Jinko Solar Imp. & Exp. Co. v. United States

2026 CIT 30

Opinion

Slip Op. 26-30

UNITED STATES COURT OF INTERNATIONAL TRADE

JINKO SOLAR IMPORT AND EXPORT CO. LTD., ET AL.,

Plaintiffs, and

JA SOLAR TECHNOLOGY YANGZHOU CO., LTD., ET AL.,

Plaintiff-Intervenors, Before: Claire R. Kelly, Judge v. Consol. Court No. 22-00219 UNITED STATES, Defendant, and

AMERICAN ALLIANCE FOR SOLAR MANUFACTURING,

Defendant-Intervenor.

OPINION AND ORDER

[Sustaining in part and remanding in part Commerce’s redetermination.]

Dated: March 30, 2026

Ned H. Marshak, Dharmendra N. Choudhary, Jordan C. Kahn, Brandon M. Petelin, and Elaine F. Wang, Grunfeld Desiderio Lebowitz Silverman & Klestadt LLP, of New York, NY and Washington, D.C., for plaintiffs Jinko Solar Import and Export Co. Ltd., Jinko Solar Co., Ltd., Jinkosolar Technology (Haining) Co., Ltd., Yuhuan Jinko Solar Co., Ltd., Zhejiang Jinko Solar Co., Ltd., Jiangsu Jinko Tiansheng Solar Co., Ltd., Jinkosolar (Chuzhou) Co., Ltd., Jinkosolar (Yiwu) Co., Ltd., and Jinkosolar (Shangrao) Co., Ltd.

Jonathan M. Freed, Robert G. Gosselink, and Kenneth N. Hammer, Trade Pacific PLLC, of Washington D.C., for consolidated plaintiffs Trina Solar Co., Ltd., Trina Solar (Changzhou) Science & Technology Co., Ltd., Changzhou Trina Solar Yabang Energy Co., Ltd., Turpan Trina Solar Energy Co., Ltd., Trina Solar (Hefei) Science & Technology Co., Ltd., Changzhou Trina Hezhong Photoelectric Co., Ltd.

Jeffrey S. Grimson, Sarah M. Wyss, Bryan P. Cenko, Jill A. Cramer, Kristin H. Mowry, Savannah Rose Maxwell, and Yixin (Cleo) Li, Mowry & Grimson, PLLC, of Washington D.C., for consolidated plaintiffs and plaintiff-intervenors JA Solar Technology Yangzhou Co., Ltd., and Shanghai JA Solar Technology Co., Ltd.

Craig A. Lewis, Lindsay K. Brown, and Nicholas W. Laneville, I, Hogan Lovells US LLP, of Washington D.C., for consolidated plaintiff BYD (Shangluo) Industrial Co., Ltd.

Gregory S. Menegaz, Alexandra H. Salzman, and Vivien J. Wang, The Inter-Global Trade Law Group PLLC, of Washington D.C., for consolidated plaintiff and plaintiffintervenor Risen Energy Co., Ltd.

An Hoang and Kara M. Westercamp, Commercial Litigation Branch – Civil Division, U.S. Department of Justice, of Washington, D.C., for defendant United States. Of counsel were Fee Pauwels, Jack Dunkelman, and William M. Purdy, Office of the Chief Counsel for Trade Enforcement & Compliance, U.S. Department of Commerce, of Washington, D.C.

Timothy C. Brightbill, Laura El-Sabaawi, Stephanie M. Bell, and Paul A. Devamithran, Wiley Rein, LLP, of Washington D.C., for defendant-intervenor American Alliance for Solar Manufacturing.

Kelly, Judge: Before the Court is the U.S. Department of Commerce’s (“Commerce”) Final Results of Redetermination Pursuant to Court Remand, Sept. 26, 2025, ECF No. 147 (“Second Remand Results”) pursuant to this Court’s remand order in the 2019–20 administrative review of the antidumping duty (“AD”) order covering crystalline silicon photovoltaic cells, whether or not assembled into modules, from the People’s Republic of China (“China”). See Jinko Solar Imp. & Exp. Co. v. United States, 789 F. Supp. 3d 1275 (Ct. Int’l Trade 2025) (“Jinko II”); see also 87 Fed. Reg.

3

38,379 (Dep’t Commerce June 28, 2022), as amended by Crystalline Silicon Photovoltaic Cells, Whether or Not Assembled Into Modules, From the People’s Republic of China, 87 Fed. Reg. 48,621 (Dep’t Commerce Aug. 10, 2022) (amended final results) (“Final Results”) and accompanying Issues and Decision Memo., Oct. 5, 2022, ECF No. 24-5 (“Final Decision Memo.”).

BACKGROUND

The Court presumes familiarity with the facts set forth in Jinko Solar Import and Export Co. v. United States, 701 F. Supp. 3d 1367 (Ct. Int’l Trade 2024) (“Jinko I”) and Jinko II, and recounts only those pertinent to the instant matter. See generally Jinko I, 701 F. Supp. 3d 1367; Jinko II, 789 F. Supp. 3d 1275. On December 7, 2012, Commerce published the AD order on solar cells from China. See generally Crystalline Silicon Photovoltaic Cells, Whether or Not Assembled Into Modules from the People’s Republic of China, 77 Fed. Reg. 73,018 (Dep’t Commerce Dec. 7, 2012) (amended final determination). On February 4, 2021, Commerce initiated the eighth administrative review of the AD order. See generally Initiation of Antidumping and Countervailing Duty Administrative Reviews, 86 Fed. Reg. 8,166, 8,168–69 (Dep’t Commerce Feb. 4, 2021). Commerce selected Plaintiff Jinko Solar Import and Export Co., Ltd. (“Jinko” or “Plaintiff”) and Consolidated Plaintiff and Plaintiff-Intervenor, Risen Energy Co., Ltd. (“Risen”), as mandatory respondents. Respond. Select. Memo. at 1–5, PD 53, CD 5, bar code 4092031-01 (Feb. 25, 2021).

4

On December 23, 2021, Commerce published its preliminary results. See generally Crystalline Silicon Photovoltaic Cells, Whether or Not Assembled Into Modules, from the People’s Republic of China; 2019–20, 86 Fed. Reg. 72,923 (Dep’t Commerce Dec. 23, 2021) (preliminary results and partial rescission) (“Preliminary Results”) and accompanying preliminary issues and decision memo. (“Prelim. Decision Memo.”). On August 10, 2022, Commerce issued its final results. See generally Final Results; Final Decision Memo.

