Fifteen years ago, Apple exited the server market. Now, it appears to be considering a comeback. According to The Information's report, Apple is evaluating the development of an inference server aimed at AI developers, enterprises, and government clients, which would utilize its proprietary M8 Ultra chip and could potentially launch as early as 2029. The project is not yet finalized, but if it comes to fruition, this would mark Apple's first foray back into the server market since it discontinued the Xserve in 2011. The fact that one of the most successful consumer electronics companies is considering re-entering a business it abandoned years ago highlights the increasingly pivotal role of servers in the AI era.

This very shift was the starting point for the discussion on "Deterministic Opportunities in the AI Infrastructure Era," which took place at the Sina Finance AI Investment Summit on September 16th. During a marquee dialogue session, Wu Yi, Co-Head of BofA Securities' China Research and Chief China Strategist, posed a long-debated market question to Lenovo Group's CFO, Wai Ming Wong: "Given the massive global capital already deployed in AI infrastructure, how much more growth potential exists in this investment cycle, or is it already beginning to overheat?"

Wong's response was direct: "We are only seeing massive demand." Lenovo's AI server backlog, which previously stood at over $10 billion, then climbed to over $20 billion, has now surpassed $50 billion in the most recent quarter. He also revealed that on the very morning of the summit, he was in discussions with a major global private equity fund about supporting NeoCloud in building new computing infrastructure. Just days earlier in Los Angeles, during a public event discussing AI development, Nvidia CEO Jensen Huang received a direct call from Donald Trump, who explicitly opposed any notion of the US AI industry "hitting the brakes."

As the global AI infrastructure race accelerates, the rankings in the global server market are also changing. According to the latest IDC data, in the second quarter of 2026, LENOVO GROUP's x86 server shipments surpassed Dell to claim the number one spot globally for the first time. Thirteen years after Lenovo first topped the global PC market, it now finds itself in a similar position in the server market. The number one ranking in 2013 ultimately led the market to accept Lenovo as a global PC company; the number one ranking in 2026 presents the market with a new question: Is LENOVO GROUP still a PC company with a server business, or is it on its way to becoming a company that possesses both the cash flow of a PC giant and the scale of a global AI infrastructure player?

Where Does the 'Determinism' of AI Infrastructure Come From?

Wong's assessment is not an isolated one. From an industry investment perspective, this round of AI infrastructure build-out is still in its expansion phase. In March of this year, McKinsey described it as "one of the largest infrastructure build-outs in modern history," projecting that global data center-related investment could reach approximately $7 trillion by 2030, with around $5.2 trillion of that dedicated to AI workloads. The flow of capital extends far beyond servers themselves; encompassing chips, storage, networking, and also drawing in land, power, cooling, and data center construction. Morgan Stanley's September research further projects that by 2028, global AI computing power could reach approximately 120GW, roughly four times the level seen in 2025.

As the capital expenditure base continues to grow, the market's focus is also shifting from "whether AI spending will continue" to "whether such massive investments can generate sufficient returns." This was a central theme repeated throughout the summit discussions. Before Wong took the stage, Peng Wensheng, a practicing professor and chief economist at the Shanghai Advanced Institute of Finance (SAIF), had already raised a sharper question from a macroeconomic perspective: The risk premium on US AI assets has been compressed to very low levels, indicating that the market is pricing in high expectations for future returns. Whether such robust capital expenditure can ultimately deliver on these expectations is the true test AI investment must face. Peng characterized large models as a form of "asset-heavy, scaled production." Unlike past internet platforms that expanded with near-zero marginal costs, the advancement of large models entails increasing investments in computing power, data, electricity, and data centers. The digital economy, as a result, is exhibiting increasingly pronounced industrialization characteristics.

Wu Chaoze, President of the Future Industry and Policy Research Institute at China Securities, who spoke next, placed this wave of investment within a longer industrial cycle for comparison. She estimated that capital expenditure by major North American cloud providers has reached $700-800 billion in 2026. Even if the growth rate decelerates in the future, the absolute scale of new investment remains substantial. In her view, the real risk the market needs to watch is the time lag between massive underlying capital expenditure and commercial returns, rather than a sudden halt in model capability evolution. This points to the true meaning of "determinism" discussed at the AI investment summit. It doesn't imply that AI capital expenditure can grow indefinitely or that all participants will achieve returns. Rather, it points to a relatively clear fact: amidst significant uncertainty in model competition and application adoption, there are currently no clear signs of a reversal in the underlying demand for computing power.

However, computing power demand is just the first layer. What's truly noteworthy is that as model and infrastructure architectures evolve, the system value embedded in each unit of computing power is also increasing.

The Value Boundaries of AI Infrastructure Are Expanding

The transformation of AI infrastructure isn't just about spending more money; the servers themselves are changing. Chen Zhenkuan, Vice President of Lenovo and General Manager of its China Infrastructure Business Group, articulated this change more concretely during a panel discussion themed "SuperNodes Ignite a New Cycle." In the past, servers, storage, networking, and operating systems were relatively independent products; in the SuperNode era, they need to be reassembled into a single integrated system. A SuperNode can be deconstructed into computing cabinets, power cabinets, storage cabinets, and connection cabinets, involving chip interconnect, liquid cooling, power, AI storage, and both Scale-Up and Scale-Out architectures. "Today, a pure server vendor cannot create a SuperNode on its own," Chen stated.

