On June 25, at a closed-door meeting of the China Strategic Think Tank, Zhao Yanjing, a professor at Xiamen University, reframed the diagnostic logic of China's economic predicament with a "macro balance sheet": the constraint lies not on the asset side (labor and production capacity), but on the liability side (capital and equity claims).
Under this framework, he made a highly impactful judgment: China does not need to defeat the U.S. head-on in the AI race. Instead, it should stick to the open-source route to compress the profit margin of closed-source models, and wait for the U.S. AI bubble to burst before making counter-cyclical bottom-fishing of technology assets. Through low-price and open-source strategies, China is shifting the battlefield from computing power—where the U.S. has chosen to compete—to the AI valuation bubble, which is the U.S.'s weak spot.
In June 2026, the U.S. stock AI sector experienced a severe correction—the Nasdaq fell by over 4% in a single day, NVIDIA dropped 6.2%, Broadcom fell 7.92%, and Micron plunged 13.25%. Xinhua News Agency issued a warning under the headline "U.S. Stocks Plunge: Is the AI Bubble Facing a 'Wake-Up Call'?" Meanwhile, DeepSeek V4 has approached or even surpassed top closed-source models in multiple benchmark tests, narrowing the technological gap between open-source and closed-source models to 3-12 months.
Is the bubble bursting? Is China ready for the second "bottom-fishing window"? This article is excerpted from Zhao Yanjing's speech at the closed-door meeting, with Kegong Liliang (Industrial Insight) supplementing industrial facts and differentiated judgments.
A Balance Sheet to Reframe the Coordinates of "AI Competition"
Zhao Yanjing's analytical tool is a macro balance sheet. The left side is the asset side, corresponding to labor—current assets are savings, and fixed assets are investment and consumption (i.e., domestic demand); the right side is the liability side, corresponding to capital—liabilities include bank loans and export surpluses, while equity includes stocks, bonds, real estate, and natural resources.
The two sides are identical, and the insufficient side determines the upper limit of the economic scale. Currently, China's constraint lies on the liability side: deflation, unemployment, and the "flexible employment" population increased from 240 million in 2024 to 320 million in 2026, accounting for over 40% of urban employment. Meanwhile, labor and production capacity on the asset side are severely overcapacity—which means that any further labor-saving technological progress will not only fail to drive growth but will instead exacerbate unemployment and hyper-competition.
AI is precisely a labor-saving technology. Zhao Yanjing's judgment is therefore extremely sharp: in an economy like the U.S., where the constraint is on the asset side (labor shortage), AI improving labor productivity means expansion; but in an economy like China, where the constraint is on the liability side (labor surplus), AI will only lead to greater unemployment and more persistent deflation.
"The most dangerous policy misjudgment in China is believing that high-tech and 'new quality productive forces' can replace real estate to drive the Chinese economy," Zhao Yanjing pointed out. "New quality productive forces" are on the asset side, while real estate is on the liability side.
Under the constraint of the liability side, progress on the asset side cannot automatically bring about expansion on the liability side—"new quality production" must create "new quality demand."
But does this mean China can only resign itself to defeat in the AI race?
"No Need to Win Head-On, Just Wait for the Bubble to Burst"
Zhao Yanjing's answer is negative. His logical chain is as follows:
First, the massive capital investment by the U.S. in AI leaves "little suspense" about its victory, but the huge capital investment also makes the bursting of the bubble a "highly probable event."
Second, China adheres to the open-source route, making it impossible for U.S. closed-source large models to profit—the performance of open-source models such as DeepSeek, Qwen, and Llama has approached that of top closed-source models, while the cost is only one-tenth of the latter. When enterprise customers can use open-source solutions to complete 80% of standardized tasks, the subscription price premiums and API service fees of closed-source models will face systemic pressure.
Third, once China's equity market (stock market + real estate market) recovers and steps out of the U.S. dollar-dominated economic cycle into a different one, it will be possible to engage in counter-cyclical bottom-fishing of technology assets when the U.S. stock AI sector crashes. The historical precedent for this "wait + bottom-fish" strategy is a series of overseas M&As by China after the 2008 financial crisis: Geely’s acquisition of Volvo (2010): Ford bought Volvo for more than $6 billion in 1999. By 2010, Ford had suffered massive losses and was forced to slim down. Chinese private automaker Geely acquired Volvo’s passenger car business, including its technology, R&D and global sales channels, for $1.8 billion — less than one-third of Ford’s original purchase price.
The internationalization path of Joyson Electronics: It went public through a reverse merger around 2010 and successively acquired Germany's PREH (acquiring world-class core automotive electronics technology) and Germany's Quin, and later bottom-fished the bankrupt Japanese Takata assets and swallowed the U.S. KSS company, ultimately growing into the world's second-largest automotive safety system supplier. Sany Heavy Industry's acquisition of Putzmeister (2012): The German construction machinery giant "Elephant" faced financial pressure during the European debt crisis. Sany Heavy Industry, together with CITIC Industrial Fund, acquired 100% equity for 360 million euros, directly obtaining global marketing channels, core technologies, and high-end brands.