Because Commerce treats China as a nonmarket economy (“NME”) when calculating the dumping margins for the mandatory respondents, Commerce determined the surrogate value (“SV”) of the respondents’ entries of subject merchandise by using data from a surrogate market economy country (“surrogate country”) to value the factors of production (“FOP”). See Section 773(c)(4) of the Tariff Act of 1930, as amended, 19 U.S.C. § 1677b(c)(4). 1 Commerce chose Malaysia as the primary surrogate country for valuing all of respondents’ FOPs to construct normal value for purposes of calculating dumping margins. Prelim. Decision Memo. at 16– 19, 23–28; Final Decision Memo. at 18.

Commerce preliminarily valued respondents’ anti-reflective-coated (“AR-

coated”) solar glass using Romanian HTS 7007.19.80 import data, rather than Malaysia’s HTS 7007.19.90 data, finding that the Romanian data was more specific,

1 Further citations to the Tariff Act of 1930, as amended, are to the relevant provisions of Title 19 of the U.S. Code, 2018 edition.

5

reliable, and accurate to the input. [Commerce] Prelim. [SV] Memo. at 3, PD 403, bar code 4194750-01 (Apr. 16, 2021) (“Commerce Prelim. SV Memo.”); Final Decision Memo. at 15. Commerce applied partial facts available with an adverse inference to value Risen’s missing data after finding that Risen failed to provide all requested data and, by continuing to use suppliers that did not cooperate with Commerce’s requests, failed to cooperate to the best of its ability. Prelim. Decision Memo. at 15– 16; Final Decision Memo. at 8–13. Commerce calculated a rate using facts otherwise available with an adverse inference, which it determined to be “sufficiently adverse” to incentivize cooperation. Prelim. Decision Memo. at 15–16; Final Decision Memo. at 8–13.

Plaintiffs, Consolidated Plaintiffs, and Plaintiff-Intervenor challenged various aspects of the Final Decision Memo. before this Court, and on May 1, 2024, the Court issued Jinko I.2 See ECF Nos. 35–41; Jinko I, 701 F. Supp. 3d 1367.

2 In Jinko I, the Court sustained Commerce’s determinations regarding: (1) Consolidated Plaintiffs Trina Solar Co., Ltd., Trina Solar (Changzhou) Science & Technology Co., Ltd., Changzhou Trina Solar Yabang Energy Co., Ltd., Turpan Trina Solar Energy Co., Ltd., Trina Solar (Hefei) Science & Technology Co., Ltd., and Changzhou Trina Hezhong Photoelectric Co., Ltd.’s separate rate status; (2) the valuation of Plaintiff’s electricity, ocean freight, backsheet, and ethylene vinyl acetate (“EVA”); (3) the use of JA Solar Malaysia’s financial statements to calculate surrogate financial ratios; (4) the deduction of Section 301 duties; and (5) the application of facts available with an adverse inference in calculating Risen’s rate. Jinko I, 701 F. Supp. 3d at 1397. The Court remanded Commerce’s determinations regarding (1) the use of Romanian HTS 7007.19.80 to value solar glass; (2) the valuation of air freight; (3) the methodology for constructing Risen’s adverse inference rate and its consistency with 19 U.S.C. § 1677e; and (4) the separate rates for Plaintiff-Intervenors BYD and JA Solar in view of any changes to Jinko’s or Risen’s margins. Id. at 1397.

6

Commerce filed its first remand results on August 29, 2024. See Final Results of Remand Redetermination Pursuant to Court Remand, Aug. 29, 2024, ECF Nos. 89–90 (“Remand Results”). Plaintiff, Consolidated Plaintiffs, and Plaintiff- Intervenor submitted comments in opposition to Commerce’s remand determination on October 30, 2024. See ECF Nos. 97–101. On May 30, 2025, the Court issued Jinko II. 3 See generally Jinko II, 789 F. Supp. 3d 1275.

On August 18, 2025, Commerce filed its Draft Second Remand Results. Draft Results of Remand Determination, PD 2RCR 1, CD 2RPR 1, bar code 4812049-01 (Aug. 18, 2025) (“Draft Second Remand Results”). On September 26, 2025, Commerce filed its Second Remand Results. See generally Second Remand Results. On January 5, 2026, Risen, Jinko, BYD (Shangluo) Industrial Co., Ltd. (“BYD”), JA Solar Technology Yangzhou Co., Ltd., and Shanghai JA Solar Technology Co., Ltd. (“JA Solar”) submitted comments in opposition to Commerce’s determinations in the Second Remand Results. See generally Pl. Int./Consol. Pl. Cmts.; Pl. Cmts.; BYD Cmts.; JA Solar Cmts., Jan. 5, 2026, ECF Nos. 159–165. On January 6, 2026, Defendant-Intervenor American Alliance for Solar Manufacturing submitted comments in support of the Second Remand Results. See Def. Int. American Alliance

3 In Jinko II, the Court again remanded, finding Commerce (1) failed to substantiate Romanian data specificity and (2) did not adequately explain why it could not convert respondents’ glass consumption data to match Malaysia’s square meter-reported import values. Jinko II, 789 F. Supp. 3d at 1283–84. The Court sustained Commerce’s air freight valuation, but remanded Commerce’s application of facts available with an adverse inference and selection of the surrogate Romanian HTS data to value solar glass. Id. at 1285–86.

7

for Solar Manufacturing’s Cmts. Supp’n. [Second Remand Results], Jan. 6, 2026, ECF Nos. 166–167 (“Def. Int. Cmts.”). On February 25, 2026, Plaintiff, Defendant- Intervenor, and Defendant replied to comments on the Second Remand Results. See Jinko Reply to Cmts. on Remand Results, Feb. 25, 2026, ECF Nos. 172–173; American Alliance Reply to Cmts. on Remand Results, Feb. 25, 2026, ECF Nos. 174, 177 (“Def. Int. Reply”); Def. Reply to Cmts. on Remand Results, Feb. 25, 2026, ECF Nos. 175– 176 (“Def. Reply”).

