According to a leading chief economist, the AI bull market is entering its "autumn" phase, and while winter may not be far off, investors shouldn't abandon the opportunity to harvest the season's bounty simply because colder days lie ahead. Instead, the focus should be on actively preparing for the final gains.

The economist projects that for at least the next one to one-and-a-half years, until the intense competition among large language models subsides, the diffusion of AI applications is likely to sustain the broader AI ecosystem's momentum.

There's no need to avoid food for fear of choking; this AI bull market will likely only establish its major peak after fully reflecting industrial trends. Going forward, the key watchpoint is the proliferation of AI applications—only when capital expenditure and real-world deployment form a virtuous cycle, combining AI infrastructure with efficiency gains in the real economy, can the entire AI ecosystem continue to thrive.

By 2028, whether AI capital spending reaches $2 trillion or $3 trillion, the critical question will be whether it generates sufficient returns through improvements in global total factor productivity. While it seems difficult to justify the math, this remains currently unprovable. From a one-year perspective, investors should approach this AI bull market with both caution and optimism.

At least in the short-to-medium term, there's little need to worry about long-end US Treasury yields quickly pricking the AI bubble, as US government officials are even more eager than the market to suppress rates. This is indeed a long-term risk, as US debt concerns, inflation risks, and the AI bubble are fundamentally difficult to reconcile over time.

In the third quarter, long-end US Treasury yields are likely to peak and then retreat—rising first before declining. It's quite possible that the 10-year Treasury yield could initially climb to around 5%, before settling back to roughly 4.3% by the fourth quarter.

In the later stages of this AI bull market, investors should pay more attention to established companies blooming anew, similar to a convertible bond investment strategy. The economist favors companies with deep value and healthy cash flows that can develop second growth curves by leveraging the global arms race among large model developers to enhance their own business models—this represents the new opportunities from AI diffusion.

These insights come from the chief economist's latest presentation at a major investment bank's 2026 autumn strategy conference, under the theme "Sowing for Autumn."

Where to begin with these evolving market dynamics?

After transitioning from "summer winds" to autumn market conditions, this economist's latest theme encapsulates both a fundamental judgment of the current market and a medium-term outlook for the global AI bull market. Autumn is harvest season, but winter won't be far behind. Compared to his bearish stance a month ago, he's clearly more positive on short-to-medium-term prospects.

Over the coming quarters, as AI applications deploy, large models continue competing, and AI-plus initiatives empower various industries, the AI bull market is expected to persist and broaden. However, from a long-term perspective, he's turning cautious, suggesting this global AI bull market may already be entering its later stages.

As the AI-driven capital expenditure cycle enters its second half, the bull market's logic will shift. In the early phase, the market believed most strongly in AI computing power scarcity—the supply constraints in optical communications, storage, and advanced manufacturing. In the second half, as AI infrastructure costs escalate without corresponding returns from applications, productivity gains, and real economy efficiency improvements, the AI narrative loses its sustainability.

Unlike previous years, AI's explosive technological growth has transformed it from merely a micro-level emerging industry concern into a critical macroeconomic variable. AI giants now borrowing heavily for capital expenditure compete with US Treasuries for demand, pushing long-end yields higher and eventually constraining AI development. Additionally, post-construction, electricity demand has emerged as a core constraint, with power generation equipment, transformers, grid connections, data center power upgrades, and energy storage potentially becoming hard limits on US AI expansion.

Why focus on just a few key sectors?

Once the AI bull market enters its later phase, the emphasis must shift to finding new fundamental logic—concentrating on AI application deployment and AI-plus initiatives that empower the real economy. Turning to the Chinese market, with risk-free rates at historic lows and equity implied returns still exceeding bond yields and rental income, domestic wealth allocation still favors the stock market.

More importantly, comprehensive capital market investment and financing reforms are opening another avenue. Over the next year, the elasticity of Chinese assets won't come solely from endogenous growth driven by AI applications, but also from exogenous expansion through mergers and acquisitions, asset injections, securitization, and equity-based fiscal policies.

Therefore, "sowing for autumn" means seeking more cost-effective new opportunities in the second half of the AI investment cycle and the later stages of the AI bull market. The TMT sector will experience divergence; crowded AI hardware directions require distinguishing the genuine from the false. What truly deserves attention are assets that align with AI application diffusion while offering earnings delivery, industry barriers, and attractive valuations.

