming a standards setter”?
A Different Economic Universe than Stargate
In other words, one begins to wonder if so many vast, expensive data centers will be needed in a world where smaller AI applications might run on personal computers or even gaming PCs.
Consider the observation that—
DeepSeek isn't the one that needs to make a ROI on half a trillion dollars worth of data centers (or whatever fraction of that amount actually materializes) with a product that's now offered free by the competition. And that's probably exactly the point of DeepSeek's strategy: to fundamentally change the economics of the market so as to make OpenAI's model obsolete.
It sounds a lot like how Tinci Materials—the world’s largest electrolyte producer, based in Guangzhou, China—followed the China Playbook a few years ago to capture much of the electrolyte market by devising a new way to process LiPF6 (a salt used in almost all battery electrolytes, called lithium hexafluorophosphate), eliminating costly manufacturing steps.
In other words, by innovating.
“Chinese companies,” observed Steve LeVine in reference to the China Playbook, “appear already to be operating in a different economic universe from Western rivals.”
Daft in Davos
As MIT Technology Review points out, “As well as prioritizing efficiency, Chinese companies are increasingly embracing open-source principles. Alibaba Cloud has released over 100 new open-source AI models, supporting 29 languages and catering to various applications, including coding and mathematics. Similarly, startups like Minimax and 01AI have open-sourced their models.”
Is China pursuing a strategic encirclement of expensive proprietary AI models so favored by some corporations in the West? You wouldn’t know it from much of the chatter at Davos last week.
As physicist and startup founder Steve Hsu notes, “Stargate and lots of Davos blather about massive build-out of compute infra and data centers. Very jarring as coincides with a tour de force demonstration that human brainpower applied to algos/models can give you ~30x improvement in training and inference costs.”
In fact, some think Stargate-style AI model-building is “a money trap.” For Chamath Palihapitiya, “What this (DeepSeek) means is that it’s very difficult to plan spending programs to create advances and will likely render entire months and possibly years of energy, hardware, R&D, OpEx and CapEx wasted and useless as new innovations are discovered. Open-source is the clear winner.”
And still another AI watcher offered the following epiphany—“I just realized DeepSeek R1 JUST made reasoning cheaper than a cup of coffee, open-source unlike GPT4, and somehow outperforms Claude 3.5 Sonnet. ‘Made in China’ AI now costs $0.50/hour while U.S. minimum wage is ~$15/hour. Intelligence just got priced like an electricity bill.”
Yann LeCun, Chief AI Scientist at Meta, put it this way: “DeepSeek,” he wrote, “has profited from open research and open-source (e.g. PyTorch and Llama from Meta). They came up with new ideas and built them on top of other people's work. Because their work is published and open-source, everyone can profit from it. That is the power of open research and open-source.”
A Chinese Flywheel?
But that’s not all. We might ponder the potential effects of all this on the China Playbook itself. Now, as yet another observer notes, “Imagine the impact of DeepSeek on the China speed of spillover effects and synergy to other connected parts of China's industrial ecosystem of electric vehicles, drones, semiconductor chips, smartphones, batteries, robots, etc.”
Strategic encirclement expands outward and proceeds apace.
China’s strength across multiple overlapping industries creates a compounding effect for its industrial policy efforts…. And if you’re already strong in multiple overlapping industries, then this creates a mutually reinforcing feedback loop that further strengthens your position in all of these connected industries…. As a result, Chinese tech companies are increasingly becoming tech Swiss Army knives, starting in one industry but then quickly branching out into a range of adjacent tech domains: smartphones, EVs, autonomous vehicles, generative AI, drones, robotics.
For those in the BPO industry, might we add... the contact center?
Playing Two Different Games
One should also begin to wonder… are China and the United States playing two entirely different games in the pursuit of technological supremacy? Is it just me, or is the AI race beginning to look a bit like the ancient Chinese game of Wei qi (pronounced “way chee”), also known as “GO.” As Henry Kissinger explains in his book On China, Wei qi implies a concept of gradual strategic encirclement. It’s very different than the decisive approach of attack embraced by the West in the game of chess.
When it comes to GO, says Kissinger—
At the end of a well-played game, the board is filled by partially interlocking areas of strength. The margin of advantage is often slim, and to the untrained eye, the identity of the winner is not always immediately obvious. Chess, on the other hand, is about total victory. The purpose of the game is checkmate, to put the opposing king into a position where he cannot move without being destroyed…. If chess is about the decisive battle, ‘Wei qi’ is about the protracted campaign. The chess player aims for total victory. The ‘Wei qi’ player seeks relative advantage…. Where the skillful chess player aims to eliminate his opponent’s pieces in a series of head-on clashes, a talented ‘Wei qi’ player moves into ‘empty’ spaces on the board, gradually mitigating the strategic potential of his opponent’s pieces. Chess produces single-mindedness; ‘Wei qi’ generates strategic flexibility.