JURISDICTION AND STANDARD OF REVIEW The Court has jurisdiction under 19 U.S.C. § 1516a(a)(2)(B)(iii) and 28 U.S.C.

§ 1581(c), which authorize judicial review of Commerce’s final determination in an administrative review of an AD order. 19 U.S.C. § 1516a(a)(2)(B)(iii); 28 U.S.C. § 1581(c). The Court will sustain Commerce’s determination unless it is “unsupported by substantial evidence on the record, or otherwise not in accordance with law.” 19 U.S.C. § 1516a(b)(1)(B)(i). “Substantial evidence is ‘such relevant evidence as a reasonable mind might accept as adequate to support a conclusion.’” Huaiyin Foreign Trade Corp. (30) v. United States, 322 F.3d 1369, 1374 (Fed. Cir. 2003) (quoting Consol. Edison Co. v. NLRB, 305 U.S. 197, 229 (1938)). Commerce must “examine the relevant data and articulate a satisfactory explanation for its action including a ‘rational connection between the facts found and the choice made,’” so that a reviewing court can discern whether Commerce reasonably connected the record evidence to its decision. Motor Vehicle Mfrs. Ass’n v. State Farm, 463 U.S. 29,

8

43 (1983); Burlington Truck Lines, Inc. v. United States, 371 U.S. 156, 168 (1962). The Court reviews a remand determination for compliance with its remand order. Xinjiamei Furniture (Zhangzhou) Co. v. United States, 968 F. Supp. 2d 1255, 1259 (Ct. Int’l Trade 2014).

DISCUSSION

On second remand, Commerce continues to value respondents’ solar glass using Romanian HTS 7007.19.80 rather than data from the primary surrogate country, Malaysia, under HTS 7007.19.90. See generally Second Remand Results. Commerce also continues to calculate Risen’s rate using partial facts available with an adverse inference. See generally id. For the following reasons, Commerce’s use of Romanian HTS data is remanded for further consideration, Commerce’s calculation of Risen’s rate using facts available with an adverse inference is sustained, and Commerce’s determination of separate rates is remanded for reconsideration. I. Solar Glass Valuation Jinko and Risen argue that Commerce’s decision to value solar glass using Romanian HTS 7007.19.80, rather than Malaysian HTS 7007.19.90, is unsupported by substantial evidence. See Jinko Cmts. at 6–9, 23; Risen Cmts. at 5–8. Jinko and Risen argue: (1) Commerce’s interpretation of the exclusions in Romanian HTS 7007.19.80 is not supported by substantial evidence, and AR-coated solar glass is excluded from Romanian HTS 7007.19.80; and (2) Commerce fails to provide a reasonable explanation for why it cannot convert respondents’ record data from per-

9

piece to square meters. See Jinko Cmts. at 6–9, 23; Risen Cmts. at 5–8. Commerce’s determination is remanded for further consideration.

A. Romanian Data Commerce determines antidumping duties by calculating the amount by which the normal value of subject merchandise exceeds the export price or constructed export price. 19 U.S.C. § 1673. Normal value generally is the price at which the producer or exporter sells the subject merchandise in the ordinary course of trade in its home market, or, in certain circumstances, in a third-country market. 19 U.S.C. § 1677b(a)(1). When the merchandise is exported from an NME country, however, Commerce does not base normal value on sales. 19 U.S.C. § 1677b(c)(1). Instead, Commerce determines normal value using “the value of the factors of production utilized in producing the merchandise,” plus “an amount for general expenses and profit,” and including “the cost of containers, coverings, and other expenses.” Id.

To value a respondent’s factors of production, Commerce relies on surrogate values from market economy countries that are both economically comparable to the NME country and significant producers of comparable merchandise. 4 19 U.S.C. § 1677b(c)(4). Commerce’s regulations, consistent with the statutory scheme, express

4 In selecting a primary surrogate country, Commerce considers: (1) economic comparability to the NME country; (2) production of comparable merchandise; (3) whether such countries are significant producers of that merchandise; and (4) the quality and availability of FOP data. See Import Admin., U.S. Dep’t of Commerce, Non-Market Economy Surrogate Country Selection Process, Pol’y Bulletin 04.1 at 1 (Mar. 1, 2004) (“Pol’y Bulletin 04.1”).

10

a preference to value all factors, to the extent possible, using a single surrogate country. 19 C.F.R. § 351.408(c)(2). In selecting surrogate values, Commerce must use “the best available information regarding the values of such factors in a market economy country or countries considered to be appropriate.” 19 U.S.C. § 1677b(c)(1). Commerce evaluates whether proposed surrogate data constitute the best available information by considering specificity to the input, tax and duty exclusivity, contemporaneity with the period of review (“POR”), whether the data reflect a broad market average, and public availability. Pol’y Bulletin 04.1 at 1.

In Jinko I, the Court remanded Commerce’s decision to use Romanian HTS data to value Jinko’s and Risen’s solar glass. Jinko I, 701 F. Supp. 3d at 1367, 1381– 82. The Court held that Commerce did not support its determination with record evidence. Id. at 1397. In particular, the Court noted that Romanian HTS 7007.19.80 specifically excludes glass with an absorbent layer:

Enamelled, Coloured Throughout The Mass, pacified, Flashed Or With An Absorbent Or Reflecting Layer, Glass Of Size And Shape Suitable For Incorporation In Motor Vehicles, Aircraft, Spacecraft, Vessels And Other Vehicles.

Id. at 1383; see also Am. Alliance for Solar Mfr. Pre-Prelim. Cmts. & Subm. [SV] at Ex. 9, PD 334, CD 433, bar code 4149598-03 (Aug. 3, 2021) (“Pet. Pre-Prelim. SV Cmts. at Ex. 9 (Romanian HTS)”) (containing the Romanian HTS subheading); see also Final Decision Memo. at 13 n.42). Commerce defines absorbent as “something that takes in without releasing” light. Jinko I, 701 F. Supp. 3d at 1383 (citing Final Decision Memo. at 17). Record evidence indicates Jinko’s solar glass, because of the

11

AR-coating, “captures and retains light.”5 Id. at 1383 (citing Jinko’s Sect. A & App’x XI Questionnaire Resps. (“Jinko AQR”) at Ex. A-8C, A-12, PD 120–26, CD 77–106, bar code 4108310-01 (Apr. 8, 2021) (Jinko’s polycrystalline module “feature[es] new glass technology [that] improves light absorption and retention”)). Thus, the Court reasoned Jinko’s AR-coated glass falls within “both Commerce’s definition of ‘absorbent’ and also the Romanian HTS exclusion.” 6 Jinko I, 701 F. Supp. 3d at 1383. The Court held that Commerce’s method for valuing solar glass was unsupported and remanded its decision for reconsideration or further explanation. Id.