The "AI-plus" theme is broadening, with particular promise in pharmaceuticals, energy technology, advanced manufacturing, non-ferrous resources, non-banking financials, and defense—established sectors blooming anew.

Don't be blinded by short-term earnings; focus on long-term growth rates

First, let's address whether the AI bull market can persist and whether new fundamental logic can provide support. The answer is yes, it can continue. From an investment standpoint, approximate correctness trumps precise error. Approximate correctness refers to major trends, while precise error often comes from over-reliance on short-term data—like quarterly operational metrics or micro-level observations—which can obscure the bigger picture. A broader perspective and macro-level vision are essential for trend-based investment decisions.

Many textbook pricing models serve as useful references, most notably the perpetual growth model, whose simplified version is the Gordon Growth Model. This model incorporates several factors: near-term cash flows in the numerator; risk-free rates, equity risk premiums, and terminal growth rates in the denominator. Many investors obsess over "next period" earnings, calculating future quarters meticulously, believing precise short-term profit forecasts predict stock movements. They don't—these are just one of four factors that are relatively easy to derive bottom-up.

Attention must also be paid to asset pricing, fundamentals, and industry logic, which don't always align perfectly; markets often bottom or top before fundamentals do. That's because asset pricing also incorporates risk-free rates, risk premiums, and long-term growth expectations. Following this logic, the focus shouldn't be on next quarter's cash flow or next year's earnings growth, but rather on the long-term growth rate—the trajectory of the AI industry's long-term trends.

According to the capital expenditure cycle, the AI bull market has entered "autumn"

Fundamental judgments are typically categorized by time horizon into inventory cycles, capital expenditure cycles, and long-wave cycles. Since the mobile internet era, inventory cycles in major economies have become less reliable, with low-amplitude fluctuations that no longer provide strong timing signals. Long-wave cycles spanning 60 years are too extended to be practical. For medium-to-long-term fundamental trends, the capital expenditure cycle—equipment replacement investment cycles typically lasting 8-10 years—has become increasingly important.

This current cycle bottomed and turned upward in early 2021, accelerated in 2023, and is expected to enter its second half around 2026, potentially peaking by 2028. Historically, stock markets lead fundamentals by roughly six months at both troughs and peaks. Based on this, the AI bull market's sustainability can be projected: trusting the capital expenditure cycle and recognizing this AI wave as an epic technological revolution, this cycle should extend at least to early 2028, meaning the AI bull market can continue at least into next year, with vigilance warranted around mid-next-year for potential peaks.

The conclusion is that the AI bull market is entering "autumn"—winter may not be distant, but investors shouldn't forfeit the harvest because of what's coming. Instead, they should actively prepare for the bounty.

AI applications are beginning to accelerate their spread, as large models enter an intense "life-or-death" competition

Four dimensions illustrate this AI bull market's autumn. First, AI applications now have the foundation for diffusion. Historically, this period resembles 1998 in the internet 1.0 era, when US personal internet penetration was around 30%. According to Stanford's AI research institute, by early 2026, US adult generative AI adoption had reached 28.3%, now exceeding 30%. On the enterprise side, US business AI adoption is approximately 20% based on Census surveys. As penetration rises, AI enters an accelerated expansion phase of application demand.

The next phase of AI applications will spread from AI coding tools like Anthropic's to broader domains. Of course, the most direct AI application is the large model itself. In the internet era, applications required software or tech companies to build business models from scratch. Now, models iterate continuously, becoming more capable with lower hallucination rates, enabling AI to empower countless industries.

The "AI-plus" era has arrived. Whether it's enterprise applications, industrial AI, government AI, or AI-driven drug development—including direct participation in tumor mutation screening and candidate antigen identification—the effects are increasingly evident. More than 30% of China's large-scale manufacturing enterprises have adopted AI technologies. Industry automation foundations directly determine AI deployment speed. Equipment manufacturing leads in AI investment share, while mining and chemicals lag. This indicates AI isn't confined to concepts but is progressing from pilot exploration to deeper industry applications.

From another angle, capital expenditure is shifting from infrastructure investment toward applied revenue realization and then to global productivity dividends. AI empowers existing business models to improve efficiency. According to a 2025 OECD report, AI could contribute 0.2 to 1.3 percentage points annually to G7 labor productivity growth over the next decade under different adoption scenarios, with the neutral case at 0.5-1 percentage point. With 2025 global GDP at approximately $118 trillion, a 1% AI-driven productivity gain corresponds to about $1.2 trillion in annual returns—and that's under neutral assumptions.

