In August 2026, the capitalization of hard technology in China was intensively realized: Changxin Technology listed on the sci-tech innovation board, its market value once surged to nearly 30 trillion yuan, while Yushu Technology issued a market value of about 61 billion yuan and was priced at over 200 times its earnings, about to begin subscriptions. These two companies represent breakthroughs in domestic storage and humanoid robotics respectively. Placing them back into the global coordinates — comparing with domestic SMIC, Cambricon, and Haiguang Information, as well as the American stock market giants NVIDIA, TSMC, the three major storage companies, and robotics players — investors can make clearer judgments: how far can global tech stocks go, how much bubble exists, and which signals to watch for.

Changxin Technology: Ten Years of Losses for Cyclical Resilience

Zhu Yiming left Zhaoyi Innovation and devoted himself full-time, maintaining a zero salary until turning profitable in 2025, with a core philosophy of respecting storage cycles and showing patience in next-generation research and manufacturing.

In operations, it actively shrank its low-end DDR4, fully shifted to DDR5 and LPDDR5X, advanced production capacity in Hefei and Beijing, and extended to Shanghai, aiming for a monthly output of about 350,000 pieces by the end of 2026, while initiating HBM sample testing and signing long-term supply agreements with Tencent and ByteDance.

• In 2025, revenue is estimated at about 61.8 billion yuan, with a net profit attributable to parent company of 1.875 billion yuan.
• In the first quarter of 2026, revenue reached 50.8 billion yuan, with a net profit attributable to parent company of 24.762 billion yuan.
• For the first half of the year, revenue is expected to be between 110 billion to 120 billion yuan.

It has fully understood the general DRAM shortage driven by AI servers, but profits highly depend on rising average selling prices, while HBM is still in the catch-up stage.

Yushu Technology: Real Sales, High Margins, But Growth Shift and High Valuation Coexist

Wang Xingxing started from a student project, adhering to “being able to work” rather than showing off technology, emphasizing that athletic ability is a prerequisite, predicting that the “ChatGPT moment” for embodied intelligence will arrive in two to three years — robots can accomplish about 80% of tasks in 80% unfamiliar scenarios on instructions alone.

Yushu Technology started with four-legged robotic dogs and quickly shifted to humanoid robots, with humanoid shipments exceeding 5,500 units in 2025 (leading globally), and cumulative four-legged sales exceeding 30,000 units. Its financial performance is rare in the embodied intelligence track:

• In 2023: Revenue was about 159 million yuan, net profit loss of about 11 million yuan (non-recurring loss of about 18 million yuan).
• In 2024: Revenue was about 393 million yuan, turning profitable (net profit of about 95 million yuan, non-recurring about 78 million yuan).
• In 2025: Revenue was about 1.7 billion yuan (up approximately 335%), non-recurring net profit attributable to the parent company around 591-600 million yuan, with main business gross margin rising to about 60%.
• In the first quarter of 2026: Revenue was 423 million yuan (up 68%), but non-recurring net profit was about 40 million yuan (down 52.5% year-on-year); for the first half of the year, revenue is expected to be 1.052 to 1.128 billion yuan (up 35% to 45%), but non-recurring net profit may still slightly decline year-on-year.

The revenue structure has switched: humanoid robots contributed more than half of the revenue in 2025, with a high proportion from research and education scenarios. High margins come from vertical integration (self-developed motors, reducers, etc.) and cost control. The IPO plans to raise about 6.1 billion yuan (exceeding original plans), with about 85% allocated to R&D (embodied large models, ontology, etc.), reflecting the “mass production first, then iteration” approach.

In comparison with UBTECH, the contrast is more pronounced: UBTECH's 2025 revenue is 2 billion yuan (up 53%), with full-size humanoid revenue skyrocketing over 22 times to 821 million yuan, yet still posting an annual net loss of about 790 million yuan, with a gross margin of about 37.7%. Yushu has already achieved scaled profitability and positive operating cash flow, but growth rates have fallen from triple digits, and expenses (especially R&D and marketing) are rising rapidly, putting pressure on profits.

It has created early barriers through real deliveries, but still needs to prove the industrial and household scene closed-loop after the model breakthrough.

Domestic Hard Technology Comparison

SMIC, as the leader in wafer foundry, has an estimated revenue of about 9.3 billion dollars (about 67 billion yuan) in 2025, with net profit attributable to parent company of about 5 billion yuan, capacity utilization rebounding to over 90%, and a gross margin of about 21% to 22%, which is growing more steadily but with limited elasticity, resembling a mature infrastructure.

Cambricon is expected to have a revenue of about 6.5 billion yuan in 2025, turning a profit with net profit attributable to the parent company of about 2.06 billion yuan, primarily focusing on self-developed NPUs, but with high inventory, customer concentration, yield rates, and capacity of advanced processes still as bottlenecks.

Haiguang Information is on track to have 14.4 billion yuan in revenue in 2025 and a non-recurring net profit of about 2.3 billion yuan, relying on x86 licensing and DCU dual lines, seeing solid growth in trust creation and AI inference servers.

UBTECH is expected to achieve revenue of 2 billion yuan in 2025, still posting a loss of about 790 million yuan; humanoid income has surged, but the commercialization path relies more heavily on service and educational scenarios.
Overall, domestic hard technology has transitioned from “burning money to catch up” to “partial profitability + dual drivers of policy and domestic replacement,” but valuations are generally overstretched, and the gaps in technological iterations and limited supply chains are common constraints.

