From October 20 to 22, by the banks of Shanghai's Huangpu River, the 2026 Sustainable Global Leaders Conference returns once again. Nobel Prize and Turing Award winners will take the stage to discuss the future and trends of technology, and the internet-famous professor who "rescues" people from anxiety and the era of intense competition — Macfarlane — will hold a live dialogue with Xiang Biao!
Spots are limited, so act fast. Let us gather for this sustainable appointment together! [Click to register] The "heavyweight bomb" dropped by NVIDIA (NVDA) did not land in Silicon Valley, but in the Southern Hemisphere. Recently, NVIDIA announced partnerships with eight Australian cloud computing and data center AIDC enterprises to build up to 2GW of AI data center capacity by 2027. What does 2GW mean? Using the rough conversion that a single NVIDIA H100 server consumes about 10kW of power, 2GW could support the operation of approximately 200,000 H100-class servers. The world's largest single data center currently consumes only around 200-300MW. In other words, NVIDIA's move in Australia alone represents the compute scale of nearly ten ultra-large-scale AIDC data centers. So how exactly does this compute chain from chips to AIDC data centers operate? Taking the ChiNext Compute ETF China (158041) as an example, it anchors the compute infrastructure sector, focusing on data center AIDC construction and operations, compute leasing, liquid-cooled servers, and other core compute areas. Among these, the "data center (AIDC)" concept exposure is the highest in the entire market (91.03%), making it the sharpest tool for capturing compute infrastructure dividends. Using this as a framework, let us break down the full compute landscape.
Full Compute Landscape Deconstruction: A Three-Layer Architecture Breakdown
Many people think chips are compute, but compute is far more than that. The compute industry chain is not a linear chain, but a layered, nested, three-dimensional ecosystem. From the perspective of value creation, it can be clearly divided into three layers: Upstream: Compute chips and core components — the "brain" of compute. GPU/ASIC/NPU and other compute chips serve as the computing core, paired with HBM high-bandwidth memory to break through memory bottlenecks, supplemented by liquid cooling to solve the heat dissipation challenges brought by high-density compute. Together, these three form the source of compute supply. Midstream: Integration, interconnection, and infrastructure construction — the "skeleton and blood vessels" of compute. Upstream core components are assembled into compute servers, and through high-bandwidth, low-latency high-speed interconnections within racks, large numbers of compute chips are integrated into a single ultra-large compute unit. Servers are connected through switches and optical modules to form high-speed interconnect networks, enabling efficient scheduling and transmission of compute between nodes, thereby opening the channel from "single-point compute" to "cluster compute." Downstream: Compute operations and application deployment — the "limbs" of compute. Compute operations and application deployment use AIDC data centers as the underlying carrier. The construction of AIDC itself is a systematic engineering project combining resources, energy, hardware, and software, giving rise to various delivery forms such as rack colocation, compute leasing cloud services, and intelligent computing solutions, transforming into actual services that enterprises can invoke, price, and deploy. Upstream supply, midstream interconnection, downstream monetization — the three layers are interlocked, forming a complete compute industry landscape. The key to understanding the compute chain lies in the "system" rather than "single points." A single chip is like an engine, while compute infrastructure is more like an integrated system that can operate continuously: how servers are assembled, how data is stored and retrieved, how networks are connected, how machine rooms are powered, cooled, and continuously maintained — if any link becomes a bottleneck, the entire system cannot operate efficiently. When rack power density rises, delivery cycles lengthen, and AIDC data centers become scarce resources, server racks, power supply and distribution, temperature control and cooling, IDC operations, and network connectivity all become part of system capabilities. This is why compute market rotations repeatedly cycle between "chips → optical modules → servers → liquid cooling → AIDC."
