SOLAR · DEEP DIVEVERIFIED

Solar
Solar's Glut, Verified: Capacity, Output and Installations Are Three Different Numbers

Almost every English-language account of China's solar "overcapacity" quotes a single comparison: factories that can make more than a thousand gigawatts of modules a year, set against a world that installs roughly six hundred. The ratio is real, and it is also the wrong number for almost every decision it is used to support — because the thousand and the six hundred do not measure the same thing, and between them sit two more numbers that the coverage routinely skips. Chinese statistics keep the three apart with three different words. Capacity (产能, nameplate) is what the industry's furnaces and lines could produce in a year if they ran flat out. Output (产量) is what they actually produced. Installations (新增装机, new grid-connected capacity) is what got built and connected. The glut lives in the gaps between those three numbers — and the gaps are not the same size, do not move for the same reasons, and do not imply the same thing for a buyer, an investor or a policymaker. This note reads the three series from their primary sources — the China Photovoltaic Industry Association (CPIA, 中国光伏行业协会) for the first two, the National Energy Administration (NEA, 国家能源局) for the third — and shows what each one actually counts.

>1,100 GWModule nameplate capacity at end-2025 (CPIA) — against ~612 GW of expected global installations in 2026
38–56%Implied 2025 running rates by chain link — our arithmetic on CPIA's printed capacity and output figures
315.07 GWNet solar additions in 2025 (NEA) — followed by 72.07 GW in H1 2026, down 66.04%, as the deadline rush annualised out

01 · THREE NUMBERSThree series, three publishers, three different verbs

Start with who publishes what, because in this industry the publisher is part of the definition. Manufacturing capacity and output come from the CPIA, the industry association whose annual roadmap — 《中国光伏产业发展路线图(2025–2026年)》, the China PV Industry Development Roadmap (2025–2026), compiled with the China Center for Information Industry Development (赛迪智库集成电路研究所) under the guidance of the Ministry of Industry and Information Technology's electronics department (工业和信息化部电子信息司) — is the reference document for the supply side. Installations come from the NEA, which counts grid connections, not factory gates: its monthly power-statistics releases (全国电力统计数据) and its annual renewables grid-connection reports (可再生能源并网运行情况). Exports sit in between, recorded by customs and summarised in CPIA's seminar reporting.

The units already tell you the three series are not interchangeable. Polysilicon is counted in tonnes; wafers, cells and modules in gigawatts. And the module gigawatt is a DC nameplate rating — the sum of the panels' rated peak output — while the NEA counts a solar plant in the same table as coal and nuclear plants rated in AC terms. A gigawatt of solar in the NEA's capacity table is not a gigawatt of coal, and a gigawatt of module capacity in CPIA's table is not a gigawatt of anything until someone builds a plant with it. English coverage that lines the three up on one axis has already made its first error before reaching the numbers.

02 · CAPACITYThe nameplate: what could be made, if every line ran

CPIA's February 2026 outlook seminar put China's end-2025 nameplate capacity at more than 3.5 million tonnes of polysilicon, more than 1,500 GW of wafers, more than 1,400 GW of cells and more than 1,100 GW of modules — three to six times the end-2020 levels (polysilicon had been 460,000 tonnes, wafers 240 GW), and 96%, 96.2%, 91.3% and 80.1% respectively of the world's capacity at each link. These are the figures behind every "China can make twice what the world needs" headline, and as far as they go, they are correctly sourced. What they count deserves more care than they usually get.

Nameplate capacity is a stock of potential. It counts every line that exists, whether or not it ran last year: mothballed lines, lines producing last generation's product, lines whose owners have stopped feeding them. It assumes full-rate, full-year operation that no real plant sustains. And it is measured against a demand figure that is itself an installation forecast: CPIA's own July 2026 update expects 612 GW of new installations worldwide in 2026 — so China's module nameplate alone is roughly 1.8 times the entire world's expected build (our arithmetic on CPIA's two numbers). That ratio is a statement about how much building-out happened in 2021–2023, when capacity at every link tripled or more in three years. It is not a statement about how much product will reach a market this year. For that, you need the second number.

03 · OUTPUTWhat the furnaces actually made — and the wedge where the glut lives

The CPIA roadmap records 2025 output of 1.34 million tonnes of polysilicon (down 26.4% year on year), about 680 GW of wafers (down 9.7%), more than 660 GW of cells (up 0.9%) and more than 620 GW of crystalline-silicon modules (up 5.4%). Set output against the end-2025 nameplate and the implied running rates are roughly 38% for polysilicon, 45% for wafers, 47% for cells and 56% for modules — our arithmetic, offered as an approximation (the capacity figure is an end-year stock, the output a full-year flow), and consistent with the utilisation rates CPIA itself reported for the first half of 2025: 38.7% for wafers, 45.7% for cells, 48.3% for modules. In other words, the industry's answer to a 1,100 GW nameplate was to run it at about half speed. The first half of 2026 shows the adjustment deepening: CPIA's July seminar reported output of 538,000 tonnes of polysilicon (down 9.8%), 293 GW of wafers (down 7.3%), 260.7 GW of cells (down 21.9%) and 201.3 GW of modules (down 35.1%) — module output annualising to roughly 400 GW against that same 1,100 GW-plus nameplate, a running rate near 37%.