US Stocks and Global Leaders: Higher Barriers and Stronger Pricing Power

NVIDIA has almost defined this round of AI hardware dividends: expected revenues of about 216 billion dollars in the 2026 fiscal year, with net profits around 120 billion dollars, data center business accounting for a very high proportion, with gross margins maintained over 70%; in the first quarter of the 2027 fiscal year, revenue already reached about 81.6 billion dollars. Huang Renxun's philosophy is “accelerating platformization of computing,” continuously iterating architecture, binding ultra-large-scale cloud manufacturers with sovereign AI demands, with the moat lying in software ecology and system-level optimization.

TSMC, as the absolute leader in advanced process foundry, has an estimated revenue of about 122.4 billion dollars and a net profit of about 55.2 billion dollars, with a quarterly revenue of about 40.2 billion dollars in the second quarter of 2026, and a net profit margin exceeding 55%; advanced processes (7nm and below) account for more than 70%, with 2nm already in mass production, and capital expenditures increased to 60 to 64 billion dollars. Its logic lies in “technological differentiation + capacity discipline.”

The three major storage companies directly compete with Changxin:

• SK Hynix leads in the HBM field, with HBM market share exceeding 50%, and operating profit margins once exceeding 70%
• Samsung Electronics' DRAM market share has rebounded to about 39%, HBM is accelerating catch-up and starting HBM4 mass production
• Micron, through expansion in the US and long-term contracts, has occasionally surpassed 40 billion dollars in revenue in some quarters, with gross margins soaring over 80%

All three prioritize the highest quality capacity for HBM, with supply of general DRAM being relatively restrained, thus pushing prices higher.

In the robotics track, US stocks focus more on the long-term narrative. Tesla's Optimus is still primarily focused on internal learning and low-volume production, with Elon Musk frequently lowering short-term targets, emphasizing that scaling production is the biggest challenge. Figure AI, valued at about 39 billion dollars, leads pure humanoid companies, having completed over 1,000 hours of real production line operations in BMW's factory, but its commercialization scale remains limited. In contrast, Yushu has achieved scaled shipments and profitability but still needs to catch up in models and high-end industrial scenarios.

Commonalities and Differences

The commonality is that AI computing power and data center demands remain the core engine — training and inference are driving the consumption of advanced logic chips, HBM, and general DRAM.

The differences lie in: US stock leaders have higher technical barriers, stronger pricing power, and more mature customer ecosystems, with profit quality and cash flow far surpassing most domestic companies; domestic targets benefit from domestic replacement and policy windows, having greater elasticity, but face higher risks in iteration gaps, yield rates, supply chains, and geopolitical risks. In storage, Changxin relies on the boom of general DRAM cycles, while the three giants earn excess profits from HBM; in robotics, Yushu proves the feasibility of commercialization through deliveries and gross margins, while Tesla and Figure use capital and model narratives to price the long-term.

How much bubble exists?

Current valuations are partially overstretched. NVIDIA and TSMC's high growth is still supported by demand and orders, but capital expenditure competition and rising customer concentration; the super cycle of storage prices is unsustainable. Once supply catches up or AI capital spending slows, ASP declines will quickly compress profits; humanoid robots are highly valued; more on the assumption that “the critical point is approaching,” and if model generalization delays or scene implementation is slower than expected, there is considerable room for pullback. Domestic companies generally have “policy premium + narrative premium”; if performance realization is less than expected or external friction intensifies, fluctuations will be more severe.
This is not a comprehensive bubble but rather a structural overheating — companies with real moats that can traverse cycles still have room, while those entirely conceptual or highly dependent on a single cycle face higher risks.

Signals Investors Need to Watch

• Demand Side: Capital expenditure guidance from ultra-large-scale cloud manufacturers and sovereign AI, AI server shipments and utilization rates, visibility of HBM and advanced process orders.
• Supply Side: Storage ASP and inventory turnover, wafer fab capacity utilization and execution of capital expenditures, actual deployment hours of robots and customer renewal situation.
• Technical Realization: HBM mass production progress and yield rates, advancements in advanced process nodes, success rates of embodied large models in unfamiliar scenarios.
• Macro and Geopolitical: Changes in export controls, trade frictions, and the impact of interest rates and liquidity on high-valuation assets.
• Company Itself: Whether operational cash flow continues to improve, R&D conversion efficiency, customer concentration and long-term contract proportions, interests and governance quality of founders/management.

How far global tech stocks can go depends on the speed at which AI shifts from training booms to broader inference, agency, and physical world applications, as well as whether companies can convert short-term cyclical dividends into long-term technological and customer moats. Changxin and Yushu demonstrate that China's hard technology can achieve scaled profitability and real deliveries, while NVIDIA and TSMC show that top players are still expanding their advantages.

For investors, a holistic perspective means: do not just look at the narrative of a single company or market, but track whether demand is real, supply is uncontrollable, technology is realized, and whether valuations leave a safety margin. Long-term holdings are suitable for funds that truly understand industry rules, can endure volatility, and gradually allocate capital; short-term chasing is likely to be harmed at the cycle turning point or return of narrative tides.

No matter when and where, always remind: the market carries risks, and investment should be cautious.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。