TMTB Asia: Overnight Wrap

2026-08-13 08:08Z · Yahoo

In-line US CPI, a bid across North Asia tech, and Hong Kong the odd one out. Korea led on memory; Taipei stayed risk-on through 46k; HSI slipped on Tencent even as HSTECH held up.

Setup。 美国 CPI 符合预期，北亚科技有买盘，香港是例外。韩国靠存储领涨；台北风险偏好还在，指数过 46k；恒指被腾讯拖累，HSTECH 还撑着。收盘 desk 口径：long Korea, short Hong Kong。

Korea — KOSPI +3.56% / KOSDAQ +0.29%. The index entered its first technical bull market since July’s rout, now ~+22% from the July 30 low. It was the fourth straight up day and the highest close since July 23. Memory did the work: Samsung Electronics +4.9% was the largest index contributor, while SK Hynix rose +6%. SEMCO +12% led the broader electronics complex. KOSDAQ lagged, with today’s bid concentrated in large-cap semis. The recovery may get tougher from here, with the 100-day MA the next test and turnover in memory-focused ETFs already cooling.

韩国 — KOSPI +3.56% / KOSDAQ +0.29%。 7 月暴跌以来首次进入技术性牛市，较 7 月 30 日低点约 +22%；连涨第四日，收盘为 7 月 23 日以来最高。存储干了活：三星电子 +4.9% 是指数最大贡献，SK 海力士 +6%。三星电机 +12% 带电子板块。KOSDAQ 落后 — 今天买盘集中在大盘半导体。街边：往上可能更难（下一个是 100 日均线；存储主题 ETF 成交已经在降温）。

Taiwan — TAIEX +1.11%. Winning streak, index through 46k. Mediatek +5.2% and Delta +5.3% led; Hon Hai −3% on post-earnings selling. Rotation: MLCC and wafers bid, ABF/CCL and CPO took a breather. Turnover +18% DoD. Non-tech was mixed and lagged — CPI and the MSCI review didn’t pull them along.

台湾 — TAIEX +1.11%。 连涨，指数过 46k。联发科 +5.2%、台达电 +5.3% 领涨；鸿海 −3% 业绩后被砸。轮动：MLCC 与硅片有买盘，ABF/CCL 与 CPO 歇一口气。成交额环比 +18%。非电两极分化、没跟上 — CPI 和 MSCI 季度调整也没把它们带起来。

Japan — NKY +1.16%. Quiet day, mostly tracking the US. Momentum outperformed by 1.5 sigma. AI infra was bid anyway — US peer prints, Korea’s chip rebound, NVIDIA’s new funding scheme — and in-line CPI took some of the “don’t add risk” worry off. Software and AI-disruption names lagged; some desks said the US software bounce already looked hot. Semis held up, autos didn’t.

日本 — NKY +1.16%。 安静一天，基本跟美股。Momentum 超 1.5 个 sigma。AI 基建没本地催化剂也有买盘 — 美股同业财报、韩国芯片反弹、NVIDIA 新融资安排；CPI 符合预期，也减轻了一点「别加风险」。软件和 AI 颠覆标的落后，有 desk 说美股软件反弹已经看起来过热。半导体还行，汽车没有。

China — CSI300 −0.57%. An early +0.6% gain reversed, with poor breadth: 1,679 of 2,225 Shanghai names fell. The TMT tell was a rotation into domestic AI infrastructure and semiconductor names: Ruijie Networks +13.5%, Huaqin +8.8%, Suzhou TFC Optical +7.1%, and Unisplendour +7.0%. The A/H premium on dual-listed mainland names widened to a near one-year high of ~23%.

中国 — CSI300 −0.57%。 早盘 +0.6% 没守住，宽度差（沪市 2,225 只里 1,679 只下跌）。TMT 看点是切向国产 AI 基建和半导体：锐捷网络 +13.5%、华勤 +8.8%、天孚通信 +7.1%、紫光股份 +7.0%。AH 溢价扩到近一年高位约 23%。

Hong Kong — HSI −0.17% / HSTECH +0.33%. Bucked the North Asia bid. Tencent −4.4% was the story — brokers cutting PTs after 2Q showed a capex step-up that flipped FCF negative (Hunyuan / WorkBuddy / Xiaowei). Lenovo ripped ~+20% to an ATH after 1H revenue +43% YoY. That’s the relative: Seoul chipmakers vs HK digesting Tencent spend.

