It takes as little as18 months to go from groundbreaking to GPU deployment for a data center; however, the lead time for delivering a heavy-duty gas turbine has now extended beyond three years. For the first time, the pace of AI expansion is being constrained by power equipment. In 2025, global demand for gas turbines exceeded 100 GW, while supply stood at only 50–60 GW. Mitsubishi Power's production schedule extends to 2031, and GE Vernova's capacity is booked out until after 2029. Quotes for the same model have accumulated an increase of approximately threefold over three years.
Cloud providers can pay premiums to secure chips, but they cannot simply buy new gas turbine production lines with money: high-temperature alloys, large castings and forgings, and skilled labor cannot be scaled up quickly. GPUs depreciate over three years, while power assets depreciate over thirty years. Consequently, the competition for computing power has been pulled back into the time scale of heavy industry. In the past, Chinese manufacturing has repeatedly reshaped delivery landscapes in asset-heavy industries; this time, it is the turn of gas turbines.
01. Aging Grids Meet New Computing Power: The Global Gap in Stable Power Sources Is Opening Up to Chinese Capacity
The problem facing the Western power industry lies in the simultaneous loss of control over unit age and load curves. The United States has a cumulative installed capacity of approximately 1.33 billion kW, with main units serving for 28 to 32 years; net additions in 2025 were only 45 million kW. Japan's installed capacity is about 320 million kW, with units serving for 22 to 26 years. Russian units have an average service life of 35 to 40 years. These systems can maintain existing stock but struggle to provide dispatchable incremental capacity. Wind and solar provide energy volume but not capacity; nuclear power has a construction cycle of ten years; coal-fired power faces policy constraints in Europe and the US. Among options that can supplement stable capacity within three to five years, gas turbines rank first.
AI has further widened this gap. McKinsey predicts that global data center electricity demand will rise from 82 GW in 2025 to 219 GW in 2030. Data center loads have shifted from interruptible to around-the-clock operations, requiring cloud providers to balance baseload stability with rapid start-stop capabilities. GE Vernova had a backlog of approximately 100 GW by the end of 2026, with capacity fully booked for 2029–2030. Siemens Energy revised its annual demand forecast upward to 110–120 GW. Global orders in Q2 2026 reached 38 GW, a 29% quarter-over-quarter increase and a 71% year-over-year increase. Price hikes best illustrate the issue: prices for combined-cycle units delivered in 2031 are already triple last year's levels. Combined-cycle plants offer high efficiency, fast start-up, and seamless integration with existing grids, precisely addressing the certainty most valued by cloud providers. With expansion cycles generally exceeding three years, this gap cannot be filled by overtime work alone.
Consequently, cloud providers are beginning to move power generation from the grid side to their own campuses. The value of on-site generation, self-owned power stations, and combined-cycle units has risen accordingly: they shorten not only construction timelines but also uncertainties related to grid connection queues and electricity price negotiations. For the gas turbine chain, data centers have transformed from ordinary large customers of the grid into direct equipment buyers willing to place orders. This has made the delivery capability of power equipment directly relevant to the schedules of computing projects.
Only China possesses a complete chain capable of simultaneously providing castings and forgings, high-temperature alloy blades, complete machines, and EPC (Engineering, Procurement, and Construction) services. China's cumulative installed capacity is approximately 3.89 billion kW, accounting for more than 40% of the global total. Net additions in 2025 were about 543 million kW, nearly two-thirds of the global increment. China accounts for approximately 70% of global power equipment production capacity. Over the past two decades, China has industrialized power equipment at scale; when overseas markets once again require stable power sources, scale translates into delivery capability. Completeness is more critical than single-point technology; what overseas lacks is not just machine production lines, but also materials, castings/forgings, and skilled workers. If any link breaks, delivery schedules slip.
Orders are already flowing along this chain. Dongfang Electric handles complete machines and EPC, with its G50 autonomous gas turbine securing over 40 overseas orders. Jereh Group focuses on generator sets and oil/gas equipment integration, accumulating over USD 3.1 billion in gas turbine orders, including a USD 1.465 billion contract with a cloud provider in July. Yingliu Shares produces high-temperature alloy blades and precision castings for aero/industrial engines, generating CNY 887 million in revenue from these businesses in the first half of the year, accounting for 49.56% of its main business revenue. Overseas revenue share best indicates exposure: Jereh at 50.38%, Yingliu at 48.87%, Paike New Materials at 22.42%, and Dongfang Electric at 18.03%. Customers are also changing; early orders primarily served oil/gas fields and industrial parks, but now cloud providers appear directly, marking the entry of Chinese equipment into the core budgets of computing infrastructure.
