For the past two years, the AI narrative has powered one of the strongest rallies in global capital markets. But as capital expenditures approach the trillion-dollar scale, market patience is wearing thin—investors are no longer satisfied with hearing stories; they want to see results. At today's AI Investment Summit, how and when this round of AI investment will translate into real returns became a key concern for economists, investors, and industry participants alike.
As a leading global AI infrastructure company, Lenovo Group is on the front line of this AI infrastructure expansion and value delivery, and it stands as one of the few companies able to answer with real numbers. "The explosion of the AI industry is just beginning," was a recurring judgment shared by macroeconomists, capital market experts, and industry insiders at the AI Investment Summit hosted by Sina Finance on September 16. However, "just beginning" does not mean all AI assets will continue to ride the same upward curve. On the contrary, as computing power investments are counted in the hundreds of billions of dollars and model iterations are measured in months or even weeks, the AI industry is rapidly moving from single-point technological breakthroughs to systemic competition. The capital market's focus is also shifting from the sustainability of the AI story to the ultimate delivery of growth.
Latest IDC data shows that in the second quarter of 2026, Lenovo's x86 server shipments reached 286,000 units, up 45.6% year-over-year, leaping to the top spot globally. In terms of revenue, Lenovo's x86 server revenue surged 96% year-over-year to $8.26 billion, trailing the number-one ranked Dell by just 0.6 percentage points. Additionally, Lenovo's latest quarterly results showed that its Infrastructure Solutions Group (ISG) revenue grew 98% year-over-year to RMB 57.9 billion, with operating margins hitting a record high of 9.1%. From shipments and orders to revenue and profits, the accelerating conversion of AI infrastructure demand into real business results also makes Lenovo a representative sample for observing the delivery of this round of AI investment.
AI Industry Boom Faces Return-on-Investment Scrutiny
What kind of new economy is AI actually creating? Wang Wensheng, practicing professor at Shanghai Advanced Institute of Finance (SAIF) at Shanghai Jiao Tong University and chief economist of SAIF Think Tank, analyzed from the perspective of economies of scale, stating that large models have achieved the "scaled production of cognition" for human society, which is revolutionary. But unlike the asset-light digital economy of the past, AI large models exhibit prominent asset-heavy characteristics, as improving model performance requires continuous investment in computing power, data, and electricity. This means AI is pushing the digital economy into an industrial phase with significantly front-loaded capital expenditures.
Wang pointed out that compared to internet platforms, large models, as asset-heavy technological tools, have non-zero variable costs and relatively weaker network effects. Some of the returns they generate will spill over as productivity dividends for society through knowledge and technology externalities. For investors, whether massive capital expenditures can ultimately yield expected returns depends on how AI creates economic value and how that value is ultimately distributed.
From an investment perspective, Wu Zequan, member of the Party Committee and Executive Committee of CSC Financial and president of the Future Industries and Policy Research Institute, noted that compared to the mobile internet era, AI capital expenditures have already jumped by an order of magnitude. Data shows that in the first quarter of 2026, the combined quarterly capital expenditures of four major North American cloud providers—AWS, Microsoft, Google, and Meta—reached $131.6 billion, up 70.25% year-over-year, with full-year total capex projected at approximately $710 billion. AI servers, GPU clusters, data centers, power, and network infrastructure continue to expand. However, Wu does not believe the AI industry itself is in a bubble; what truly warrants attention is the gap between underlying capital expenditures and closed-loop commercial applications amid rapid technological development. Therefore, in her view, as long as models continue to iterate and expand, computing power demand will remain highly certain, and whether this round of intensive investment can be sustained ultimately depends on whether applications can generate sufficient commercial value.
Liu Yuhui, a council member of the China Chief Economist Forum, shifted the focus back to the financial cycle. He noted that from large language models and chains of thought to agents and continuous learning, AI is approaching a self-iterating "intelligence singularity," which is precisely the most capital-intensive phase of the entire AI process. Meanwhile, this technological sprint coincides with a period of high funding costs in the dollar monetary and financial cycle. Beyond free cash flow, AI giants are increasingly relying on debt and off-balance-sheet financing to support massive investments. Liu believes that despite high capital costs, financial capital continues to bet on AI technological breakthroughs, meaning the AI technology cycle is becoming more deeply intertwined with the financial cycle. This implies that the certainty of an industry trend does not naturally equate to the certainty of every AI asset. As capital expenditures enter the trillion-dollar scale, the core question of AI investment is further narrowing to commercial implementation and delivery.
