Canadian Solar Inc. (CSIQ) | The Buildout — AI Infrastructure

——Mkt cap — · 52-wk —–— · YTD — · delayed
Updated Sep 22, 2026Q2 FY2026 reviewed
Canadian Solar makes solar modules and grid-scale battery storage systems, including storage for data-center grid support.
Storage 3.7 GWh
Shipped above the 2.8-3.2 GWh guide; recognized 3.3 GWh.
U.S. backlog >13 GWp
Worth north of $4.5B, with deliveries through 2029.
Storage backlog $3.5B
Includes 34 GWh under long-term service agreements.
Q2 margin 13.9%
Down from 25.1%; Q1 carried an 860 bps tariff refund.
The Buildout Takeaway
The order book is deep and storage keeps beating guidance, but profitability has not followed: the Q2 net loss widened to $77M from $32M. The question is whether Section 232 repricing and the roll-off of Jeffersonville ramp costs turn contracted volume into margin.
33 analysts·15 Buy12 Hold6 Sell
Coverage is thin — only 1 price estimate, so no target is shown

FY2026 U.S. volumes reiterated at 6.5-7 GW of modules and 4.5-5.5 GWh of storage · FY2026 CapEx around $1.3B · Q3 2026 guided to $1.3-1.5B revenue, 13.5-15.5% gross margin, 3.5-3.8 GW of modules and 3.4-3.8 GWh of storage deliveries.
Important: The Buildout is a data analytics platform. Content is generated by algorithms and AI agents using public filings, earnings transcripts, and market data. This is not personalized investment advice.
Our View

The Verdict

Canadian Solar is an energy-infrastructure supplier to the power system, not an AI computing company. It makes solar modules, solar cells and battery energy storage systems through its Manufacturing segment, branded CS PowerTech in the United States, and it develops, builds, sells and operates utility-scale solar and storage projects through Recurrent Energy. Its connection to the AI buildout is power rather than compute: data centers need firm electricity and a stable grid, and the company's e-STORAGE business sells battery systems that management says support data-center grid infrastructure and resiliency, plus an undated storage product in development for inside the AI data-center server room. The solar module and cell business, the center of the U.S. reshoring story, is not framed by management as AI-driven, and management does not break out AI or data-center revenue or backlog, so the AI share of the business is not known.

Market Cap—
Revenue (TTM)$5.0B
Revenue Growth−15.7%
EBITDA Margin (TTM)2.0%
Net Debt$5.8B
Earnings Beats3 of 7
P/E (TTM)—
EV/EBITDA (TTM)—

What We Like

  • Over 13 GWp of contracted U.S. modules worth north of $4.5B, with deliveries scheduled through 2029 and multiple long-term master service agreements with U.S. utilities, IPPs, developers and EPCs, per management.
  • A $3.5B storage backlog that includes 34 GWh of contracted projects under long-term service agreements, a recurring service layer on top of equipment sales.
  • A 500 MW / 2.5 GWh battery storage project with a major U.S. utility described as designed to support data-center grid infrastructure and resiliency; management says demand from data centers is transitioning 'from conversations to contracted opportunities.'
  • Jeffersonville Phase 1 (2.1 GWp) is dated to reach full-scale production on October 1, 2026, with the first HJT module due to a customer in Q3 2026; management says HJT was chosen for the U.S. cell plant partly because the HJT IP is cleaner than TOPCon.
  • Management says HJT cells carry a price premium of roughly 10-15% over TOPCon, and the storage business is built on in-house LFP prismatic cells with proprietary power conversion, energy-management controls and full EPC.

What We’re Watching

  • Section 232 on polysilicon and derivatives, with a proclamation implementation referenced for around December 4, 2026; management expects an H2 delivery rush and says the contracted U.S. module backlog does not yet include the 232 adjustment.
  • Capacity timing: Jeffersonville Phase 2 trial production is now dated to early 2027, while Phase 1 full-scale production is dated to October 1, 2026 and Mesquite is set to double to 10 GWp in H2 2026.
  • Storage margin is not disclosed; management says margins are projected to normalize and flagged partial exposure to lithium carbonate pricing.
  • Recurrent Energy is expected to finalize the project sales delayed from Q2 in Q3 2026; Q2 included a $24M Latin America impairment.
Bottom Line

The thesis is mixed rather than broken. The demand side strengthened in the latest quarter, with a U.S. module book above 13 GWp through 2029, a $3.5B storage backlog, and data-center demand described as moving into contracts, while the profit side weakened, with the net loss widening to $77M and gross margin resetting to 13.9%. Management's credibility splits along the same line: shipment and revenue guidance has been met or beaten, while the Jeffersonville Phase 2 trial-production date has moved out to early 2027. The open question is whether policy repricing and the October 1 Jeffersonville ramp completion turn the contracted book into margin, or whether freight, ramp costs and debt service keep the profit line where it is.