香港 — HSI −0.17% / HSTECH +0.33%。 没跟上北亚。腾讯 −4.4% 是主线 — 二季度资本开支上台阶、FCF 转负（混元 / WorkBuddy / 小微），券商砍目标价。联想 1H 营收同比 +43%，涨约 +20% 创历史新高。相对就是：首尔芯片 vs 香港在消化腾讯开支。

Tencent (0700.HK, HK$441 −4.5%) — AI spend cycle: headline FCF vs monetization timing. Capex steps up from 2Q and stays high through 2027. The debate is whether monetization shows up before the spend hits earnings.

腾讯（0700.HK，441 港元 −4.5%）— AI 投入周期：报表自由现金流 vs 变现节奏。 二季度资本开支上台阶，强度看到 2027；争论在变现能否在开支拖累盈利之前跟上。

• 自由现金流 — 报道 −138 亿；剔算力预付款约 +376 亿。账面像烧钱，实质多半是 AI 基建采购时点 — 不是核心塌了。摩根大通用这个区分说明核心仍能资助，并提到 7000 亿+ 股权投资组合可作为可选资金。

• 资本开支周期 — 二季度约 530 亿 vs 一季约 320 亿；支付约 590 亿。不是一次性：下半年仍高，至少看到 2026–27。高盛最陡：2026–28E 2100 / 2450 / 2470 亿（原 1510 / 1650 / 1740），下半年再上台阶。摩根士丹利 2026 与 2027 各约 2000 亿（2026 原约 1400 亿）。摩根大通以 FY26 约 2000 亿为锚，明显超支会更谨慎。

• 钱花在哪 — 优先自用混元训练（Hy4/Hy5）、WorkBuddy / CodeBuddy、元宝 / 小微推理与微信 agent — 不是近端 AI 云出租。多余 GPU *可以* 走腾讯云，但管理层认为不太可能。

• 云 — 增速二十出头。高盛：TokenHub 8 月日 token 约 25 万亿（两月 5 倍）；AI 需求体现在 GPU 租用、MaaS、WorkBuddy/CodeBuddy，5 月后云价更硬。摩根大通上修 2026/27E 云至 21% / 33%（原 14% / 30%）。

• 变现节奏 — 摩根大通最乐观：「资助 AI 谷底」，2026 盈利增速约 2%，2027 起 10–15%；上修 2027E 广告至 27%（原 14%）；早期信号包括国内 token 正毛利、WorkBuddy 付费档毛利接近腾讯云。高盛目标价 700→670 港元，押长周期 franchise，认为近端 EPS 压力大半已在股价里。摩根士丹利目标价 650→550 港元，近端最悲观：2H26–27 非 IFRS 营业利润/净利大致持平，营业利润率 35.5% / 32.4% vs 2025 37.3% — 先消化，不是马上重估。

• FCF — Reported −Rmb13.8bn; ~+Rmb37.6bn excluding compute prepayments. Headline looks like cash burn; it’s mostly timing on AI infra buys, not the core collapsing. JPMorgan leans on that split and also flags the Rmb700bn+ equity portfolio as optional funding. Management: compute ordered a couple of months ago can now be sold at more than 30% profit.
• Capex cycle — 2Q ~Rmb53bn vs ~Rmb32bn in 1Q; payments ~Rmb59bn. Not a one-off: spend stays high into 2H26, multi-year through at least 2026–27. Goldman is steepest: 2026–28E Rmb210 / 245 / 247bn (from 151 / 165 / 174), another step-up in 2H26. Morgan Stanley ~Rmb200bn in both 2026 and 2027 (2026 from ~Rmb140bn). JPMorgan anchors FY26 at ~Rmb200bn and gets more cautious if spend runs well above that.
• What the spend is for — Own Hy training (Hy4/Hy5), WorkBuddy / CodeBuddy, Yuanbao / Xiaowei inference and agentic Weixin — not near-term AI Cloud rental. Excess GPUs *could* go through Tencent Cloud; management says that’s unlikely.
• Cloud — Growth in the low-twenties. Goldman: TokenHub ~25trn daily tokens by August (5x in two months); AI showing up in GPU rental, MaaS and WorkBuddy/CodeBuddy, pricing firmer since May. JPMorgan takes 2026/27E cloud growth to 21% / 33% from 14% / 30%.
• Monetization timing — JPMorgan is the most constructive: “funding the AI trough,” ~2% earnings growth in 2026 then 10–15% from 2027. Raises 2027E ads to 27% from 14%; early tells include positive GM on domestic tokens and WorkBuddy paid-tier GM roughly in line with Tencent Cloud. Goldman cuts PT 700→670, underwrites a long-duration franchise, sees the near-term EPS hit as largely in the tape. Morgan Stanley 650→550, most downbeat near term: flat non-IFRS OP/NP through 2H26–27, OPM 35.5% / 32.4% vs 37.3% in 2025 — digestion first, not a quick re-rate.