Orders and Profits Surge: Global AI Infrastructure Leader Delivers Value
In a fireside chat titled "Global Pricing of AI Infrastructure: Finding Certain Assets in AI Investment," Wu Yi, co-head of China Research and chief China strategist at BofA Global Research's Asia-Pacific strategy team, directly posed the question of AI investment delivery to Lenovo's Senior Vice President and Chief Financial Officer Wong Wai Ming: Is global AI infrastructure capex still in its early stages, or has it already become locally overheated? And how can companies convert massive investments into real orders, profits, and sustainable shareholder returns?
Wong Wai Ming emphasized on-site that Lenovo will continue to balance growth and profitability, with long-term net margins heading toward 5% and beyond, even 8% or higher. His response was first grounded in the company's front-line orders. From over RMB 140 billion in the fourth quarter of fiscal 2025/26 to more than RMB 360 billion in the first quarter of fiscal 2026/27, Lenovo's AI server order backlog has continued to grow rapidly. In Wong's view, while short-term markets may debate cycles and financing models, overall demand remains long-term. As the world gradually builds "AI factories" and "computing power factories," infrastructure will remain a key direction for long-term AI industry investment. The next major growth driver comes from enterprise-level inference. Wong stated that large international corporations are still in the early stages of AI implementation, needing first to address internal data, organizational structures, and business process integration—especially for companies that have gone through mergers and acquisitions, where integrating disparate systems takes time. But as these foundational tasks are completed, there will be substantial inference demand in the future. "This is the opportunity for AI over the next decade, and we are still at the very beginning."
Beyond demand growth, the capital market ultimately focuses on profitability. In fact, at the Investor Day held on June 25, Lenovo already laid out a clear medium-to-long-term profitability roadmap: revenue target of $100 billion with a net margin target above 3% in the next one to two years; revenue target of $130 billion with net margin rising above 5% over three to five years; and beyond five years, annual revenue target of $150 billion with net margin planned to reach 8% or more. This means that while AI infrastructure scales rapidly, sustained profitability improvement has also become a key component of Lenovo's medium- and long-term growth targets. At this summit, when asked whether Lenovo's goal of reaching a 5% net margin could be achieved ahead of schedule, Wong reiterated that the company will continue to balance growth and profitability, with long-term net margins moving toward 5% and even 8% or above.
In the AI infrastructure space, Lenovo still has considerable room for customer expansion. What's more noteworthy is that this scale growth is happening alongside profitability improvements. The latest IDC data shows that in the second quarter of 2026, Lenovo's x86 server shipments have already ranked first globally. In the most recent quarter, ISG revenue grew 98% year-over-year to a record RMB 57.9 billion, with operating margins hitting an all-time high of 9.1%. AI server order backlog reached RMB 360 billion, more than doubling from the previous quarter. Discussing Lenovo's long-term valuation, Wong candidly stated that compared to Dell, the gap in revenue and net profit between the two companies is far smaller than the valuation gap. "Our current valuation is roughly 1/7 of Dell's, but our revenue and net profit are not seven times smaller. In this regard, we are completely undervalued by the market." Wong also emphasized that the growth opportunities Lenovo currently sees are backed by real orders, and there are numerous customer projects under negotiation. Meanwhile, there is further optimization potential in customer expansion, pricing, storage, and services. In this sense, the pricing of AI infrastructure assets is entering a new phase: orders prove demand, margins prove the business model, and cash flow and sustained returns ultimately determine how much of the industry's prosperity can be converted into corporate value.