From an optimistic standpoint, AI capital expenditure does generate returns. From a pessimistic view, US AI capex is roughly $800 billion this year, with projections ranging from $1.3 trillion to $1.6 trillion next year, potentially reaching $2-3 trillion by 2028. If AI capital spending becomes too heavy while total returns remain comparatively small, the weakest link will break first. This means by 2028, the current massive AI investment path may become unsustainable.

By 2028, regardless of whether AI capex reaches $2 trillion or $3 trillion, the crucial test will be whether it delivers sufficient returns through total factor productivity improvements—can the math work? It seems challenging, but it remains unprovable today. From a one-year perspective, investors should maintain half-clarity and half-inebriation toward this AI bull market.

Even if AI capex returns don't add up, its short-term existence has rationality. A company doesn't necessarily need profits to survive—as long as labor costs are covered, machines keep running, and wages are paid, it continues operating rather than facing liquidation. In this cycle's second half, the most evident trend in AI applications is large models engaged in "mortal combat"—including Anthropic, OpenAI, and Chinese model developers—fighting relentlessly, potentially running massive losses until debt-financed capital spending becomes unviable.

Therefore, for at least the next one to one-and-a-half years, until large model competition subsides, AI application diffusion should sustain the AI ecosystem. Capital markets price the future ahead of time; industrial development and stock prices don't always move in lockstep.

Historical lessons and the macro cycle

History provides the best reference: the internet bull market. The internet experienced staged evolution—infrastructure first, then software and application diffusion, followed by enterprise process restructuring and productivity realization. From 1993 to 1999, US real investment in information processing equipment and software grew at an average annual rate of 20%, with communications equipment at 16%, as underlying infrastructure expanded rapidly. By 2000, internet applications extended to ERP, procurement, sales, and customer service operations. US retail e-commerce sales grew from about $15 billion in 1999 to roughly $28 billion in 2000—a 91% year-over-year increase—yet still under 1% of total retail sales, classic early-stage acceleration.

Correspondingly, Netscape, Yahoo, Amazon, and eBay went public between 1995 and 1998, with the Nasdaq peaking in March 2000. Capital markets consistently price the future ahead of time. The internet investment cycle extended into early 2001, but the market topped nearly a year earlier. When the Nasdaq peaked in 2000, many internet companies had already topped in 1999. However, internet application diffusion didn't end with the market peak.

History can be referenced from both industrial and macro perspectives. Once an emerging industry grows from infancy into an economic pillar influencing national finance, fiscal policy, and societal productivity, it's no longer just a micro-level issue—it must be understood through macro logic. The inescapable macro cycle ultimately asserts itself. Just as interest rate hikes pricked the tech bubble then, the macro cycle will ultimately end this AI bull market. But the conclusion of a tech bull market doesn't signal the end of the tech industry—industrial development and stock price rhythms are distinct. Indeed, from 2001 to 2007, US real estate-related assets rallied while the internet industry continued advancing; gaming diffusion and application expansion didn't stop.

Capital market pricing often compresses years of expectations into current valuations, with the most important short-term indicator being the equity risk premium. In June and July, when bearish views were expressed, global AI-related asset risk premiums were exceptionally low, and even now they remain at historical lows, corresponding to risk appetite at historical highs. As the macro cycle's gears grind forward, reverence for the eventual AI bubble burst is warranted. But there's no need to avoid food for fear of choking—this AI bull market will only establish its major peak after fully reflecting industrial trends. The key forward watchpoint remains AI application diffusion; only when AI capex and application deployment form a virtuous cycle, combining AI infrastructure with real economy efficiency gains, can the entire ecosystem continue.

On earnings, US corporate results still support the AI bull narrative, but upstream and downstream need recalibration

Second, from an earnings perspective, the four major cloud providers, semiconductor firms, and tech hardware companies continue posting high growth, supporting substantial capital expenditure short-term and providing the foundation for AI's expansion across industries. In Q2 2026, North American cloud provider revenues reached $116.2 billion combined, up 43% year-over-year. S&P 500 adjusted net income grew 39.5% year-over-year, with AI hardware remaining the strongest earnings contributor—semiconductors, technology hardware, and storage all maintaining high growth. Hyperscalers and emerging cloud providers continue expanding revenue, orders, and backlogs, indicating sustained computing demand and ongoing conversion of AI capex into industry revenue and profits.