Next upJeffersonville Phase 1 is dated to enter full-scale production on October 1, 2026, the first test of whether the ramp cost drops out of the cost line. The Section 232 proclamation implementation referenced for around December 4, 2026 is the next policy test, ahead of which management expects an H2 delivery rush.
Last Quarter — Q2 FY2026

Earnings

Revenue was $1.2 billion, at the high end of guidance, and gross margin was 13.9%, in line, down from 25.1% in Q1 after that quarter's 860 bps tariff-refund accrual. Storage was the line above plan: 3.7 GWh shipped against a 2.8-3.2 GWh guide, with 3.3 GWh recognized. Higher unit shipping costs and ramp-up expenses at the Jeffersonville cell plant drove a $49 million operating loss in Manufacturing and a net loss attributable to shareholders of $77 million.

MetricQ2 FY2026Q1 FY2026Q2 FY2025YoY
Revenue$1.2B$1.1B$1.7B−28.7%
Gross margin13.9%25.1%29.8%-1590bps
EBITDA$57M$73M$273M−79.2%
EPS$-1.13$-0.47$0.11−1153.9%
Energy storage shipments (GWh)3.7 shipped / 3.3 recognized2.1 recognizedn/a—
We see demand from data centers transitioning from conversations to contracted opportunities.— Colin Parkin, CEO, 2026-08-27

Management tone: The register stayed constructive on strategy and careful on anything outside the company's control. Management flagged Q1's margin benefit as nonrecurring before the reset landed, then described the Q2 gross margin of 13.9% as in line rather than strong, citing nonrecurring prior-period items, elevated freight and Jeffersonville ramp costs. Colin Parkin now leads the call as CEO, with founder Shawn Qu as Executive Chairman and CTO; the Recurrent Energy CEO changed from Ismael Arias to Dylan Marx with no explanation in the material. Management said it was premature to speculate on the magnitude of the Section 232 pricing benefit and withheld the customer name in its space-PV discussions.

Management Guidance

Management guided Q3 2026 to $1.3-1.5B of revenue, 13.5-15.5% gross margin, 3.5-3.8 GW of solar modules and 3.4-3.8 GWh of storage deliveries, and said it expects each remaining quarter to deliver higher volumes than the last in the second half. FY2026 U.S. volumes were reiterated at 6.5-7 GW of modules and 4.5-5.5 GWh of storage, with CapEx of around $1.3B, weighted to the second half for Jeffersonville Phase 2, the Mesquite doubling and the Southeast Asia storage scale-up. Management expects to finalize the project sales delayed from Q2 in Q3, which it says will drive a sequentially stronger third quarter. The stated assumptions are that freight costs decline as volumes shift onshore, that Jeffersonville ramp-up costs normalize as Phase 1 and Phase 2 progress, and that storage margins normalize, with partial exposure to lithium carbonate pricing.

Business Trajectory

Trajectory

Revenue has been choppy near $1.1-1.2B across the last three quarters — $1.2B in Q4 FY2025, $1.1B in Q1 FY2026 and $1.2B in Q2 FY2026 — while the trailing twelve months are down 15.7% year over year at $4.99B. The profit line moved more sharply: gross margin fell from 25.1% to 13.9% once Q1's 860 bps tariff-refund accrual dropped out, and EBITDA margin on the operating-income-plus-D&A basis slipped from 6.8% to 4.7%. Management attributes the step-down to nonrecurring prior-period items, elevated freight and Jeffersonville ramp-up costs, and guides Q3 to $1.3-1.5B of revenue at 13.5-15.5% gross margin. Free cash flow has been negative in each quarter shown since Q4 FY2024, including -$375M in Q2 FY2026 and -$1.58B on a trailing basis.