CXMT's acquisition of Qimonda (2016): Qimonda, a DRAM company under Germany's Infineon, went bankrupt in 2009 due to the financial crisis. Through the acquisition of Qimonda’s parent company Polaris Innovations Limited, CXMT obtained over 10 million technical documents and 2.8TB of data, improved the 46nm process to the 10nm level, and obtained the "entry ticket" for DRAM R&D and mass production.
The commonalities of these cases are clear: they all occurred during the window period of economic crisis and asset price collapse of competitors; they all relied on China's independent economic cycle—the cyclical independence in 2008 was provided by the real estate boom, and the cyclical independence in 2026 must be provided by the recovery of the equity market. "Without the prosperity of real estate, many of today's high-tech industries in China would not exist at all," Zhao Yanjing wrote. "For China to replicate the success of that year, it must restore the expansion of the liability side as soon as possible and step out of the haze of deflation. Otherwise, even if the U.S. AI bubble bursts ten thousand times, China will not be able to seize the opportunity to bottom-fish."
New Variables in 2026: Is the Bubble Bursting?
In June 2026, the U.S. stock AI sector experienced a severe round of correction. On June 5, the Nasdaq Composite Index fell by 4.18% in a single day, marking the largest single-day drop since April 2025. NVIDIA fell 6.2%, Broadcom continued to fall 7.92% on top of the 12.59% drop on the 4th, Micron Technology dropped 13.25%, Advanced Micro Devices (AMD) fell 10.86%, and Oracle dropped 9.59%. The technology sector of the S&P 500 led the decline with a drop of 5.78%. The next day, Xinhua News Agency published an article asking: "U.S. Stocks Plunge: Is the AI Bubble Facing a 'Wake-Up Call'?" The direct trigger for this round of correction was the heating up of high-interest-rate concerns—strong employment data led to a decline in market expectations for the Federal Reserve's interest rate cuts. But the deeper market anxiety lies in: do the commercialization returns of AI match the valuations?
A research branch of MIT previously released a report pointing out that as many as "95% of organizations are getting zero return on their generative AI investments," and "only 5% of integrated AI pilot projects have generated millions of dollars in value." Sam Altman, CEO of OpenAI, also publicly admitted that the AI sector is in a bubble, and "some investors might lose a lot of money."
Warren Buffett's response is even more signal-worthy. Berkshire Hathaway's cash reserves reached as high as $397 billion in the first quarter of 2026. It remains on the sidelines regarding pure AI tech stocks and general AI projects, deploying technology only in scenarios that can create substantial value for existing businesses (such as using AI to optimize dispatching at BNSF Railway and using AI to process standardized claims at GEICO Insurance).
Meanwhile, China's AI sector is forging an independent trend. A Goldman Sachs report in March 2026 pointed out that there is no bubble in China's AI concept stocks, which have risen by an average of 50% since the "DeepSeek Moment," but current valuations are still undervalued by 50%-100% compared to potential value. The 52-week rolling return correlation between China’s AI concept stocks and U.S. tech stocks is only 23%, far below the 69% level between the U.S. and other regions. The "strangulation" logic of the open-source route is emerging.
DeepSeek V4 has achieved a verification rate of 83.7% on SWE-bench and a 99.4% test rate on AIME 2026, surpassing GPT-4.1 in coding and reasoning. The technological gap between open-source and closed-source has narrowed from several years in 2023 to 3-12 months. DeepSeek's permanent price cuts have further broken the original competitive balance of the industry, and foundation models are entering a K-shaped divergence—mid-tier general model manufacturers face the greatest pressure, lacking both the capability moat of closed-source frontier models and the cost-performance advantage of open-source alternatives.
Anthropic's annualized revenue from April to May 2026 broke through $3 billion to $4.5 billion, with the gross margin on reasoning services reached as high as 70%. The "cost democratization" of open-source and the "capability premium" of closed-source are forming a layered coexistence, rather than a simple substitution relationship.
The Premise of Bottom-Fishing Is Not "Waiting," but "Repairing"
Zhao Yanjing's framework is penetrating, but simplifying "counter-cyclical bottom-fishing" to "waiting for the U.S. bubble to burst" has three blind spots that need to be calibrated by industrial facts.
First, the valuation anchoring of AI targets is completely different from that of physical manufacturing in 2008. Volvo had factories, revenue, and PSA; Takata had airbag market share; Putzmeister had a global channel for concrete machinery—the "residual value" of these assets can be relatively objectively evaluated during a crisis. But the valuation of AI startups is based on "future possibilities," and their core technology assets (model weights, data pipelines, talent teams) may not be "sold at a discount" after the bubble bursts, but directly dissipate. The difficulty of "bottom-fishing" in 2026 is far higher than in 2008.
Second, geopolitical scrutiny constitutes a rigid barrier. When Geely acquired Volvo in 2008, the U.S. government did not strongly intervene; but in 2026, for Chinese capital to acquire core U.S. AI assets, CFIUS (Committee on Foreign Investment in the United States) review is almost an inevitable obstacle. In the national strategic context where AI is positioned as "the largest mobilization of scientific resources since the Apollo Program," the "acquirability" of technology assets itself has become a variable.