In its first determination on remand, Commerce continued to use the Romanian HTS for its solar glass valuation. See Remand Results. In Jinko II, the Court explained that on remand, Commerce claimed that the excluded exemplars under Romanian HTS 7007.19.80 all share “light-limiting characteristics.” See Jinko II, 789 F. Supp. 3d at 1281–82 (citing Remand Results). Commerce found that Jinko’s solar glass lacks “light-limiting characteristics” and therefore, it concluded that the

5 Jinko explains that the anti-reflective coating “pull[s] in light photons instead of reflecting them back.” Jinko Cmts. at 10 (citing Jinko First Remand Comments (Oct. 30, 2024), ECF Nos. 98–99 (citing Jinko First Supplemental Questionnaire Response, PD 277-88, CD 365-91, Ex. SA-2D barcode 4146749-02, (Jul. 27, 2021)). The coating “triggers greater absorption of light photons and consequently, a higher transmission, as more photons pass through the glass and strike the underlying cells, thereby improving the performance of the solar panels.” Jinko Cmts. at 10. Jinko follows Commerce’s definition of absorption as relating to a sponge. Id. at 11. Like a sponge with water, the subject solar glass absorbs light and then releases it. Id. 6 In Jinko I, the Court identified that Commerce argued common characteristics

among the exemplars but altogether failed to address “the remainder of the exclusionary language contained in Romanian HTS 7007.18.80.” Jinko I, 701 F. Supp. 3d at 1383 n.15.

12

glass is not excluded from the Romanian HTS. See id. (citing Remand Results). The Court concluded that Commerce had failed to address record evidence that detracts from its determination because Commerce’s conclusion that all exemplars had “light limiting characteristics” ignored the included exemplars of “glass suitable for incorporation in motor vehicles, aircraft, spacecraft, vessels and other vehicles” from the HTS. Id. at 1282 (citing Remand Results); see also Pets. Pre-Prelim. SV Cmts. at Ex. 9 (Romanian HTS).

On second remand, Commerce fails to explain why the Romanian data is specific to solar glass and does not address evidence undermining its conclusion. The Romanian HTS exclusion covers glass with an absorbent layer, and record evidence indicates that the subject glass’s AR coating, “captures and retains light.” Jinko I, 701 F. Supp. 3d at 1383 (citing Jinko AQR at App’x XI, PD 148–52, CD 186–68, bar code 4108310-01, (Apr. 8, 2021)). Indeed, a “key feature” of Jinko’s solar glass is that it “improves light absorption and retention.” See Jinko AQR at App’x XI, PD 148–52, CD 186–68, bar code 4108310-01, (Apr. 8, 2021); [Risen’s] Sect. D Questionnaire Resp. at App’x XIII:7, PD 147, CD 122, bar code 4116609-01 (Apr. 30, 2021) (“Risen Sect. D Resp.”).

Commerce’s attempt to construe “absorbent” contrary to its plain meaning by arguing that all, or some subset, of the exemplars are “light-limiting” lacks merit. In both the Final Decision Memo. and the Remand Results, Commerce explains that,

13

although the phrase “absorbent or reflecting layer” is not itself defined, it appears within a broader exclusion, specifically:

Enamelled, Coloured Throughout The Mass, Opacified, Flashed Or With An Absorbent Or Reflecting Layer.

Final Decision Memo. at 17–18. Commerce continues, stating “Enameled, colored, opacified (made opaque), and flashed (colored) are all treatments to the glass that limit the amount of light that passes through the glass.” Id. Although Commerce stated that it quoted the exclusion in full, it omitted part of the exclusion language. The Romanian HTS subheading, in its entirety, reads:

Toughened [Tempered] Safety Glass (Excl. Enamelled, Coloured Throughout The Mass, Opacified, Flashed Or With An Absorbent Or Reflecting Layer, Glass Of Size And Shape Suitable For Incorporation In Motor Vehicles, Aircraft, Spacecraft, Vessels And Other Vessels.

Pets. Pre-Prelim. SV Cmts. at Ex. 9 (Romanian HTS). In the Final Decision Memo., Commerce interpreted “absorbent” in the Romanian HTS to mean “light-limiting,” and on that basis concluded that “glass with an absorbent layer” takes in light without allowing it to pass through the glass. Final Decision Memo. at 9. Commerce therefore determined that the subject solar glass was not excluded. Id. Jinko I questioned Commerce’s conclusion that the plain meaning of “absorbent” did not encompass the glass at issue and how Commerce could maintain that all the exemplars supported its interpretation of “absorbent,” given that Commerce failed to address all of them. Jinko I, 701 F. Supp. 3d at 1383 n.15 (noting Commerce omitted

14

from its analysis of the language “Glass Of Size And Shape Suitable For Incorporation In Motor Vehicles, Aircraft, Spacecraft, Vessels And Other Vessels”).

Again, in the first Remand Results, Commerce argued that the Romanian HTS exclusions were “light-limiting” exclusions, and, on that basis, that “absorbent” means light-limiting. Remand Results at 9. Jinko II directed Commerce on remand to explain how its view that the exemplars all share “light-limiting characteristics” could be reconciled with “the remaining exemplars in HTS 7007.19.80, i.e., glass suitable for incorporation in motor vehicles, aircraft, spacecraft, vessels and other vehicles.” Jinko II, 789 F. Supp. 3d at 1282.