From "Stacking Chips" to Systemic Competition, Anchoring AI's Long-Term Certainty
As AI demand continues to grow, the infrastructure itself is also changing. In a roundtable discussion titled "SuperNodes Ignite a New Cycle," Chen Zhenkuan, Vice President of Lenovo Group and General Manager of Lenovo's Infrastructure Business Group in China, gave a very clear assessment: as model parameters continue to grow, traditional 8-GPU servers will increasingly struggle to independently carry next-generation large models. SuperNodes are no longer an option but a mandatory evolution for AI infrastructure. Notably, several AI chip companies have already begun supporting SuperNodes. Chen estimates that by next year, leading AI chip companies will generally have the relevant capabilities. "Next year will definitely be the big year for SuperNodes. If this year we are still debating feasibility, next year we will see large-scale commercial deployment."
The impact of this shift extends beyond servers. Chen cited examples where the significant increase in SuperNode weight and power consumption is already forcing upgrades in factory load-bearing capacity and power infrastructure across the manufacturing system. A single SuperNode production base may require 30MW or even higher power support. At the product level, the once-clear boundaries between servers, storage, networking, operating systems, and software are disappearing. A SuperNode requires compute, power, storage, and connectivity to work together, bringing technologies such as optical interconnect, liquid cooling, AI storage, scale-up, and scale-out into a single integrated system. "Today, a pure server vendor cannot build a SuperNode on its own. It requires comprehensive capabilities across the board. This is a certainty," Chen stated.
If SuperNodes answer how AI infrastructure will evolve, the second roundtable, themed "The New AI Cycle: The Symbiotic New Landscape of Computing Power, Large Models, and Industrial Commercialization," pushed the discussion further into industrial commercialization. Currently, large models are moving from simple Q&A to task execution, from training-centric to a balance of training, fine-tuning, and continuous inference. At the same time, as AI enters production systems, enterprises are no longer just focused on model leaderboards but on concurrency, latency, cost, and data isolation. From an industrial investor's perspective, Song Chunyu, Vice President of Lenovo Group, Chief Investment Officer of Lenovo Capital and Incubator Group (LCIG), and senior partner, pointed out that AI has been the number-one track heavily invested in by Lenovo Capital over the past decade, and it's also the track with the best results in terms of number of investments, capital deployed, and returns. To date, LCIG has invested in over 330 companies, 26 of which have gone public. In the AI field, it has invested in more than 150 companies, building a full-stack portfolio spanning AI chips, AI infrastructure, large models, intelligent agents, autonomous driving, and embodied intelligence.
Song noted that AI is just getting started; the market is still primarily in the first wave of infrastructure construction, and true entry into thousands of industries has not actually begun. Based on long-term industrial investment observations, Song pointed out that Silicon Valley investment circles are discussing the trend of foundation models "eating agents." As models' long-range reasoning and agent capabilities rapidly strengthen, some application spaces that once existed independently may be absorbed by foundation models. Therefore, for investment institutions, judging the boundary between model capabilities and application startups is becoming a key question in finding the next round of opportunities. But regarding the long-term direction of the AI industry, Song is unequivocal. He believes AI has only just begun, and this is indeed the early stage of the fourth industrial revolution. Given that generative AI has truly exploded for less than four years, what the market currently sees is mainly the first wave of infrastructure building. "Actually, entering thousands of industries hasn't truly started yet."
This means that from computing power to models and applications, a new value chain has formed in today's AI investment. Computing infrastructure provides the foundation, model capabilities continuously lower the threshold for intelligent production, and what ultimately determines how far this round of AI investment can go is whether technology can truly enter thousands of industries, creating sustained usage and commercial returns. This also makes Lenovo's value coordinates as a global AI infrastructure leader clearer. As AI infrastructure moves from standalone machines to SuperNodes, from point computing to system engineering, competition is no longer just about servers themselves but about comprehensive capabilities in computing, storage, networking, liquid cooling, power supply, software, and global delivery. And as AI further enters industries, infrastructure needs to form more complete synergy with models, terminals, and industry scenarios.
From ranking first globally in x86 server shipments to an AI server order backlog of RMB 360 billion, Lenovo is capturing this wave of value migration as AI moves from "stacking computing power" to systemization, and from infrastructure construction to industrial implementation. This may also be the increasingly clear meaning of "certain assets" after the divergence in AI trading. Whether the long-term AI industry trend can be continuously converted into systemic capabilities and operational results is the ultimate question that certainty must answer.
Comments