Meanwhile, US earnings growth is broadening beyond the Magnificent Seven. S&P 500 earnings increments are spreading from a few tech giants to more companies and sectors. Since August, earnings expectations for real estate, healthcare, and industrials—non-TMT sectors—have also been revised upward notably. This may represent the early characteristics of the AI rally shifting from upstream computing leadership toward application diffusion.

However, the short-term and long-term logic differ. In the near term, computing giants' earnings growth appears unassailable, but long-term, this may not be positive for the overall AI ecosystem—it could be the high-altitude rocks that trigger glacial debris flows. Only when upstream and downstream recalibrate, and AI capex and applications re-divide the pie, will the ecosystem become healthier and more sustainable. If optical communications, storage, and computing costs keep rising, then as AI becomes a macro variable, the "brute force, spend regardless of returns" narrative will prove a dead end. Continuous AI hardware price increases won't help the large model arms race—they'll accelerate its conclusion.

Of course, the good news is that we don't know when the large model arms race will end. Until it does, model developers will continue investing like moths to a flame, striving to create omnipotent AGI and survive the competition. Anthropic's current behavior resembles humanity building the Tower of Babel. AI remains in a wild-growth phase; without proper rules and governance, the AI arms race may not be a blessing but could turn AI into Skynet. The internal competition among global model companies, from a profitability perspective, is no different from China's food delivery wars—ultimately producing ever-larger losses. The benefit is that model costs will keep falling, models will improve, and they can empower countless industries.

In early August, the economist published a report titled "Autumn Market: Different from Spring Light, Yet Surpassing Spring Light," quoting Mao Zedong's poetry. At this point, the focus should shift to new narratives beyond AI hardware—narratives about AI empowering established industries. Markets may have forgotten these "old guard" companies, but they're not without merit; traditional macro and industry logic issues are already priced into their low valuations.

In the late stages of the AI bull market, greater attention should be paid to established companies blooming anew—similar to a convertible bond strategy. The economist favors "old guard" companies with deep value and healthy cash flows that can develop second growth curves by leveraging the global model developers' arms race to enhance their business models. This represents the new opportunity from AI diffusion. The global model arms race gives these companies access to better, cheaper large models, or through M&A, they can acquire AI tech assets—making opportunities in pharmaceuticals, real estate, financials, and resources potentially more attractive than crowded tech hardware. We're already in the autumn of the AI bull market; we must operate in the present while preparing for the post-AI era.

Both China and the US are escalating their AI policy commitment, with the sector in a policy honeymoon period

Third, the policy dimension remains positive, suggesting the AI bull market's autumn seems manageable, as both China and the US are simultaneously intensifying AI policy with growing emphasis on industrialization. In China, in July, the President's keynote at the World AI Conference emphasized AI as a new engine for global economic growth and an accelerator for transitioning between old and new growth drivers, calling for comprehensive promotion of AI technological innovation, industrial development, and application scenarios. Subsequently, the NDRC and State Council have issued comprehensive AI policy packages. Over the next one to two years, "AI-plus" supporting measures will accelerate, driving China's intelligent economy forward.

The US is similar. In June, the President signed an executive order on advancing AI innovation and security, rejecting over-regulation that would stifle innovation, aiming to consolidate US global AI dominance through a permissive business and R&D environment. In mid-July, the White House launched the "Golden Eagle" initiative, an AI-driven cybersecurity information-sharing platform. From a policy standpoint, both nations treat AI as the primary arena for technological competition, doing their utmost to protect AI industrial development.

Given recent rises in long-end yields, many investors worry whether rising rates might prick the AI bubble as they did the internet bubble. With both China and the US ramping up AI policy, AI remains in a policy honeymoon period. At least short-to-medium term, there's little concern about long-end Treasury yields quickly pricking the AI bubble—US officials are more anxious than the market to suppress rates. This is indeed a long-term risk, as US debt, inflation, and AI bubble risks are fundamentally irreconcilable over time. Especially by 2028, as US elections approach, certain left-wing politicians' anti-AI and anti-capital tendencies may intensify, creating genuine policy pressure and negative shocks.

Aggregate demand still supports AI, but beware of a 2000-style story repeating a year from now

Fourth, aggregate demand. If both Chinese and US economies face major problems, can AI persist? Can industry logic continue? Can the stock bull market survive? These become significant question marks.