One can’t help but sense that the United States and its European allies seem trapped in a hopelessly ideological mindset, perceiving the competition with China as a struggle between black and white pieces on a chessboard, between democracy and autocracy, good versus evil, and that China’s continued rise must be defeated with a set of blunt instruments—threats, military posturing, export controls, and economic sanctions. This is the United States playing the role of what Professor of International Relations John Mearsheimer calls a “Crusader State,” with China as cartoon villain.
The Chinese, meanwhile, sit across the game table as bemused realists and keep playing Wei qi. The Americans seem unable to understand that decades spent glorifying moneymaking has a real cost when it comes to competing with a civilization that has spent decades building up a multidimensional industrial ecosystem.
Kissinger, in On China, is clear about what’s going on here. “Where the Western tradition prized the decisive clash of forces emphasizing feats of heroism, the Chinese ideal stressed subtlety, indirection, and the patient accumulation of relative advantage.”
Engineering Innovation Across Time
The myth that the Chinese can’t innovate at the level of the West is absurd. In China, a top-down construct managed by an imperial bureaucracy and tax system has long produced highly disciplined innovation ecosystems across time. Thousands of years ago, the ancient Qin Dynasty literally created China by pulling off a stunning feat of engineering, building vast canals and irrigation systems. A lead engineer named Li Bing designed Dujiangyan, which still irrigates the Sichuan plain today.
David Goldman, Washington Fellow at the Claremont Institute's Center for the American Way of Life, explains—
Engineers in the service of Emperor Qin Shi Huang, who gave his name to what we now call ‘China,’ built an artificial island in 256 BCE to divide the Ming River near the city of Chengdu, and cut a 66-meter channel through Mount Yulei using heated stones and cold water to crack the living rock. In place of deadly floods, the spring runoff of the Ming River irrigated 2,000 square miles of farmland, turning the Sichuan desert into the breadbasket of China.
Chinese engineers are still innovating today. DeepSeek, based in Hangzhou, China, was founded in July, 2023, by Liang Wenfeng, an alumnus of Zhejiang University with a background in information and electronic engineering. “Like Sam Altman of OpenAI,” says MIT Technology Review, “Liang aims to build artificial general intelligence (AGI), a form of AI that can match or even beat humans on a range of tasks.”
Liang has been just as resourceful and creative as Li Bing was during the Qin Dynasty and his impact might prove equally influential. DeepSeek figured out how to reduce memory usage and speed up calculation without significantly sacrificing accuracy. “The team loves turning a hardware challenge into an opportunity for innovation. The whole team shares a collaborative culture and dedication to hardcore research,” said a former DeepSeek employee. It’s the same spirit of innovation displayed by Qin Dynasty engineers who utilized heated stones and cold water to crack living rock, achieving large-scale, formalized water management.
Goldman also points out that today’s Communist Party of China arose “in China’s classic historical pattern and governs as a new incarnation of China’s ancient Mandarin caste.” Thousands of years after the rule of Emperor Qin Shi Huang, Xi Jinping is pushing tech entrepreneurs to undertake their own vast infrastructure projects, building open-weight AI models that distribute intelligence via digital canals drawing from vast data lakes, the modern floodplains reshaping the global economic landscape in real-time.
MIT Technology Review adds that, “According to a white paper released last year by the China Academy of Information and Communications Technology, a state-affiliated research institute, the number of AI large language models worldwide has reached 1,328, with 36% originating in China. This positions China as the second-largest contributor to AI, behind the United States.”
Thomas Qitong Cao, an assistant professor of technology policy at Tufts University, adds that, “This generation of young Chinese researchers identify strongly with open-source culture because they benefit so much from it.”
The possibilities are abundant. What if open-sourced genAI delivers above expectations? What if both compute and foundation models become commodities through a distributed model of local deployment that democratizes AI that runs everywhere? Optimized smaller models could mean significantly lower energy usage and inference costs. In such a world, some expensive data center infrastructure could become a liability.
As Steve Hsu observes when looking at DeepSeek’s progress, exponential improvement through “human algo ingenuity” might be more important than improvements in hardware.
Will the road to AGI… and superintelligence… run through China?
A Revolutionary Force
The race to achieve technological superiority is heading into overdrive. The West, and the United States in particular, still possess formidable advantages, foremost among them being, in the words of David Goldman, “a disruptive creativity that challenges established modes of thought.”
But, as described in The Widening Turn, the aggressive nature of America’s “Effective Accelerationism” movement (“e/acc”), Silicon Valley’s eccentric subculture hellbent on relentless technological innovation, feels at times like a bull in a China shop. While e/acc also favors open-sourcing AI software, its overall philosophy represents nothing less than a revolutionary force in the world.
In his “Techno-Optimist Manifesto,” Marc Andreessen, one of Effective Accelerationism’s most prominent voices, cries out to America in almost Nietzschean fashion: “We believe in ambition, aggression, persistence, relentlessness—strength. We believe in merit and achievement. We believe in bravery, in courage…. We are not primitives, cowering in fear of the lightning bolt. We are the apex predator, the lightning works for us.”
And president of Y Combinator Garry Tan struck a similarly Nietzschean tone when he insisted that, “Because if we can build here, we’re gonna take over the whole country. We’re going to take over every nation in the world.” This is beginning to sound a bit like Napoleon before he undertook a multi-decade push across the European continent.