In the Second Remand Results, Commerce adheres to the same interpretive syllogism. See Second Remand Results at 22–24. It invokes the principle that, in a list of examples, all the examples share common qualities. See id.; see also Remand Results at 36–37. 7 Commerce asserts that all the listed exemplars share light- limiting characteristics and then applies that premise to conclude that every exclusion, including “absorbent,” is “light-limiting.” See Final Decision Memo. at 17; Remand Results at 34–35; Second Remand Results at 23. Twice, the Court has explained to Commerce that not all exemplars are “light-limiting.” See Jinko I 701 F. Supp. 3d at 1383; Jinko II 789 F. Supp. 3d at 1282. Arguably, “absorbent” is not light-limiting, but light-capturing. “Glass suitable for incorporation in motor

7 One formulation of that principle is “noscitur a sociis,” meaning that “a word is known by the company it keeps.” Noscitur a sociis, Black’s Law Dictionary (12th ed. 2024).

15

vehicles, aircraft, spacecraft, vessels and other vehicles” does not appear to be light limiting. See Pets. Pre-Prelim. SV Cmts. at Ex. 9 (Romanian HTS). Commerce’s latest effort to preserve its flawed syllogism is to divide the exclusion’s list of exemplars into groups, which abandons its major premise that all the exemplars share a single common characteristic. See Second Remand Results at 23. The exemplars are not all light-limiting. Commerce’s division of the Romanian HTS exclusions into two general categories of glass is illogical. See id. (explaining that some exemplars are defined by purpose, where others are defined by “treatments that limit the amount of light that is transmitted through the glass”). Commerce’s approach is an unreasonable interpretation of the exclusions and cannot overcome the plain meaning of the term “absorbent.”

Commerce has not provided an explanation, substantiated by record evidence, that would permit the Court to find its use of the Romanian HTS reasonable. Although the Court has twice remanded this issue for further explanation, Commerce still does not reasonably explain why the plain meaning of “absorbent layer” in the Romanian HTS exclusions does not encompass the solar glass at issue, where record evidence indicates that the glass has a coating that absorbs light. 8 See Risen Sect. D

8 In its reply to comments on the Second Remand Results, Defendant-Intervenor argues that Commerce’s use of Romanian HTS 7007.19.80 is reasonable because it “covers tempered glass used in solar panels; the specific type of glass used by solar panel producers.” Def. Int. Reply to Comments on Second Remand Results at 8, Feb.

(footnote continued)

16

Resp. at App’x XIII:7. Because the record contains evidence that the Romanian HTS classification excludes anti-reflective solar glass, Commerce cannot reasonably rely on the Romanian HTS to calculate the normal value of Jinko’s solar glass. On remand, if necessary, Commerce may reopen the administrative record to obtain additional information needed to select a surrogate value for solar glass.

B. Reliability of Malaysian Data Even if the Romanian kilogram-based HTS data did not exclude solar glass, Commerce still has not explained why the Malaysian HTS 7007.19.90 square meter data is unreliable. In Jinko I, the Court held that Commerce failed explain why data from the primary surrogate country, Malaysia, is unreliable. Jinko I, 701 F. Supp. 3d at 1382 (citing 19 C.F.R. § 351.408(c)(2)). The Court remanded for Commerce to reconsider or further explain why it could not convert respondents’ solar glass consumption from pieces to square meters, despite record evidence indicating that conversion was possible. Id. (citing Risen Sect. D Resp. Ex. D-34; Jinko AQR Ex. AD-

25, 2026, ECF No. 174 (“Def. Int. Reply”). However, Defendant-Intervenor relies on record evidence from a prior administrative review covering the 2017–18 POR. See id. at 8 (citing Issues & Decision Memo. accompanying Crystalline Silicon Photovoltaic Cells, Whether or Not Assembled Into Modules, From the People’s Republic of China, 85 Fed. Reg. 62,275 (Dep’t Commerce Oct. 2, 2020) (final results of antidumping duty admin. rev. and final deter. of no shipments; 2017-2018) at 26 n.119). As the Court noted in Jinko I, “‘each administrative review is a separate segment of proceedings with its own unique facts,’ and thus stands on its own record.” Jinko I, 701 F. Supp. 3d at 1382–83 n.14 (quoting Peer Bearing Co.-Changshan v. United States, 587 F. Supp. 2d 1319, 1325 (Ct. Int’l Trade 2008)). On this record, no evidence shows that the Romanian HTS subheading Commerce used is specific to the type of glass used by solar panel producers.

17

9). In Jinko II, the Court again directed Commerce to explain why it could not use record glass dimensions to convert Jinko’s and Risen’s per-piece glass consumption into square meters and use the Malaysian HTS 7007.19.90 surrogate values, which are reported in square meters. Jinko II, 789 F. Supp. 3d at 1283–84; Second Remand Results at 3.

In the Draft Second Remand Results, Commerce applies an average conversion ratio to convert respondents’ glass consumption data into square meters for comparison with the Malaysian HTS values, but it does not perform the conversion on a CONNUM-specific basis using the actual record dimensions rather than an average ratio. Draft Second Remand Results at 3–4. At the same time, Commerce acknowledges that the record contains respondents’ glass consumption data in square meters, states that the Malaysian values could be used, and demonstrates the conversion needed to align respondents’ reported data with the Malaysian square meter values. See id. Commerce explained:

Upon further review, the record contains the necessary information to determine Jinko and Risen’s solar glass consumption in square meters, which is the unit of measure used to report import values under Malaysian HS number 7007.19.90. Specifically, the record contains inventory records for Jinko and Risen that show the number of pieces of each type of glass withdrawn from inventory for production during the POR and the length and width of each piece of glass, in millimeters.

Using this information, we were able to calculate the total square meters of glass consumed in production by each respondent during the POR.

18

Id. at 3. Despite having respondents’ glass dimensions, Commerce chose to do an average conversion ratio. 9

Because each respondent already converted their total glass consumption in pieces to total glass consumption in kilograms for reporting to Commerce in the underlying review, we used the ratio of the total glass consumption in square meters to the total glass consumption in kilograms to calculate the number of square meters of glass used for each kilogram of glass consumed in production. We multiplied this square-meter per kilogram conversion ratio by the number of kilograms of glass consumed per watt of finished product, which was reported to Commerce on a control number (CONNUM)-

specific basis, to calculate the square meters of glass consumed per watt of finished product for each CONNUM reported in the FOP database.

Given that we now have glass consumption in square meters for both respondents, we can use import values under Malaysian HS number 7007.19.90, which are reported in Malaysian Ringgits (RM) per square meter, to value the respondents’ consumption of solar glass. Therefore, consistent with the CIT’s opinion, Commerce’s selection of Malaysia and the primary surrogate country, and Commerce’s preference to value all FOPs using SVs from a single surrogate country, we have recalculated the value of Jinko and Risen’s solar glass consumption using the value of imports during the POR under Malaysian HS number 7007.19.90 in this redetermination.