Revenue & Margin Trajectory
RevenueGross margin$0$1.0B$2.0B$657M$668M$677M$692M$912M$1.1B$1.4B$651M$768M$901M$485M$1.0B$760M$920M$826M$696M$914M$1.0B$1.1B$1.4B$1.2B$1.5B$1.3B$2.3B$1.9B$2.0B$1.7B$2.4B$1.8B$1.7B$1.3B$1.6B$1.5B$1.5B$1.2B$1.7B$1.5B$1.2B$1.1B$1.2B18%14%Q3'16Q4Q1'17Q2Q3Q4Q1'18Q2Q3Q4Q1'19Q2Q3Q4Q1'20Q2Q3Q4Q1'21Q2Q3Q4Q1'22Q2Q3Q4Q1'23Q2Q3Q4Q1'24Q2Q3Q4Q1'25Q2Q3Q4Q1'26Q2
RevenueGross margin$0$1.0B$2.0B$657M$668M$677M$692M$912M$1.1B$1.4B$651M$768M$901M$485M$1.0B$760M$920M$826M$696M$914M$1.0B$1.1B$1.4B$1.2B$1.5B$1.3B$2.3B$1.9B$2.0B$1.7B$2.4B$1.8B$1.7B$1.3B$1.6B$1.5B$1.5B$1.2B$1.7B$1.5B$1.2B$1.1B$1.2B18%14%Q3'16Q4Q1'17Q2Q3Q4Q1'18Q2Q3Q4Q1'19Q2Q3Q4Q1'20Q2Q3Q4Q1'21Q2Q3Q4Q1'22Q2Q3Q4Q1'23Q2Q3Q4Q1'24Q2Q3Q4Q1'25Q2Q3Q4Q1'26Q2
Gross margin as reported.
Share Price — 12 Months
$10$20$30$052-wk high $30Sep '25DecMar '26JunSep '26
52-week range $12–$30.
Share Price — 12 Months
$10$20$30$052-wk high $30Sep '25DecMar '26JunSep '26
52-week range $12–$30.
The Numbers

The Model

The model projects FY+1 revenue of $5.4B with EBITDA of $321M (5.95% margin), and FY+2 revenue of $6.65B with EBITDA of $509M (7.65% margin). The near term is anchored by the reiterated FY2026 U.S. volume guidance of 6.5-7 GW of modules and 4.5-5.5 GWh of storage, a U.S. module backlog above 13 GWp delivering through 2029, and the possibility that Section 232 repricing lifts a contracted book management says excludes that adjustment. FY+2 depends on capacity landing and ramping: Jeffersonville reaching 6.3 GWp of U.S. cell nameplate in 2027, Mesquite doubling to 10 GWp, Southeast Asia battery cell and SolBank lines online in 1H 2027, SolBank 4.0 shipping in 2027, and the 34 GWh under long-term service agreements converting into recurring revenue.

Revenue & EBITDA Projections
REVENUE$5.6B$5.4B$6.7BFY25FY+1 (E)FY+2 (E)EBITDA & MARGIN$189M$321M$509M7.7%FY25FY+1 (E)FY+2 (E)
REVENUE$5.6B$5.4B$6.7BFY25FY+1 (E)FY+2 (E)EBITDA & MARGIN$189M$321M$509M7.7%FY25FY+1 (E)FY+2 (E)
Solid bars are reported actuals; outlined bars are model projections — not company guidance.
MetricFY2025Next FY (E)Following FY (E)
Revenue$5.6B$5.4B$6.7B
YoY Growth—−3.5%+23.1%
EBITDA$189M$321M$509M
EBITDA Margin3.4%6.0%7.7%

Projections are the median of 4 independent model runs. The model’s revenue sits 19.8% below analyst consensus.

Management guided Q3 2026 to $1.3-1.5B of revenue, 13.5-15.5% gross margin, 3.5-3.8 GW of solar modules and 3.4-3.8 GWh of storage deliveries, and said it expects each remaining quarter to deliver higher volumes than the last in the second half. FY2026 U.S. volumes were reiterated at 6.5-7 GW of modules and 4.5-5.5 GWh of storage, with CapEx of around $1.3B, weighted to the second half for Jeffersonville Phase 2, the Mesquite doubling and the Southeast Asia storage scale-up. Management expects to finalize the project sales delayed from Q2 in Q3, which it says will drive a sequentially stronger third quarter. The stated assumptions are that freight costs decline as volumes shift onshore, that Jeffersonville ramp-up costs normalize as Phase 1 and Phase 2 progress, and that storage margins normalize, with partial exposure to lithium carbonate pricing.