However, financial reports initially reflect inventory and advances received, with profits lagging behind. In the first three quarters, Jereh reported revenue of CNY 11.97 billion and net profit attributable to parent shareholders of CNY 1.887 billion; Q3 standalone revenue was approximately CNY 4.324 billion with net profit of CNY 691 million, yielding a net margin of about 15.8%. Contract liabilities stood at CNY 7.051 billion, and inventory at CNY 8.604 billion. Dongfang Electric reported first-three-quarter revenue of CNY 55.44 billion and net profit attributable to parent shareholders of CNY 3.839 billion, with a net margin of about 6.9%; contract liabilities were CNY 46.67 billion, and inventory CNY 31.26 billion. Yingliu reported first-three-quarter revenue of CNY 2.943 billion and net profit attributable to parent shareholders of CNY 371 million, with a net margin of about 12.6%; inventory was CNY 2.363 billion, and operating cash flow saw a net outflow of CNY 114.7 million. Revenue is recognized based on milestones, while costs precede them: purchased components, special materials, shipping, and commissioning labor queue up ahead of the financial statements. The time lag between orders and profits is essentially the financial expression of a three-to-five-year industrial cycle. Jereh and Yingliu's Q3 performance both exceeded the H1 pace, indicating accelerated deliveries; Dongfang Electric, with the largest scale, finds it difficult for incremental gas turbine business to alter the thin-margin structure of its EPC segment. Order elasticity in upstream material and certification segments will first manifest in production scheduling before reflecting in profits.
02. This Narrative Is Familiar to Chinese Manufacturing
In 2004, orders received by Chinese shipyards were still concentrated on bulk carriers, while overseas discussions focused on when China could build LNG ships. The subsequent narrative is well known: first mastering supporting technologies, then shortening berth cycles, and finally using delivery capability to reshape market share. Solar PV, home appliances, and power batteries followed the same path. Gas turbines look like the next chapter, but they differ critically from previous industries: those waiting for Chinese products this time are the wealthiest clients on Earth.
Cloud providers calculate economics differently from grid companies. Grid companies compare the cost per kWh; cloud providers compare how much less they earn if online one year later. A large data center sitting idle for a year loses not just the electricity price differential, but also model training opportunities, customer contracts, and capital market narratives. Therefore, even though the three major gas turbine manufacturers' production schedules extend to 2031, clients are still willing to pay premiums to lock in slots. The tripling of quotes over three years is fundamentally time becoming more expensive.
Along this chain, three Chinese companies each guard a segment. Dongfang Electric originates from the complete machine sector; the G50's journey from R&D to over 40 overseas orders follows the most familiar path for domestic equipment. Jereh Group acts more like a scenario-savvy integrator, packaging gas turbines, generator sets, and oil/gas field experience; among its USD 3.1 billion in orders, the USD 1.465 billion deal in July came directly from a cloud provider. Yingliu Shares holds a position further upstream; high-temperature alloy blades must operate long-term in gas temperatures exceeding 1,000 degrees Celsius. Its first-half revenue from aero/industrial engine businesses was CNY 887 million, accounting for nearly half of its main business. Further upstream, Paike New Materials, Tunan Special Steel, and Steel Research High-Tech guard forgings and alloys—names rarely appearing in AI discussions.
This type of business has an invisible threshold: certification. Before blades enter a client's system, they undergo lengthy testing and batch verification. Once passed, the cost of switching suppliers is extremely high. Therefore, Yingliu's CNY 2.363 billion in inventory and negative cash flow resemble the bill for an expansion phase rather than a signal of weak demand. Orders, inventory, and advances arrive first; revenue and profits follow. This is the common rhythm of this industrial chain. Jereh's CNY 7.051 billion and Dongfang Electric's CNY 46.67 billion in contract liabilities represent parts of the song that have been written but not yet sung.
This round also features variables absent in the past. Solar exports faced tariffs; gas turbine exports face communities. After the US midterm elections, data centers must answer two questions simultaneously: where does the electricity come from, and who pays the bill? The former determines gas turbine demand; the latter determines whether projects can break ground. For Chinese manufacturers, the worst-case scenario is overall project delays; another scenario is clients bypassing long grid connection queues by increasing the proportion of on-site generation. Both scenarios may occur simultaneously; order structure reveals direction better than order volume.
Chinese manufacturing has proven many times that once it enters asset-heavy, long-cycle, delivery-driven industries, time tends to stand on the side of complete capacity. Gas turbines are tougher, requiring simultaneous readiness in materials, complete machines, integration, and overseas services; any shortcoming will drag down delivery. Going forward, three small things beyond orders are more critical: whether the overseas revenue share continues to rise, whether inventory converts to revenue quarterly, and whether gross margins rise alongside deliveries. When these three align, gas turbine exports will transition from logic to industry.