Id. at 3–4 (footnotes omitted). 10 In its final Second Remand Results, Commerce explains that, upon further review, it determined the conversion method used in the

9 Commerce explained its choice in the final Second Remand Results, stating that “[g]iven the complexities of determining CONNUM-specific conversion rates, in the Jinko II Draft Redetermination, we determined that it was reasonable to use an average conversion rate.” Second Remand Results at 21. 10 Commerce’s Draft Second Remand Results answers the Court’s question whether

respondents’ data can convert consumption to square meters. See Draft Second Remand Results at 3–4; see also Jinko II, 789 F. Supp. 3d at 1283–84 (directing Commerce to explain why it could not convert to square meters). Commerce

(footnote continued)

19

Draft Second Remand did not account for the fact that the respondents’ solar glass varies in thickness and therefore, conversion to Malaysian square meters produced “distorted consumption quantities.” Second Remand Results at 4–5, 19–21 (“. . . we find that the factor used in the Jinko II Draft Redetermination to convert the respondents’ solar glass consumption quantities into a unit of measure that was compatible with the Malaysian import prices for solar glass did not take into account the varying thicknesses of the respondents’ solar glass, and thus, results in distorted consumption quantities for solar glass”). Commerce also explains that nothing on the record reveals the thickness of Malaysian glass. Id. at 19. Therefore, even if one could construct the thickness of the respondents’ glass from the data’s dimensions on the record, it is possible that using the Malaysian FOP data to value the glass might result in value distortions. See id.

acknowledges that respondents reported consumption in pieces, and that the record contains the length and width needed to calculate area per piece and total square meters consumed. See Second Remand Results at 4 (citing Risen Case Br., PD 421, CD 481 bar code 4206747-01 (Jan. 28, 2022); Jinko Case Br., PD 494, CD 446, bar code 4246241-01 (May 27, 2022)); Risen Cmts. at 24, Ex. 1. Commerce also acknowledges that the record captures thickness variations through respondents’ conversions to weight in kilograms. Second Remand Results at 6. As Commerce explained, Jinko and Risen calculated glass weight by multiplying glass density (kg/m3) by the cubic meters of glass consumed (m3), using each piece’s length, width, and thickness. Id. Commerce further states that the record contains inventory data showing the number of pieces withdrawn for production during the POR, along with each piece’s length and width, which Commerce used to calculate total square meters (m2) consumed. Second Remand Results at 4. Commerce then used the ratio of total square meters consumed to total kilograms consumed to determine the number of square meters per kilogram (m2/kg) of glass consumed in production. Id.

20

The weights in kilograms of respondents’ glass are on the record, and therefore, as Commerce demonstrated in the Draft Second Remand Results, an average conversion factor method is feasible. Commerce chose to use an average conversion factor in its calculation because it concluded that a CONNUM-specific conversion would be unreasonable on this record due to its “complexity,” but then rejects its own methodology as inaccurate. See id. at 21 (stating that Commerce used an average conversion factor “given the complexities” of using CONNUM-specific conversion rates). At no point does Commerce identify or explain the complexities that prevented it from calculating CONNUM-specific rates on this record. See generally Second Remand Results. Thus, it is unclear to the Court why CONNUM-specific rates are too complex to be reasonable, and the matter is remanded for further consideration or explanation on this point.

Commerce’s rejection of the methodology used in the Draft Second Remand Results as distortive also requires further consideration or explanation. Commerce’s concern that it lacks thickness data overlooks the CONNUM-specific weight, length, and width data already on the record, which is sufficient to account for thickness when converting respondents’ glass data into square meters. See Second Remand Results at 19–21. In the Draft Second Remand Results, Commerce derives an overall conversion factor in square meters per kilogram (m²/kg) from respondents’ total glass consumption, applies that factor to each CONNUM’s reported glass usage in kilograms per watt (kg/watt) to obtain CONNUM-specific quantities in square meters

21

per watt (m²/watt), and then multiplies those quantities by Malaysia’s surrogate values in Malaysian ringgits per square meter (RM/m²) to value the glass input for each CONNUM in RM/watt. See Draft Second Remand Results at 3–4. Because that conversion factor is based on both area and weight, it necessarily captures differences in glass thickness. See id. Commerce therefore does not reasonably explain its conclusion in the Second Remand Results that converting respondents’ data to square meters is distortive because glass thicknesses vary by CONNUM. See Second Remand Results at 19. 11 To be fair, Commerce’s second remand analysis also turns on its view that Malaysia’s surrogate values, expressed in square meters, are not a reasonable proxy for respondents’ solar glass because there is no way of knowing the thickness of the glass associated reported in square meters. Id. at 19–21.

Moreover, there is nothing on the record demonstrating that the overall average value of glass per square meter calculated from the Malaysian import data is accurate for Risen or Jinko since there is no information on the record showing that the glass imported into Malaysia during the POR had the same thicknesses (and the same quantities of glass for each

11 In its comments on the Second Remand Results, Risen proposes a two-step methodology linking “job orders” in “Step 8.1” to thickness and weight data in “Step 8.2” to derive CONNUM-specific thickness and conversion ratios. Risen Cmts. at 5– 8. In its reply, Defendant argues that Risen’s proposed methodology is flawed because some CONNUMs reported for sales of subject merchandise during the POR do not appear in Step 8.2 of Risen’s cost reconciliation. Risen Cmts. Ex. 1. (Step 8.2).” Def. Resp. at 13. However, Commerce’s draft methodology already suggests that respondents’ own data could yield a reliable overall m²/kg factor and, from that, a square meters per watt conversion for Malaysian inputs. Draft Second Remand Results at 3–4. Commerce did so without relying on Step 8.2 and without identifying missing CONNUM-level square meter data as necessary to perform the conversion. Id.

22

thickness) as the glass used by Risen and Jinko. Therefore, even on an overall average basis, and not a CONNUM-specific basis, we have no reason to conclude that including a glass value in normal value based on Malaysia import data is accurate.