What Could Go Right — and Wrong

What good looks like
  • Section 232 repricing is applied to the contracted U.S. module book management says excludes the adjustment, lifting revenue and price on that order book.
  • Jeffersonville Phase 1 reaches full-scale production on October 1, 2026 and the ramp cost drops out of cost of revenue, supporting the guided 13.5-15.5% gross margin band.
  • Data-center storage wins broaden beyond the disclosed 500 MW / 2.5 GWh contract, converting management's 'conversations to contracted opportunities' language into revenue.
  • The storage backlog resumes growth while the 34 GWh under long-term service agreements compounds into recurring service revenue.
  • Recurrent Energy closes the project sales delayed from Q2 and keeps monetizing assets, narrowing its operating losses and the drag on group EBITDA.
What could go wrong
  • Section 232 slips or the Commerce-managed manufacturing offsets come in smaller than expected; the H2 delivery rush becomes a pull-forward and early 2027 compares against an inflated base.
  • Freight and Jeffersonville ramp costs persist into 2027, so the margin recovery stays argued rather than shown and EBITDA stays near the 4.7% of revenue reported in Q2.
  • Recurrent Energy keeps impairing rather than monetizing — Q2 included a $24 million Latin America impairment — and the development business stays a drag on earnings.
  • Storage margins normalize as volumes grow, with management flagging lithium carbonate exposure and disclosing no storage margin; competitor storage margins are compressing industry-wide.
  • Net interest expense of $43 million in Q2, up from $36 million in Q1, consumes a large share of quarterly EBITDA of $56.7 million as total debt stands at $7.31 billion.
What’s Next

Looking Ahead

The next twelve months are built around dated milestones. The first HJT module is due to a customer in August-September 2026, Jeffersonville Phase 1 reaches full-scale production on October 1, and Mesquite is scheduled to double to 10 GWp nameplate in H2 2026. Recurrent expects to finalize the project sales delayed from Q2 in Q3 and to keep monetizing assets, while Phase 2 equipment installation at Jeffersonville is slated for Q4 2026 ahead of trial production in early 2027. Around December 4, 2026 the Section 232 proclamation implementation anchors the policy test, and in 1H 2027 the Southeast Asia battery cell and SolBank lines are due online. The plan assumes around $1.3B of FY2026 capital spending, weighted to the second half.

Catalysts
  • Aug-Sep 2026First HJT module delivery — Tests whether the HJT cell line reaches paying customers.
  • Oct 1, 2026Jeffersonville Phase 1 full-scale — Tests whether the ramp cost line rolls off on schedule.
  • H2 2026Mesquite doubles to 10 GWp — Tests whether all future U.S. volumes can come from Texas.
  • Q4 2026Phase 2 equipment installation — Installation begins Oct-Nov, ahead of the early-2027 trial run.
  • ~Dec 4, 2026Section 232 implementation — Tests the expected H2 rush and the backlog repricing.
  • Early 2027Phase 2 trial production — Slipped once already; a second test of the schedule.
Numbers

Financials

Annual Summary

MetricFY2024FY2025TTMYoY
Revenue$6.0B$5.6B$5.0B-6.6%
Gross Margin16.7%17.2%16.4%+50bps
EBITDA$397M$189M$100M-52.4%
EBITDA Margin6.6%3.4%2.0%324bps
Net Income$36M−$104M−$186M-388.4%
Free Cash Flow−$2.5B−$1.7B−$1.6B—
Net Cash————

Key Ratios (Trailing)

Valuation
  • P/E TTM—
  • EV/EBITDA TTM—
  • EV/Revenue TTM—
  • Price/FCF TTM—
Profitability
  • Gross Margin (TTM)16.4%
  • EBITDA Margin (TTM)2.0%
  • Net Margin (TTM)-3.7%
  • ROIC-0.3%
  • FCF Conversion-1577.8%
  • SBC / Revenue0.0%
Reference

The Company

Canadian Solar describes itself as one of the world's largest solar technology and renewable energy companies: a manufacturer of solar photovoltaic modules, a provider of battery energy storage solutions, and a developer of utility-scale solar and battery storage projects. It reports two segments. Manufacturing, branded CS PowerTech in the U.S., covers the design and production of solar ingots, wafers, cells, modules and battery energy storage products, and contains e-STORAGE, the storage business. Recurrent Energy develops, builds, sells and operates solar and storage projects and provides power services (O&M) and asset management. Products include the CS7N-TB-AG and CS6.2-66TB TOPCon modules, the CS6.2-66HB HJT module, and the SolBank 3.0, EP Cube, FlexBank 1.0 and KuBank 2.0 storage systems. Cumulative deliveries reached approximately 174 GW of solar modules and more than 18 GWh of battery storage as of December 31, 2025.

The operating strategy is domestic manufacturing, which management calls the first pillar of its global strategy. Mesquite, Texas makes modules, with initial annual capacity of 5 GW and a doubling to 10 GWp peak planned for H2 2026. Jeffersonville, Indiana is the solar cell plant: 2.1 GWp in Phase 1, adding 4.2 GWp in Phase 2 for 6.3 GWp in 2027. Shelbyville, Kentucky is the planned utility-scale battery storage manufacturing and R&D hub. Global nameplate at end-2025 was 51.3 GW of modules, 32.4 GW of cells, 37.0 GW of wafers, 31.0 GW of ingots, plus 15.0 GWh of BESS and 3.0 GWh of battery cells. Revenue is now roughly half American: the U.S. was 49.6% of FY2025 revenue, up from 19.5% in 2023, while China fell from $2.15B to $0.71B over the same two years.