AIDC (AI Data Centers): The "Super Token Factory" of the AI Era
Global compute infrastructure is embarking on an "arms race," and AIDC is the main battlefield of this competition. Just like this 2GW deal — it is not NVIDIA personally building machine rooms, but rather the eight partners each operating their own "AI factories." What is AIDC? It solves the question of "where AI models are trained, where they are inferred, and how compute is efficiently monetized from hardware into capability." It is the only physical endpoint where capital expenditures on compute chips, servers, optical modules, and liquid cooling are converted into usable compute — without AIDC, all upstream hardware procurement cannot become deliverable Token throughput capacity. What kind of business is AIDC? A high-performance AIDC is not simply buying GPUs and stuffing them into a machine room. On the surface, it is a data center machine room, but at its core, it is a systematic engineering project connecting high-density compute + high-speed networks + high-power supply + efficient cooling + AI scheduling platforms + capital. Heavy asset investment is the ticket to entry: GPUs, servers, power, cooling, land, facilities — none can be missing. High-tech integration is the moat: coordinating GPU clusters, servers, liquid cooling, high-speed networks, and scheduling systems. Strong operational capabilities determine profitability: whether compute can be reasonably scheduled and achieve utilization rates directly determines whether fixed costs can be converted into cash flow. Customer resources and stickiness determine whether revenue can be sustained. What are AIDC's business model and profit logic? AIDC's monetization forms progress through four levels: ① Rack colocation (like renting a bare room and buying your own furniture) — customers purchase their own servers and pay by rack, power, and bandwidth; heavy asset but simple model; ② Compute leasing (like renting a fully furnished apartment and moving right in) — operators build their own clusters and lease by time/node/cluster; the test is utilization rate; ③ AI cloud services (like daily rental homestays, stay when needed) — packaging underlying compute into a platform with usage-based billing; the test is user stickiness; ④ Intelligent computing solutions (like providing specific event scenarios such as proposal venues, birthday parties, team-building activities) — providing integrated services of compute + data + models + applications, delivering overall capabilities with the highest added value. From ① to ④, added value increases progressively, and operational difficulty also grows. This looks like a "rent collection" business, yet it has higher comprehensive barriers than rent collection. Among these, three value-added main lines run throughout: ① From "selling assets" to "selling efficiency": the value-added of compute operations first comes from "activating" heavy assets — through scheduling, reuse, and overselling, maximizing the utilization rate of physical compute; secondly, reducing ineffective energy consumption and saving high electricity costs. High utilization + low PUE = cost reduction and efficiency improvement, which is the primary productivity of compute operations. (PUE is the core metric for measuring data center energy efficiency; PUE = total data center energy consumption ÷ IT equipment energy consumption, always greater than 1; the closer to 1, the higher the energy efficiency.) ② From "selling resources" to "selling capabilities": bare racks, bare cards, and bare compute are homogeneous and low-margin. True value-added lies in layering software platforms, scheduling systems, and models to transform resources into "out-of-the-box capabilities." ③ From "selling cost" to "selling value": the ultimate value-added of compute operations is shifting from cost-based pricing to value-based pricing — billing by business outcomes, sharing by results, and binding by growth. Embedding compute into customers' business value chains captures the richest profits of the compute era. This is precisely where AIDC's investment value lies: it is the link in the compute industry chain with the strongest certainty and the most comprehensive barriers. The elasticity of hardware links depends on orders and iteration, while AIDC's appeal lies in the sustained cash flow formed after capital, technology, energy, and operations are all twisted together — in the compute "arms race," whoever owns AIDC that operates stably, efficiently, and safely holds the compute "rent collection rights" of the AI era.
————AIDC development is also accelerating domestically. As of the end of 2025, China has built 42 ten-thousand-card-level intelligent computing clusters, with total intelligent compute scale exceeding 1590 EFLOPS. AIDC has moved from a few experimental clusters into large-scale engineering construction. Demand no longer falls on just one type of GPU, but is transmitted to every link — chips, boards, complete machines, switching networks, power supply and distribution, cooling equipment, and campus operations. The decisive factor in the future compute market is not whose chip parameters are prettier, but who can make each kilowatt-hour and each watt of power produce more Tokens. Related ETF: ChiNext Compute ETF China (158041) tracks the ChiNext Compute Infrastructure Index, anchoring the infrastructure sector of the AI industry chain, focusing on data center (AIDC) construction and operations, compute leasing, liquid-cooled servers, and other core compute areas. Among these, the index's "data center (AIDC)" concept exposure is the highest in the entire market (91.03%), filling the gap for investors seeking a pure compute investment tool in A-shares. Fund fee rates: Investors trade the above ETF on stock exchanges just like buying and selling stocks; the main costs are brokerage commissions and fund operating expenses (management fee and custody fee are 0.50%/year and 0.10%/year respectively). When investors subscribe or redeem fund shares, the subscription/redemption agency may charge a commission of no more than 0.3%. Risk disclosure: 1. The above fund is an equity fund, mainly investing in the underlying index constituent stocks and alternative constituent stocks. Its expected risk and expected return are higher than mixed funds, bond funds, and money market funds. It is a medium-high risk (R4) product. The specific risk rating results are subject to the rating results provided by the fund manager and sales institution. 2. The above fund has main risks such as deviation of underlying index returns from average stock market returns, underlying index volatility, and deviation of fund portfolio returns from underlying index returns. 3. Before investing in the above fund, investors should carefully read the fund's "Fund Contract," "Prospectus," and "Product Information Summary" and other fund legal documents, fully understand the risk-return characteristics and product features of the above fund, and fully consider their own risk tolerance based on their own investment targets, investment period, investment experience, asset status, and other factors. On the basis of understanding the product situation and sales suitability opinions, investors should make rational judgments and prudent investment decisions, and independently bear investment risks. 4. The fund manager does not guarantee that the above fund will definitely profit, nor does it guarantee minimum returns. The past performance and net value levels of the above fund do not predict its future performance, and the performance of other funds managed by the fund manager does not constitute a guarantee of the performance of the above fund. 5. The fund manager reminds investors of the "buyer beware" principle in fund investment. After investors make investment decisions, investment risks arising from fund operating conditions, fund share secondary market trading price fluctuations, and fund net value changes shall be borne by investors themselves. 6. The registration of the above fund by the China Securities Regulatory Commission does not indicate that it makes a substantive judgment or guarantee on the investment value, market prospects, and returns of the above fund, nor does it indicate that investing in the above fund is risk-free. 7. The above products are issued and managed by China AMC, and distribution institutions do not bear responsibility for product investment, redemption, and risk management. 8. Markets carry risks, and investment requires caution. MACD golden cross signals have formed, and these stocks are rising well!
Comments