Third, the logic of "open-source strangling closed-source" requires more refined differentiation. Currently, open-source models are indeed compressing the profit margin of closed-source models in standardized tasks, but the closed-source camp (OpenAI, Anthropic, Google) still firmly occupies high-end scenarios such as complex reasoning, coding agents, and multimodality. Anthropic's annualized revenue from April to May 2026 broke through $3 billion to $4.5 billion, with the gross margin on reasoning services reached as high as 70%. The "cost democratization" of open-source and the "capability premium" of closed-source are forming a layered coexistence, rather than a simple substitution relationship.
A deeper question lies in: the "bottom-fishing" capital in 2008 came from the expansion of the real estate liability side; what is the source of alternative capital in 2026?
Zhao Yanjing's answer is to repair the equity market (stock market + real estate market), driving savings into the capital market through large-scale interest rate cuts and debt monetization. However, Kegong Liliang (Industrial Insight) believes that this framework needs to be supplemented by the physical support dimension of industrial profits: the globalized profits of China's new energy industrial chain (in 2026, the export proportion of NEVs has exceeded 30%), the control-layer root technologies of industrial automation enterprises (Inovance, Supcon), the digital capabilities grown from heavy assets by manufacturing natives (Midea, Sany, Haier), and the new quality demands created by the landing of embodied AI scenarios—the cash flow generation capabilities of these "heavy asset sides" are constituting a new source of capital different from 2008.
The essence of counter-cyclical bottom-fishing is not to bet on the opponent's mistakes, but to confirm that one's own balance sheet is healthy enough. The health in 2008 came from the expansion of the real estate liability side; the health in 2026 must come from the dual confirmation of equity market recovery and the physical support of industrial profits. Without the latter, even if the U.S. AI bubble bursts ten thousand times, China will not be able to bottom-fish—it might even be bottom-fished in reverse by its opponents.
Reference Frame: The "Dosage" for the U.S. to Walk Out of the Great Depression During WWII
At the end of the text, Zhao Yanjing introduced an extreme but powerful historical reference: the U.S. before WWII. The Roosevelt New Deal failed to pull the U.S. out of the Great Depression; it was WWII, which truly turned the tide, that brought about rapid debt expansion. In 1940, the total U.S. federal debt was about $43 billion, accounting for 44% of GDP; by the end of the war in 1945, the total debt soared to about $258 billion, flipping about 6 times in 4-5 years, and the debt-to-GDP ratio reached a historical peak of 119% in 1946. The Federal Reserve implemented "yield curve control," pinning short-term Treasury yields at 0.375% and setting the long-term ceiling at 2.5%, which was essentially debt monetization.
Massive military orders activated idle capacity, and the unemployment rate plummeted from 14.6% in 1940 to 1.2% in 1944. By 1945, the total U.S. industrial output accounted for over 50% of the global total, laying the foundation for its post-war hegemony.
"The experience of the U.S. walking out of the Great Depression shows that if the dosage is not enough, even a correct New Deal will not be able to walk out of the recession," Zhao Yanjing wrote. "The same is true for China today; without proactive, large-scale expansion of the liability side, it is absolutely impossible for the economy to automatically walk out of the recession." He directed this judgment towards a severe option: either wait for a "Pearl Harbor-style event" to provide the U.S. with a political rationale for large-scale borrowing, or overcome psychological barriers to proactively borrow like the "Four Trillion" stimulus package. Kegong Liliang (Industrial Insight) believes that the inspiration from this historical reference lies not in the "war demand" itself, but in the concept of "dosage"—the macroeconomic effect of technological breakthroughs depends on whether the constraint side is unlocked by policy expansion of sufficient intensity.
The "Four Trillion" stimulus in 2008 indeed provided the dosage, but its side effects (local government debt, overcapacity) are also clearly visible. The challenge in 2026 lies in: how to find a new liability-side expansion tool with sufficient dosage without repeating the old prescription.
Conclusion
Does the U.S. stock AI correction in June 2026 mean that the "wake-up call" has arrived? The answer remains uncertain. But what is certain is: if China's equity market fails to complete its recovery before this round of adjustment, even if the U.S. AI bubble bursts, China will miss the window for the second counter-cyclical bottom-fishing.
From Geely's acquisition of Volvo to CXMT's acquisition of Qimonda, the generation of Chinese entrepreneurs in 2008 completed technological leapfrogging at "less than one-third of the price." Their confidence came from a healthy balance sheet—the expansion of the liability side supported by the real estate boom. The new generation of technology M&A players in 2026 need not only to wait for opponents to make mistakes, but also to repair their own balance sheets. This is not speculation, but a dual test of strategic patience and institutional capability.
This article is excerpted from Zhao Yanjing's speech at the closed-door meeting of the China Strategic Think Tank on June 25, 2026, with Kegong Liliang (Industrial Insight) supplementing industrial facts and differentiated judgments.