Looking at the US, AI investment has become a pillar of economic resilience. AI investment remains elevated while traditional investment decelerates. AI-related sectors—power, computing, communications, computers—are the brightest spots in the US economy. With AI investment persistently high, US economic resilience and momentum are actually increasing, and future AI applications will drive efficiency gains across industries. According to IMF forecasts, 2026 US growth of 2.3% may be the strongest among developed economies. Of course, there's "joy tinged with worry." AI stands truly exceptional; otherwise, traditional US investment remains subdued. Under high rates, both residential and non-residential investment have retreated from peaks.

Another issue: while inflation risks won't spiral out of control, core inflation still hasn't fallen below the Fed's target. So, with AI overheating and inflation not conducive to significant Fed easing, US tech giants borrowing heavily for capex squeezes Treasury demand, constraining US rates. In summary, short-term overseas aggregate demand still supports AI, but a year from now, beware of repeating 2000.

For China, aggregate demand concerns are less pressing. China is in a critical period of transitioning growth drivers and transforming its development model—seeking progress while maintaining stability and consolidating fundamentals is the norm, with emphasis on quality and efficiency. Rather than magnifying short-term macro "troika" data, focus on technological innovation quality, industrial capability, and sustainable economic development. The key is the new growth engine—new quality productive forces—with "AI-plus" as the core.

Since 2018, high-tech industry value-added growth has outpaced manufacturing overall. In July 2026, high-tech industrial value-added rose 16.9% year-over-year, notably exceeding overall manufacturing. Export momentum is also shifting; since 2021, new growth engine exports have accumulated 186% growth on a 12-month rolling basis, significantly outpacing traditional electromechanical, consumer electronics, and old-economy manufacturing. Starting in July, government bond, policy bank bond, and municipal investment bond issuance accelerated, with fiscal and monetary policy providing additional support. Policy direction isn't relying on aggregate stimulus for stability but uses new quality productive forces as the lever, combining long-term competitiveness enhancement with short-term stability.

This is the fundamental assessment of China's fundamentals. Under this framework, there will be increased integration of technological and industrial innovation, deeper implementation of "AI-plus," and development of new intelligent economic forms. Real estate policy remains more about "supporting without lifting," treating real estate-related chains as bond-like allocations while seeking M&A and rejuvenation opportunities.

From aggregate demand, this AI wave's continuity in China may outlast the US. First, high-tech gap-filling—addressing shortages like computing power and advanced manufacturing—will sustain momentum longer. Second, "AI-plus" initiatives benefit from China's unparalleled capacity to marshal social resources, channeling wealth toward "AI-plus" opportunities. Before the global AI investment cycle concludes, "AI-plus" holds substantial promise. Even when the US cycle ends, China may maintain momentum due to remaining gaps—historically evident when the Nasdaq peaked in March 2000, China's bull market lasted until June 2001, an additional year and a half.

Overseas funding conditions: crisis amid opportunity, with long-end Treasury yields likely to rise then fall in H2

The second section addresses funding conditions. Short-term, there's crisis amid opportunity. Rising long-end Treasury yields have become a source of volatility for global risk assets in August and September. The 30-year Treasury yield has risen to 5.27%, with the 10-year around 4.74%. The Treasury market is becoming the core cross-asset pricing risk source; further yield increases could pressure global risk assets further. The core reasons: market distrust of long-dated Treasuries and concerns about long-term US credit, keeping real rates elevated and term premiums rising. Additionally, someone is "competing" with Treasuries—AI giants are borrowing aggressively. Hyperscalers are accelerating bond issuance to fund capex, extending into longer tenors, increasing duration supply. With long-dated Treasury supply already elevated, AI corporate bonds and government debt combine to push up financing costs and long-end yields simultaneously. Tech giants' newest dollar bond curves have shifted upward, with long-dated coupons generally rising to around 6%. Accelerating capex and rising financing costs are squeezing tech giants' free cash flow.

However, in Q3, long-end Treasury yields are likely to peak and retreat—up first, then down. The 10-year could briefly reach around 5% before settling to ~4.3% by Q4. Treasury efforts like helping Japan intervene in yen or short-dated borrowing and buybacks are minor measures that can't suppress long-end yields. What's truly required is the Fed and Treasury cooperating to lower long-end rates—that's the essential path. Currently, the Fed remains hawkish, so long-end yields may spike further, possibly before September's FOMC meeting. Ultimately, the Fed must step in; it's the key force. Subsequently, the US may employ some form of financial innovation to help suppress long-end Treasury yields.