Business Segments

Solar modules (Manufacturing / CS PowerTech)
FY2025 revenue $3,377.7M, 60.4% of total
TOPCon and HJT modules; shipments fell to 24.3 GW in 2025 from 31.1 GW in 2024, with a $0.16/W average price.
Growth driver: U.S. domestic-content demand and Section 232 pricing
Battery energy storage (e-STORAGE, in Manufacturing)
FY2025 revenue $1,370.6M, 24.5% of total, up 68%
SolBank, EP Cube, FlexBank and KuBank systems, built on in-house LFP cells with EPC and long-term service.
Growth driver: Data-center and utility storage contract wins
Recurrent Energy
~22 GW solar / 84 GWh storage development pipeline at June 30, 2026
Develops, builds, sells and operates utility-scale solar and storage projects, plus O&M and asset management.
Growth driver: Asset monetization to recycle capital and cut debt

Competitive Landscape

The 20-F states that the company faces intense competition from a large number of module competitors, naming First Solar, LONGi, Trina Solar, JinkoSolar, JA Solar and Tongwei Solar, and from battery storage companies including Tesla, Fluence Energy and Sungrow. Competition in the U.S. now turns on domestic content: First Solar says about 41 GW of its 45 GW backlog carries a domestic-content requirement, and T1 Energy is building U.S. cell capacity with named forward offtakes. Patent risk sits on top of the price competition — First Solar says it filed a Delaware suit on May 9, 2025 alleging infringement of certain U.S. TOPCon patents, and management says HJT was chosen for the U.S. cell plant partly because the HJT IP is cleaner than TOPCon. On the evidence, if Canadian Solar's products disappeared the AI buildout would face some supply constraints for U.S.-made solar and storage, but many alternatives exist and the impact would likely be limited to higher prices or delays rather than a stall.

  • First Solar
    Named in the 20-F as a module competitor; First Solar's own disclosure says it filed a Delaware lawsuit on May 9, 2025 against Canadian Solar and its related entities alleging infringement of certain U.S. TOPCon patents.
  • Tesla
    Named in the 20-F among battery energy storage competitors.
  • Fluence Energy
    Named in the 20-F among battery energy storage competitors; also appears as a battery cell/BESS customer in the machine-derived wiring map.
  • T1 Energy
    Named in T1 Energy's own filing as one of Canadian Solar's U.S. solar manufacturing competitors; building domestic cell capacity with named forward offtakes.
  • Trina Solar
    Named in the 20-F as a module competitor; the April 2026 PTAB invalidation concerned Trina's TOPCon patents, which does not resolve the First Solar suit.
Module and storage competitors come from the FY2025 20-F competition disclosure; the First Solar litigation detail comes from First Solar's own disclosure surfaced through the supply-chain wiring map; T1 Energy's naming comes from T1 Energy's own filing. The 20-F also names LONGi, JinkoSolar, JA Solar, Tongwei Solar and Sungrow.

Supply Chain

Canadian Solar sits upstream of the power system. It buys polysilicon, wafers, equipment and electronics, makes cells, modules and battery systems, and sells to utilities, IPPs, developers and EPCs. No neighbor transcript in the packet names the company.

Supplier
3800.HK and DQ
Polysilicon and wafers (machine-derived links)
Supplier
300751.SZ
HJT solar cell manufacturing equipment (machine-derived)
Supplier
Corning
Solar glass, front (machine-derived)
Supplier
Nucor
Solar module frame aluminum (machine-derived)
→
U.S. HJT cells; in-house LFP cells
CSIQ
Makes ingots, wafers, cells, modules and battery systems; develops projects.
→
U.S. utilities, IPPs, developers and EPCs
top five customers = 17.4% of FY2025 revenue
Long-term master service agreements behind the >13 GWp module backlog
Major U.S. utility (unnamed)
500 MW / 2.5 GWh
Battery storage project for data-center grid infrastructure and resiliency
Apex Clean Energy
381 MWh
Battery storage in Michigan, co-located with solar
Microsoft
Long-term power purchase agreement behind the 150 MW Carwarp project

Analysis updated Sep 22, 2026, reviewing Q2 FY2026. Prices delayed. Built with The Buildout’s published methodology. Not investment advice. No positions held. © The Buildout 2026.

More on CSIQ: Earnings recap