Id. at 19. Although the Draft Second Remand Results methodology accounts for variation in respondents’ glass thickness, it cannot eliminate distortion arising from variation in Malaysian glass thickness because the record lacks the necessary Malaysian thickness and weight data. Even so, Jinko points to record evidence supporting the comparability of Malaysian glass thickness to respondents’ glass. See Jinko Reply at 3. Jinko notes that “all three glass’ conversion factors available on the record . . . corroborate each other.” Jinko Cmts. at 22–23. According to Jinko, this corroboration constitutes substantial evidence that glass imported under Malaysian HTS 7007.19.80, as part of the same pool of internationally traded glass, shares physical characteristics similar to Jinko’s glass, including average thickness. Id. at 23. Commerce has not addressed this evidence, which detracts from the support for its determination.

II. Calculation of Risen’s Rate Using Facts Available with an Adverse Inference

On remand, Commerce modifies and further explains its methodology for calculating Risen’s rate using facts available with an adverse inference. Second Remand Results at 25–33. Risen argues that Commerce’s revised methodology is inaccurate and based on punishment not cooperation. Risen Cmts. at 16–17. Commerce’s revised methodology, however, is in accordance with law, and its resulting rate determination is supported by substantial evidence.

23

Section § 1677e of Title 19 governs Commerce’s use of facts otherwise available.

19 U.S.C. § 1677e. Under § 1677e(a), Commerce shall apply facts otherwise available where necessary information is not available on the record, subject to the requirements of 19 U.S.C. § 1677m(d). 19 U.S.C. §§ 1677e(a); 1677m(d). If Commerce further finds that a party failed to cooperate “to the best of its ability,” it may apply an adverse inference in selecting from among the facts otherwise available. 19 U.S.C. § 1677e(b)(1)(A). When applying an adverse inference, Commerce is limited to relying on information “derived from” the record. 19 U.S.C. § 1677e(b)(2). In Jinko I and Jinko II, the Court explained that the “derived from” requirement demands a traceable, logical link between the selected inference and the underlying source material. Jinko I, 701 F. Supp. 3d at 1395–96 (citing 19 U.S.C. § 1677e(b)(2)); Jinko II, 789 F. Supp. 3d at 1287–88 (citing 19 U.S.C. § 1677e(b)(2)).

Commerce must exercise its methodological discretion reasonably, and it must explain how its selection furthers the statutory purpose of encouraging cooperation. See Vicentin S.A.I.C. v. United States, 404 F. Supp. 3d 1323, 1342–43 (Ct. Int’l Trade 2019), aff’d, 42 F.4th 1372, 1382 (Fed. Cir. 2022) (holding Commerce must explain on the record how it effectuates a statute’s remedial purpose where the same market distortion appears to be addressed in a concurrent CVD proceeding). Commerce must also explain why its imposition of an adverse inference is nonpunitive and grounded in the record. F.lli De Cecco Di Filippo Fara S. Martino S.p.A. v. United States, 216

24

F.3d 1027, 1032 (Fed. Cir. 2000); Gallant Ocean (Thai.) Co. v. United States, 602 F.3d 1319, 1323 (Fed. Cir. 2010).

In Jinko II, the Court remanded Commerce’s first remand determination regarding its application of facts available with an adverse inference to calculate Risen’s rate as contrary to law and unsupported by substantial evidence. Jinko II, 789 F. Supp. 3d at 1286–87. In the prior remand, Commerce described its methodology as follows:

For each reported input with missing data, Commerce calculated a ratio for each monocrystalline solar module CONNUM by dividing the per-

unit consumption quantity of the input reported for the CONNUM by the highest per-unit consumption quantity reported for that input for any monocrystalline solar module. Commerce then averaged all the CONNUM-specific ratios that it calculated for that input to derive one single ratio for the input (the input ratio). Commerce than calculated:

(1) a simple average of all the input ratios for all the inputs (i.e., which include direct materials, direct and indirect labor, electricity, gas, and water) that were used to produce solar modules (including, where applicable, input ratios for inputs, other than solar cells, that were used to produce components in the solar module); (2) another simple average of all the input ratios for all the inputs (i.e., which include direct materials, direct and indirect labor, electricity, gas, and water) that were used to produce solar cells; and (3) a third simple average of all the input ratios for all the material inputs that were used in packing solar modules. Commerce then multiplied each reported per-unit consumption quantity by the multiplicative inverse of the applicable simple average input ratio (i.e., 1 divided by the applicable simple average input ratio) to increase all the reported consumption quantities as an adverse inference.

Remand Results at 22 (footnotes omitted). Thus, Commerce calculated the ratio by dividing each consumption ratio by the highest consumption ratio. Id. It then took a simple average of those ratios within each of three input categories and multiplied

25

each category’s average by the inverse of the first ratio. Id. The Court faulted Commerce’s explanation because it suggested that the resulting rate increase flowed from the multiplicative inverse itself, rather than from record-based information. Jinko II, 789 F. Supp. 3d at 1286–87. Commerce did not explain the purpose of the multiplicative inverse beyond stating that it was used to “increase all the reported consumption quantities as an adverse inference.” Remand Results at 22. The Court therefore directed Commerce to explain how the inverse is grounded in record evidence, rather than an arbitrary figure, why Commerce could not rely on actual record data, and why its grouping methodology was reasonable. Jinko II, 789 F. Supp. 3d at 1286–87.

Here, on remand, Commerce makes a slight methodological adjustment, and, more importantly, provides the previously missing explanation for a key component of its formula. Commerce explains that its formula seeks to identify:

. . . by what number do we need to multiply CONNUM-specific per-unit consumption quantities for inputs that are an average . . . of the highest reported CONNUM-specific per-unit consumption quantities of the inputs to adjust them, on average, to the highest reported CONNUM-

specific per-unit consumption quantities for those inputs . . . .

Second Remand Results at 8. In other words, Commerce explains that the purpose of its formula is to measure the spread between the average consumption quantity and the highest reported consumption quantity, express that spread as a percentage, and apply that percentage as an adverse inference when filling missing information

26

with facts available. Commerce lists the steps it uses to construct this inference as follows:

We first isolated the highest reported consumption quantity for a given input in any CONNUM. Then, we averaged the reported consumption quantities for the remaining CONNUMs. From there, we calculated the percentage difference between the highest reported consumption quantity and the average of the other reported consumption quantities for each input.