China's funding picture is more positive, with equities remaining the preferred wealth allocation. From a funding perspective, China's conditions are favorable, with risk-free rates at historic lows. Comparing domestic asset classes, the inverse of the PE ratio for all A-shares—the implied return—is approximately 4.40%, notably higher than the 10-year Chinese government bond yield of 1.86%. Even with a two-times risk premium adjustment (~3.37%), equities remain more attractive. Equity implied returns also exceed rental yields; the national 100-city residential rental yield is only about 2.48%. Thus, China faces an effective asset shortage, with stocks the preferred wealth allocation—benefiting both A-shares and Hong Kong-listed equities.

Globally, MSCI China's 12-month forward PE still trades at a significant discount to the S&P 500. Once long-end Treasury yields follow the expected pattern, Chinese equities' allocative appeal should continue rising. Sub-segment analysis—insurance allocation needs, RMB appreciation-driven foreign demand, and "invisible hand" allocations from entities like Central Huijin—supports a relatively positive outlook for Chinese equities' funding over coming quarters.

A new development deserves attention. Long-term, steadfast pursuit of a Chinese-characteristic financial development path is essential. Short-to-medium term, changes in China's funding or policy landscape must be watched closely. The July Politburo meeting statement highlighted deepening capital market investment and financing reforms to enhance resilience and confidence. On one hand, the practical short-term objective is nurturing new quality productive forces and increasing the market's "tech content"—a goal for both A-shares and Hong Kong. On the other, capital markets' political, people-oriented, and functional nature must help with "debt resolution" while enhancing resilience and confidence—referencing the 2000-2001 bull market that served state enterprise reform.

The CSRC has placed refinancing, M&A, REITs, securitization, PE/VC exits, and long-term capital market entry into the same reform framework. July introduced shelf registration and other refinancing reforms to facilitate listed companies' equity financing. It's reasonable to believe that in the coming year of the AI bull market's "autumn," both A-shares and Hong Kong will increasingly empower China's new quality productive forces development. Tech IPOs will dominate both markets, and tech's weight across indices will continue rising—becoming the new normal.

Valuation-wise, A-shares and Hong Kong have already risen considerably from three years ago; they're no longer absolute bargains but reasonably undervalued—this alone no longer justifies further upside. In coming quarters, both markets will reside in the AI bull market's "autumn," with upside awaiting earnings delivery. Earnings can come two ways: organically or exogenously. Organic delivery applies to AI applications, gap-filling high-tech sectors, and industries where China has global advantages—these maintain prosperity with upward earnings elasticity. Additionally, in a stable-but-progressing economy, external growth deserves more attention for non-tech sectors. The next phase should focus on M&A, asset injections, government tenders, or institutional orders that could quickly improve non-tech or "old guard" companies' income statements.

In this environment, Chinese listed companies seeking excess earnings elasticity—drawing lessons from 2000-2001—should focus on M&A, asset injections, and securitization over the next year. Particularly in Hong Kong, many value stocks are cheap, with negatives largely priced in. These cash-generative value companies can acquire domestic or overseas AI technology assets via asset injections or restructuring, creating earnings elasticity—similar to convertible bond investing. Parallel research treats coming quarters as analogous to the bull market from 2000 to June 2001.

Why did A-shares continue rising for about a year after the internet bubble burst in Q1 2000? That bull market, superficially about the internet, was actually tied to late-1990s state enterprise reform. The core issue was 1997-2000 state enterprise losses and high leverage creating banking system non-performing assets. That reform was essentially a complete balance sheet restructuring. On the asset side: M&A, asset swaps, shareholding reforms, and equity transfers concentrated state assets into more efficient enterprises and listed platforms. The 1999 Fourth Plenum called for strategically adjusting the state economy and promoting strategic restructuring. The 2001 CSRC notice standardized major asset purchases, sales, and swaps, requiring they benefit sustainable development and shareholder interests. On the liability side: four AMCs were established to absorb and dispose of bank NPLs, with large-scale debt-for-equity swaps. By December 2000, 580 enterprises had converted debt to equity totaling 405 billion yuan, reducing their debt ratios from above 70% to below 50%. On governance: debt-equity conversion and equity restructuring drove modern enterprise systems and governance reforms. The 2002 acquisition management measures further facilitated control transfers through negotiated offers, tender offers, or exchange-traded bids, promoting securities market resource optimization.