The main driver of this change is the shift in model scale. A few years ago, a large model with 175 billion parameters could run on traditional server architectures. However, as models evolve towards trillion-plus parameter scales, traditional 8-GPU servers are increasingly unable to handle the tasks independently. SuperNodes with 32 or more GPUs are becoming a crucial form of AI infrastructure. Server vendors are thus facing an entirely different kind of business. Previously, they delivered a single machine; today, it might be a set of cabinets; in the future, customers may purchase an entire "AI Factory" encompassing compute, storage, interconnect, liquid cooling, deployment, and maintenance.

The "AI Factory" concept that Wai Ming Wong repeatedly referenced points to this transformation. Lenovo's latest financial results already reflect this trend. In the first quarter of FY2026/27, ISG revenue reached $8.5 billion, a 98% year-over-year increase; operating profit hit $777 million, with operating margin reaching a record high of 9.1%; and AI server backlog grew 157% quarter-over-quarter to $54 billion. Beyond training, the next wave of incremental growth may come from inference. In the latter part of the dialogue, Wong discussed his participation in a global CFO advisory board. A clear change is that almost all large enterprises are now considering AI investment, but many are still grappling with integrating their internal databases, processes, and systems. Wong believes that enterprise AI, particularly investment on the inference side, is still in its early stages. If training demand over the past few years was concentrated among a few hyperscalers, the buyers in the inference era might expand to include NeoClouds, governments, large enterprises, and even local data centers.

The customer base is broadening, and the value of individual systems is increasing. For LENOVO GROUP, it is no longer facing a traditional server market, but an AI infrastructure chain that extends from standalone machines to systems, from training to inference, and from the cloud deeper into the enterprise. However, a large addressable market doesn't guarantee that every company will capture the same value. AI servers are first and foremost a scale business, but ultimately they must also be a profit business.

Dell Emerges as Lenovo's New Valuation Benchmark

On the topic of profitability, Wu Yi raised a question of significant interest to the capital markets: "Lenovo announced a medium-term target for group net profit margin to reach 5% in April this year. Is it possible to achieve this ahead of schedule?" Wong did not provide a specific timeline. He stated that the $100 billion revenue target might be achieved earlier than originally planned, but the company still needs to balance growth and profitability. In the long run, the group's net profit margin will still move towards 5%, or even 8%. He then proactively mentioned Dell. "Recently, Dell has been the leader globally in this regard," Wong said. He added that Lenovo has its own advantages in technology, channels, and supply chain, and "should have the opportunity to reach their profit level."

This statement effectively provides a clearer financial benchmark for Lenovo's AI infrastructure business in its next phase. For a long time, when the capital market discussed Lenovo, it was more accustomed to using PC manufacturers like HP as a reference. Now, management is proactively placing Dell into the comparison framework, signaling a desire for the capital market to re-evaluate its business structure. Dell's fiscal Q2 2027 results, announced in early September, showed quarterly revenue of $47 billion, a 58% year-over-year increase; ISG revenue of $31.8 billion, up 89%; and AI server quarterly revenue of $16.4 billion, with new orders of $60.9 billion and a year-end backlog of $95 billion. ISG operating profit was $4.8 billion, up 225% year-over-year, with an operating margin of approximately 15%.

Whether it's the scale of the infrastructure business, recognized AI server revenue, or profit margins, LENOVO GROUP currently lags behind Dell. However, some operating metrics are closing the gap rapidly. Lenovo's ISG revenue grew 98% year-over-year in its latest quarter, outpacing Dell's 89% growth during the same period. In Q2 2026, Lenovo's x86 server shipments surpassed Dell for the first time, with related revenue reaching $8.26 billion, narrowing the revenue gap with Dell to just $570 million. This suggests the competition between the two companies is moving from a stage of clearly disparate scales to one where more metrics can be directly compared.

Wu Yi also pressed Wong on this point: Given Dell's current lead in the AI server market, what exactly is Lenovo lacking, and in which areas does it plan to catch up? Wong did not deny that Lenovo was late in entering some overseas NeoCloud markets. He mentioned that some overseas customers previously "didn't even know Lenovo." However, he believes the company still has significant room to catch up in terms of technology, supply chain, and market communication. Lenovo's chips are not just about server volumes either. The supply chain capabilities honed during the PC era are being transferred to AI infrastructure. Lenovo currently has more than 30 manufacturing bases globally, operates in 180 markets, and rose to 5th place in the 2026 Gartner Top 25 Supply Chain ranking, while Dell ranked 17th. Concurrently, the company has been strengthening its capabilities in liquid cooling, high-end storage, and rack-scale solutions, building a capacity that delivers from individual servers to full system-level solutions.

More importantly, Lenovo serves over 5,800 AI customers, extending from hyperscalers and NeoClouds to enterprise and government sectors. Combined with its SSG services business, which boasts an operating margin exceeding 24%, this allows the company to convert one-time hardware deliveries into recurring revenue streams. In other words, what Lenovo is truly attempting to replicate is not just the "number one in scale" achievement from the PC era, but to recombine its supply chain, system capabilities, global customer base, and services revenue into a competitive moat for the AI infrastructure era.

The truly sensitive question ultimately revolves around valuation. "Lenovo's share price has risen over 250% this year. As CFO, how do you view Lenovo's long-term valuation?" Wong's response was more direct here: "Our valuation is about 1/7th of Dell's, but my revenue and net profit are not 7 times worse." He added, "We are completely undervalued by the market." If you place this statement back into the context of the aforementioned operating data, it isn't just management's opinion on the stock price. Dell's current advantages are real, but Lenovo has also begun to enter the same comparison framework. Therefore, what the market truly needs to reassess may not just be how many servers Lenovo can sell, but whether, as its capabilities continue to materialize, the operational gap between Lenovo and Dell still justifies today's stark valuation disparity.