As such, China will need to rely on thousands of years of experience in sophisticated statecraft to manage American ambition, to negotiate the accelerating widening turn the world has hydroplaned into. And political leadership in the United States had better rediscover the lost art of diplomacy to skillfully steer America’s ambitions… while at the same time resisting what might be a process of ongoing strategic encirclement spreading out from the Middle Kingdom.
Sound dramatic? As Ezra Klein of The New York Times notes, “AI is breaking through at the same moment the international system is breaking down. The United States and China have drifted from uneasy cooperation to grim competition, and both intend to be prepared for war. Attaining AI superiority has emerged as central to both sides of the conflict.”
The problem is, a competition between the United States and China that descends into armed conflict could mean the breakdown of international order, perhaps the end of international society itself.
A Fevered Drive to Innovate
With no sense of irony, Meta Platforms has set up several “war rooms” to dissect DeepSeek and use the insights gleaned to improve its own open-weight model, Llama. Indeed, it seems all of Silicon Valley is in a state of discombobulation over the gauntlet that’s been thrown down by China's innovators... hand-wringing here, calls to action there, with a whiff of panic thrown in. It’s a moment that’s beginning to rival the release of ChatGPT in late 2022 in its shock value, when Google’s Sundar Pichai declared a “Code Red.”
As The Widening Turn suggested, the United States and China are two pistons in an over-heated global engine fueled by what feels like a wholly new technological “will-to-power” that would’ve made the aforementioned German philosopher Friedrich Nietzsche grin. Because Nietzsche felt that a mysterious will-to-power is a fundamental human drive. And, intriguingly, he observed that this drive isn't just about domination… it’s also a craving to overcome challenges in pursuit of growth.
"Human beings,” claimed Nietzsche, “do not seek pleasure and avoid displeasure. What human beings want... is an increase of power; driven by that will they seek resistance, they need something that opposes it—displeasure, as an obstacle to their will-to-power, is therefore a normal fact; human beings do not avoid it, they are rather in continual need of it."
Considering the amount of displeasure in Silicon Valley this week, when it comes to the pursuit of technological innovation, China might be the best thing that ever happened to the United States… so long as the two competitors can avoid a crack-up.
And yet, “I worry,” tweeted Chamath Palihapitiya, “that in this current melee, we’ve overspent billions on dumb features which these next-gen models will roll over in the next 12 months or earlier. Lots of capital losses are coming.” Indeed—
With R1, DeepSeek essentially cracked one of the holy grails of AI: getting models to reason step-by-step without relying on massive supervised datasets. Their DeepSeek-R1-Zero experiment showed something remarkable: using pure reinforcement learning with carefully crafted reward functions, they managed to get models to develop sophisticated reasoning capabilities completely autonomously…. The technical breakthrough here was their novel approach to reward modeling…. This simpler approach turned out to be more robust and scalable than the process-based reward models that others have tried.
Our Own Sweet Foolish Will
The world will need steady hands to negotiate the times we’re spinning through. More than anyone, Nietzsche understood that the will-to-power isn’t always necessarily stable. In the 19th century, one of the best illustrations of this fact came from a writer Nietzsche admired deeply, Russian novelist Fyodor Dostoevsky. In Notes from Underground, Dostoevsky’s Underground Man ponders those inexplicable moments when human beings turn suddenly from the relentless pursuit of efficiency and progress… and succumb to a disruptive spasm that defies reason and sends all our “logarithms to the devil.”
Over to you, Fyodor—
Man is stupid, you know, phenomenally stupid; or rather he is not at all stupid, but he is so ungrateful that you could not find another like him in all creation. I, for instance, would not be in the least surprised if all of a sudden, A PROPOS of nothing, in the midst of general prosperity a gentleman with an ignoble, or rather with a reactionary and ironical, countenance were to arise and, putting his arms akimbo, say to us all: ‘I say, gentleman, hadn't we better kick over the whole show and scatter rationalism to the winds, simply to send these logarithms to the devil, and to enable us to live once more at our own sweet foolish will!’ That again would not matter, but what is annoying is that he would be sure to find followers—such is the nature of man.
We’d best be attentive to the potential whims of man and machine alike. Because futurist Ray Kurzweil’s chase for “the Master Algorithm”—a single, general purpose learning algorithm—is a frantic chase, filled with unpredictable turns.
I think,” offers Meta’s Chief AI Scientist Yann LeCun, “the shelf life of the current (LLM) paradigm is fairly short, probably three to five years. I think within five years, nobody in their right mind would use them anymore, at least not as the central component of an AI system. I think… we’re going to see the emergence of a new paradigm for AI architectures, which may not have the limitations of current AI systems.”
If LeCun is correct, we could be on the verge of another AI revolution that goes beyond the generative model of pattern recognition and reaches into emerging “world models” that promise reasoning machines with the will to plumb the depths of the real world.
How many of us fully appreciate the rate of exponential technological change fueling this widening turn the world is hydroplaning into?
Who can understand the nature of man?
Image credit: from www.tripchinaguide.com
コメント