Ultimately, of the 6,599 large state-owned loss-making enterprises at end-1997, 66.5% had escaped difficulty by end-2000. By 2001, state enterprise reform had succeeded, China joined the WTO, and the bull market ended. Today's macro environment and market size differ vastly, but the underlying debt-resolution logic shares similarities—only now it's about resolving local government debt rather than state enterprise debt. Drawing historical lessons, strategic M&A, asset swaps and injections, debt-equity swaps, and equity/control restructuring can all drive Chinese stock market performance over the next year.

This enables local governments' new development approaches—the Hefei model—to yield returns, creating a virtuous cycle between capital markets and local fiscal balance. Specifically, before 2028, local government-affiliated PE, industrial investment, fund-of-funds, and VC institutions can leverage this AI bull market and capital market reforms to IPO or inject the substantial tech equity positions acquired at low valuations between 2021 and 2023, revitalizing and revaluing these assets, rapidly improving local government balance sheets. Trust this external growth opportunity early; discern the new trend of capital markets serving the real economy. As the ancient saying goes, "to take, one must first give"—during this securitization process, policy support will persist, and smart economy and "six networks" related orders are likely to consistently exceed expectations.

In coming quarters, as the AI bull market continues, focus shouldn't be limited to endogenous growth from the global AI cycle's second half, but also on the external growth momentum from China's investment-financing reforms. Embrace one new logic with two main lines: TMT differentiation and non-tech diffusion

Finally, investment strategy and allocation recommendations: embrace one new logic with two main lines. The new logic is that the AI bull market has entered its later phase where, in autumn, opportunities will be abundant—AI applications will proliferate, requiring attention to both the original tech theme and the "non-tech" theme. The "autumn" new logic is applications-led.

The two main lines: TMT, characterized by differentiation, and non-tech, characterized by diffusion—established sectors blooming anew.

TMT differentiation spans from semiconductors, storage, and optical communications toward AI applications. Tech hardware should focus on China's "gap-filling" areas and fields where China holds absolute overseas advantages. For computing power, semiconductors—particularly equipment and manufacturing—the focus is on domestic supply chains. Advanced manufacturing enters a new phase; mature processes remain the primary battleground for import substitution. Foundry operations exhibit "full capacity, rising prices" logic, with AI expansion and domestic storage capacity growth supporting record equipment market levels. Strategically, avoid overcrowded, crowded tech hardware areas.

Optical communications fundamentals remain robust, with AI computing demand sustaining high prosperity; 1.6T is accelerating, with 2026-2027 as the profit realization period, though mid-term US policy and competitive dynamics may impact. Opportunities exist but not indiscriminately—focus on truly defensible, profitable, supply-demand advantaged companies, and pay attention to ownership structure; Hong Kong and US listings offer relatively better positioning. Storage benefits from AI demand underpinning elevated conditions, though price gains may moderate toward a high-level plateau—not an immediate cyclical reversal, but markets may adjust storage valuations in advance.

It's believed that 70% of Asia-Pacific optical module and storage stocks, driven by overcrowded trades, leverage, and quant strategies, likely peaked in late June or early July, with the remaining 30% potentially achieving new highs via earnings and application expansion. Based on medium-term logic, AI tech hardware will narrow while applications broaden—favor AI tech stocks with clear industry structures, high tech content, strong competitive barriers, and better alignment between ownership structure and funding.

The non-tech line follows "AI-plus" diffusion beyond AI technology. Quality non-tech assets see established companies bloom anew through both organic and external growth, prompting revaluation of these "convertible-bond-like" assets. As AI applications diffuse, benefits spread first to pharmaceuticals—especially innovative drugs—then, with China-US "AI-plus" progress, to global expansion chains, particularly large model and token exports, plus power equipment, robotics, high-end manufacturing, and chemical new materials. Concurrently, select "AI-plus" opportunities in domestic demand sectors—resources, energy, finance, defense-related AI applications—and position early in cash-flow-healthy local SOEs for potential "AI-plus" asset injections and M&A surprises.

Finally, to sustain the AI industry and bull market, efforts to suppress US Treasury rates will ultimately enhance gold, digital assets, and fintech's allocative value.