This wedge — capacity minus output — is where the economics of the glut actually live. It is why the same industry can be simultaneously "twice oversupplied" on paper and a hard negotiator in practice: the marginal gigawatt in the capacity figure belongs to a line that is idle, obsolete or both, and it will not be offered to your procurement tender at any price. It is also the number that policy is now aimed at. The exit route Beijing has chosen runs through standards rather than quotas: three mandatory national standards published in June–July 2026 — energy-consumption limits for polysilicon and monocrystalline silicon, and efficiency limits for modules and inverters, effective 1 January 2027 — which, on estimates presented at CPIA's July seminar, could retire 25–60% of backward polysilicon capacity, around 30% of monocrystalline-silicon capacity and about 20% of module capacity. Nameplate, in other words, is about to be edited downwards by regulation. Anyone quoting the 1,100 GW figure in 2027 without asking what the standards removed will be quoting a ghost.

04 · INSTALLATIONSThe demand side runs on a policy clock, not a market one

The third number is the one the other two ultimately serve, and it is the one most often misread abroad — because in China it is set by policy timing as much as by demand. The NEA's power-statistics series records 315.07 GW of net solar additions in 2025, taking the cumulative fleet past 1,200 GW (1,201.73 GW at end-2025). The NEA's own renewables grid-connection report gives the same year a slightly different cut — 317 GW of photovoltaic additions (164 GW centralised, 153 GW distributed), cumulative PV of 1,200 GW — a small, unexplained divergence between two series from the same agency, and a useful reminder that even "the" installation number carries a series label: one is the net change in the capacity stock, the other a count of connections on a PV-only basis.

What happened next is the cleanest illustration in the dataset of why annual installation figures cannot be read as demand. The pricing reform notice 《关于深化新能源上网电价市场化改革 促进新能源高质量发展的通知》 (发改价格〔2025〕136号, NDRC and NEA, January 2025) split projects at 1 June 2025: plants connected before the line kept a guaranteed mechanism price; everything after sells into the market, with a competitively auctioned mechanism price on top. A companion rule for distributed PV (《分布式光伏发电开发建设管理办法》, 国能发新能规〔2025〕7号) grandfathered projects filed and connected before 1 May 2025. The industry did the obvious thing and connected everything it could before the deadlines: NEA figures put first-half 2025 additions at about 212 GW, and CPIA's Wang Bohua has given the pulse's shape for commercial-and-industrial distributed projects — 55.92 GW connected in the second quarter of 2025, against 6.32 GW in the third. Then came the payback. First-half 2026 additions were 72.07 GW, down 66.04% year on year (NEA, 22 July 2026) — a collapse in the headline rate that is, in substance, a return to trend: Wang Bohua noted at the July seminar that the first-half figure sits around the 2021–2024 first-half average. By September 2026 the cumulative numbers had resumed their climb: 110.83 GW added in January–September, cumulative solar capacity of 1,312.87 GW, up 16.5% year on year (NEA, 26 October 2026), in a fleet that had already passed coal in July to become the country's largest source of installed capacity. Note, finally, the basis trap for forecasters: CPIA's 2026 installation forecast for China is 180–240 GW, while BloombergNEF's analyst put the same year at about 264 GW explicitly on an AC basis, as reported from a February 2026 industry forum — two reputable numbers for the same year, printed on different sides of the inverter, and routinely compared in English coverage as if they were the same quantity.

05 · EXPORTSThe fourth number, and the one that changed by decree on 1 April 2026

Module output has only three places to go: domestic installations, exports, or inventory. In the first half of 2026, with domestic installations at 72 GW and module output at 201.3 GW, exports were the industry's main release valve — and even they need reading twice. CPIA's July seminar reported first-half PV exports of $17.183 billion, up 24.3% year on year, with export volumes moving in three different directions at once: wafers down 0.6%, cells up 36.8%, modules down 2.5% — and the module export value rising for the first time since the second half of 2023. A value series rising while the volume series falls is a price and mix story, not a demand story; and trade-press reporting of the customs data attributes the first-half value growth almost entirely to a rush to ship before a deadline. That deadline was fiscal, not commercial: 《财政部 税务总局公告2026年第2号》 (Ministry of Finance and State Taxation Administration Announcement No. 2 of 2026, 8 January 2026) cancelled the export VAT rebate on photovoltaic products from 1 April 2026 — 249 product lines, assessed by the export date on the customs declaration — with battery rebates cut from 9% to 6% over the rest of 2026 and abolished from 1 January 2027. Exporters pulled shipments into the first quarter; the "growth" in the half-year export value is substantially that pull-forward, and the same notice is a standing headwind for every quarter that follows it.