Id. at 13–14. Commerce explains that it calculates the “percentage difference between the highest reported consumption quantity and the average of the other reported consumption quantities,” and then adds that percentage difference to the input ratios used for missing data. Id. Commerce further explains that this percentage difference measure estimates the increase needed to replicate the missing data by using “the variation in Risen’s CONNUM-specific per-unit consumption quantities to calculate the amount by which the uncooperative supplier’s consumption quantities could exceed those of Risen.” 12 Id. Commerce has now provided the explanation the Court requested, showing that the rate is derived from

12 Although Commerce now uses a formula that does not rely on a multiplicative inverse, its explanation here confirms that its prior remand’s use of the multiplicative inverse was reasonable. Commerce asks, what is the percentage relationship between the highest consumption rate and all the other rates? See Second Remand Results at 9–10. That percentage is determined by dividing the average consumption rates of all the other rates by the highest rate. Id. For example, the average rate might be 0.75 (75%) of the highest rate. Commerce then asks what multiplier one would use to bring that average rate up to the highest rate. The highest rate’s relationship to itself is necessarily 1 (100%). Converting a fraction or percentage to 1 requires taking its multiplicative inverse. Commerce’s explanation that it is identifying the multiplier needed to move each consumption rate up to the highest consumption rate therefore makes clear that its methodology is reasonable.

27

record evidence reflecting variation in reported consumption quantities, rather than from an arbitrary figure.

Commerce has also adequately explained why it declined to use Risen’s highest reported consumption quantities as facts available with an adverse inference to induce cooperation. Commerce found that relying on the highest record value alone was inadequate because, across multiple reviews, uncooperative suppliers remained in the proceeding despite Commerce’s use of the highest consumption rate on the record. Id. at 31. Commerce further explained that, although it had used Risen’s highest reported consumption quantities as facts available with an adverse inference in prior reviews to induce cooperation and had warned that it would examine continued reliance on uncooperative suppliers, Risen continued using such suppliers across four consecutive administrative reviews, including the 2019–20 POR at issue. Id. at 31–32. Commerce therefore reasonably concluded that, if the highest reported consumption quantities had been sufficient to induce cooperation, Risen would not have continued using uncooperative suppliers, and that a different adverse inference methodology with a built-in increase was warranted to deter continued non- compliance. Id. at 33–34. Commerce also explained that increasing the percentage as a built-in incentive was reasonable in light of repeated noncompliance across multiple reviews. Id. at 29. Commerce may apply an adverse inference that includes a built-in incentive to encourage cooperation, so long as the selected rate remains anchored to the record and is not merely punitive. See F.lli De Cecco Di Filippo Fara

28

S. Martino S.p.A. v. United States, 216 F.3d 1027, 1032 (Fed. Cir. 2000). Contrary to Risen’s argument, Commerce’s methodology is not “divorced from the concerns of accuracy” and therefore does not result in improperly punitive margins. See Risen Cmts. at 16–17. Commerce reasonably concludes that the highest reported consumption quantity is not sufficient incentive and reasonably made a stronger adverse inference here because that increase is derived from record evidence rather than punishment for its own sake.

Finally, Commerce reasonably groups the hundreds of record inputs into three groups—solar cells, solar modules, and packing—and calculates separate average input ratios for each category. Second Remand Results at 27. Commerce explained that its FOP-specific approach is reasonable because it aligns adjustments with distinct stage-specific inputs and processes and avoids distortions that would result from applying a single average across both cell and module production. Id. at 33. Although Risen argues that this grouping is unreasonable because the inputs have different reported FOPs and surrogate values, see Risen Cmts. at 15, Commerce explains that cell and module production involve different inputs and processes, and that variability in cell inputs has no apparent relationship to module production. Second Remand Results at 33. Grouping inputs by production stage therefore reasonably promotes accuracy by avoiding skewing across stages while remaining workable on a record containing hundreds of inputs. In light of the foregoing, Commerce’s application of facts available with an adverse inference is reasonable.

29

III. Separate Rate Assignments for BYD and JA Solar Jinko II remanded to Commerce to recalculate the separate rates for BYD and JA Solar. Jinko II, 789 F. Supp. 3d at 1292. Plaintiff-Intervenor JA Solar contends that Commerce should set its separate rate equal to Jinko’s margin, and BYD contends that Commerce should adjust BYD’s separate rate to track any changes to Jinko’s and Risen’s margins. See Pl.-Int./Consol. Pl. Cmts., Jan. 5, 2026, ECF No. 159 at 4; BYD Cmts., Jan. 5, 2026, ECF No. 163 at 2, 6. Because Commerce found no margin changes on remand, it left both Plaintiff-Intervenors’ separate rates unchanged. Second Remand Results at 34. Following its redetermination, Commerce will necessarily reconsider whether it must recalculate the separate rates for JA Solar and BYD.

CONCLUSION

For the foregoing reasons, Commerce’s redetermination with respect to its decision to use Romanian glass data to calculate normal value is remanded for reconsideration consistent with this opinion. Commerce’s redetermination with respect to its methodology for calculating Risen’s rate using facts available with an adverse inference is supported by substantial evidence on the record and in accordance with the law and this Court’s remand order. In light of the foregoing, it is ORDERED that Commerce’s methodology for calculating facts available with an adverse inference is sustained; and it is further

30

ORDERED that Commerce’s determination with respect to its valuation of solar glass using Romanian HTS 7007.19.80 is remanded for reconsideration consistent with this opinion; and it is further ORDERED that Commerce’s determination of the review-specific rate applicable to JA Solar and BYD is remanded for reconsideration consistent with this opinion; and it is further ORDERED that Commerce shall file its redetermination with the Court within 90 days of this date; and it is further ORDERED that the parties shall have 30 days to file comments on the redetermination; and it is further ORDERED that the parties shall have 30 days to file their replies to the comments on the redetermination; and it is further ORDERED that the parties shall file the joint appendix, including the entire confidential record, within 14 days after the filing of replies to the comments on the redetermination; and it is further ORDERED that Commerce shall file the administrative record within 14 days of the date of filing its redetermination.

/s/ Claire R. Kelly

Claire R. Kelly, Judge

Dated: March 30, 2026 New York, New York