Key sector outlooks: biopharma, advanced manufacturing, energy technology, and traditional "old guard" assets

In biopharma, the focus is on CXO, innovative drug chains, and innovative technology platforms. CXO external demand is robust with strong global competitive advantages—leading companies continue performing well with reasonable valuations. Innovative drug chains benefit from strong external demand, with out-licensing entering a harvest period. The biopharma sector could see endogenous profitability inflection points in 2026-2027 as globally competitive products enter overseas commercialization. Innovative technology platforms—AI-driven drug developers listing in Hong Kong—make 2026 a critical year for platform value verification. These aren't simple single-pipeline stories but "platform valuation plus pipeline options" dual drivers, becoming a relatively scarce thematic sector in Hong Kong.

Advanced manufacturing includes commercial aerospace and robotics, where alpha can be found. Commercial aerospace sees demand explosion, rapid downstream application growth, communications satellite bursts, and computing satellite follow-through. The upstream bottleneck is rocket launch capacity. Domestic and international technology continues breaking through, with launch density rising; both China and the US have multiple new medium-to-heavy-lift rockets scheduled for first flights or recovery verification in H2. Prior corrections have substantially released risk; the next rally trigger may come from verifiable technological milestones or flagship listings. Robotics is at the beginning of mass production, with scenario verification determining winners. Short-term, there's a game between high-frequency catalysts and valuation digestion. Humanoid robot production is clearly accelerating; however, the market has substantially priced in shipment and volume expectations, so guidance upgrades alone cannot sustain valuation expansion. Future excess returns hinge on scenario verification—whether commercial closed loops achieve endogenous breakthroughs in real application settings.

Energy technology deserves serious attention in the AI bull market's late stage and the post-AI era. Its development logic is shifting from power generation-side structural optimization toward AI-driven demand-side restructuring. Previously, energy was viewed through generation-mix adjustment; AI's massive construction has made electricity demand the new core constraint. Energy is both AI's foundation and bottleneck. AI demands greater energy stability; high-reliability generation and grid equipment may face extended supply tightness. Traditional data center power architectures are mature, but AIDC's high-power, high-consumption, high-volatility characteristics necessitate innovation and upgrades in power architecture and components, plus energy storage.

In consumption, the core business currently offers only rebounding opportunities from decline and bond-like allocation opportunities; watch for local SOE M&A of AI assets for external growth. Since late July, Hong Kong consumption high-frequency data has improved marginally, supporting valuation recovery and stock gains, though the medium-term narrative hasn't reversed. Gradually prepare for the post-AI era; genuine consumption stock turning points may not come from AI empowerment but from the AI bubble bursting—as in 2002-2007. To paraphrase the film "Let the Bullets Fly," consumption might tell AI, "Without you, things would be better."

Among traditional "old guard" assets, watch for renewed relevance and changes in competitive dynamics. The most favored "old guard" assets are safe havens—first, resources: gold, copper, rare earths, strategic minor metals. Recent Treasury yield rises alongside gold price increases suggest gold's pricing logic is shifting. Historically, gold was priced by short-term opportunity costs, with the real yield as the core variable. This round, rising Treasury yields reflect not just Fed policy but fiscal deficit expansion and excessive debt issuance. As Treasuries are no longer universally viewed as risk-free, gold—as the ultimate monetary equivalent and sovereign credit alternative—undergoes sustained, strategic repricing. Global central bank de-dollarization and structural buying provide medium-to-long-term support. Since August 2022, emerging nations have consistently added gold reserves, driving prices upward. Concurrently, US concerns about the AI bubble diminish Treasury appeal, pushing some investors toward gold. Gold's pricing dominance has shifted from traditional short-term opportunity cost to long-term sovereign credit risk and global asset allocation restructuring. Therefore, gold still has the opportunity to reach $5,000 per ounce this year. Strategically, 2027 and 2028 offer substantial further upside.

Second, over the next one to two years, global "old guard" assets favored are defense-related technology. On one hand, they benefit from great-power competition; on the other, AI empowers military applications.

Finally, fintech. Traditional finance empowered by AI—brokers have the highest AI content. Additionally, as regulatory clarity increases for Chinese cross-border internet brokers, Chinese securities firms' second growth curve is accelerating overseas. Leading firms are deepening Hong Kong and Singapore presence across user scale and client AUM dimensions; global expansion has become a medium-to-long-term growth trajectory. Meanwhile, fintech and digital assets will markedly benefit from global financial order restructuring—with logic similar to gold.

In summary, this "sowing for autumn" requires half-clarity and half-inebriation, cherishing the bull market's autumn. Don't abandon the autumn harvest simply because winter will eventually come.