06 · WHAT IT MEANSFor procurement and strategy readers: which number to use, when

  • Never quote an "overcapacity" ratio without naming both series. Capacity ÷ installations (~1,100 GW against ~612 GW of expected global build) measures the 2021–2023 construction boom. Output ÷ installations measures this year's market. They differ by a factor of about two, and they support opposite negotiating postures.
  • For supply security and pricing, work from output and utilisation, by link. The chain's tightest link is not the one with the biggest nameplate. Polysilicon ran at roughly 38% of nameplate in 2025 against global demand that CPIA-based reporting puts at around 1.2 million tonnes — deep surplus, but concentrated in lines the new energy-consumption standard is designed to close. Watch output and the standards' enforcement, not the capacity total.
  • Treat Chinese installations as policy-clocked. The 2025 figure (315–317 GW, depending on the NEA series) contains a deadline rush worth roughly a normal half-year of demand; the 2026 "collapse" is mostly that rush annualising out. For demand planning, the NEA's monthly series and the provincial mechanism-price auction results are the live indicators — not any single year's total.
  • Check the basis before comparing any two solar numbers. DC nameplate or AC? Net stock change or gross connections? 太阳能发电 (including solar thermal) or 光伏 alone? The two NEA series differ by ~2 GW on the same year; the CPIA and BNEF 2026 forecasts differ in basis before they differ in judgement.
  • Exports are now a policy variable too. After 1 April 2026, Chinese PV export prices carry no VAT rebate; first-half export values flatter the underlying volumes. Contract pricing that assumes pre-April export economics is working from a repealed rule.

The answer, in one paragraph

China's solar glut is real, but it is a gap between three numbers, not one number. Factories carry nameplate capacity of more than 1,100 GW of modules (and 1,400–1,500 GW of cells and wafers) against roughly 612 GW of expected global installations in 2026 — yet the industry ran that nameplate at about half speed in 2025 (620 GW of modules actually produced) and nearer a third in early 2026, while installations themselves swung from a policy-driven 315 GW in 2025 to a 72 GW first half in 2026 that merely matches the pre-rush trend. Capacity tells you what was built in the boom; output tells you what the market is actually being offered; installations tell you what policy deadlines, not just demand, have scheduled. Quote the one your decision turns on — and name it.

Primary sources

《中国光伏产业发展路线图(2025–2026年)》 — China PV Industry Development Roadmap (2025–2026)

China Photovoltaic Industry Association with CCID's Institute of Integrated Circuits, guided by MIIT's Department of Electronic Information · 4 Feb 2026 · 2025 output by chain link; 2025 installations as recorded by the NEA; efficiency benchmarks

CPIA 2026 mid-year seminar — remarks by Wang Bohua, Ningbo, 23 Jul 2026

China Photovoltaic Industry Association · H1 2026 output by link; H1 2026 export value and volumes; 612 GW global 2026 installation forecast; the 2021–2024 comparison for installations

CPIA 2025 review and 2026 outlook seminar — remarks by Wang Bohua, Beijing, 4–5 Feb 2026

China Photovoltaic Industry Association · end-2025 nameplate capacity by link; China's share of global capacity; 2026 installation forecasts

《国家能源局发布2026年1-9月份全国电力统计数据》 — NEA national power statistics, Jan–Sep 2026

National Energy Administration · 26 Oct 2026 · cumulative solar capacity 1,312.87 GW; Jan–Sep additions 110.83 GW (read in a complete republication of the NEA release)

NEA press conference on renewables grid connection — transcript, 30 Oct 2026

National Energy Administration (nea.gov.cn) · cumulative PV of 13.1亿 kW split between centralised (7.12亿) and distributed (5.99亿) plant

《2025年可再生能源并网运行情况》 — 2025 renewables grid-connection report

National Energy Administration · PV additions of 317 GW (centralised 164 GW, distributed 153 GW); cumulative PV 1,200 GW

《关于深化新能源上网电价市场化改革 促进新能源高质量发展的通知》(发改价格〔2025〕136号)

NDRC and National Energy Administration · 27 Jan 2025 · the 1 June 2025 stock/incremental split behind the connection rush

《财政部 税务总局公告2026年第2号》 — Announcement No. 2 of 2026 on export tax rebates

Ministry of Finance and State Taxation Administration · 8 Jan 2026 · PV export VAT rebates cancelled from 1 April 2026; battery rebates phased out to 1 January 2027

Full note above — free to cite with attributionREADER · FREE