As a matter of fact, for the prepayment and for some of the compute orders that we had made just a couple of months ago, today, we can actually sell that at more than 30% profit.

— Tencent management

Expectations were already soft into the print, so this is less a missed “beat” than FCF going negative plus a bigger multi-year capex step — that’s the cliff feel. A lot of the buy-side chatter is where Tencent sits in a *short* framework: if Korea/Japan sentiment turns, Hong Kong can look like the funding short again, as it has before. Pure AI still sends people to the model names (MiniMax, Zhipu); cloud still sends them to Alibaba first. What’s left for Tencent is the old question with a sharper cash wrapper: the foundation model still isn’t convincing, products a bit better, monetization ROI still hard to underwrite.

业绩前情绪其实已经不太高了——所以这次更像是 FCF 转负、加上较大的多年度 capex 台阶带来的坍落感，而不是「预期差」本身。买方 chatter 里很多人在想：怎么在一个 short thesis 里给腾讯定位；如果韩国/日本情绪转好，香港可能又像以前一样变成 funding short。纯 AI exposure 市场更容易直接去找模型名（MiniMax、智谱）；云的话首先想到阿里。腾讯留下的还是老问题，只是现金流包装更刺眼：基模仍不给力，产品有些起色，货币化回报率一样不好回答。

DeepSeek V4-Pro: messy drop vs a strong Flash. @teortaxesTex calls the V4-Pro-0813 release “really weird”: a WeChat leak doesn’t even match Artificial Analysis directionally (Flash −2.7%, Pro −9%). Flash usage is still making new highs — @jayair’s chart at 15T tokens.

DeepSeek V4-Pro：相对 Flash 的一次乱发布。 @teortaxesTex 称 V4-Pro-0813 很怪 — 微信流出与 Artificial Analysis 连方向都不一致（Flash −2.7%，Pro −9%）。Flash 用量仍在创新高：@jayair 图上已经到 15T token。

Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)

@teortaxesTex

Really weird things going on with V4-Pro-0813 release
I will withhold judgement until they admit it's rolled out and post their own evals in English. Currently, the WeChat leak does not match AA even directionally (–2.7% for Flash, –9% for Pro). I doubt this is their best.

Aug 13, 2026 · Open on X

Jay

@jayair

jesus christ 15T

Aug 13, 2026 · Open on X

V4 Pro disappoints. Expectations were high *because* Flash was genuinely good, and the programmer-community read on Pro is poor. That’s some relief for listed China model names — MiniMax (0100.HK, HK$376.80 +5.4%) and Zhipu / Z.ai (2513.HK, HK$1,317 +9.0%) — which had been sitting under a “DeepSeek jumps the queue again” overhang. It doesn’t break the DeepSeek demand story: Flash is still printing new usage highs at 15T.

V4 Pro 确实 disappoint — 预期高是因为 Flash 很强，程序员社区对 Pro 的反馈不好。某种程度上，这对上市的中国模型名是 relief — MiniMax（0100.HK，376.80 港元 +5.4%）和智谱 / Z.ai（2513.HK，1,317 港元 +9.0%）——此前压着「DeepSeek 又要插队」的阴影。但不改 DeepSeek 的用量故事：Flash 还在创造用量新高（15T）。

NAND mix: server eSSD toward half of bits, YMTC #3 on shipments — Counterpoint (via @PolarisLog). Inference (KV cache / datasets) is pulling in high-capacity, low-power NAND. Server eSSD is on track for more than half of NAND bit shipments this year; consumer NAND prices are at record highs. Samsung Electronics (005930.KS, ₩268,000 +4.9%) is capacity-constrained and still prioritizing high-margin DRAM. SK hynix (000660.KS, ₩1,593,000 +5.9%) is riding Solidigm, bits +40% QoQ. YMTC ate the shortage (+22% YoY / +5% QoQ), 267L in production, 300L+ Xtacking in development. #3 by bits (Chosunbiz had ~14% share), but still #5 by revenue behind Micron (MU, $911.29 +4.9%) and Kioxia (285A.T, ¥51,800 +3.9%) — mix is still consumer-heavy. Kioxia was #3 last quarter with server mix above 30%, then lost the growth race as customers slowed on server SSD price hikes. YMTC wants more eSSD in 2H to cement #3. Through 2027, profitability is mix quality, not bits.

NAND 结构：服务器 eSSD 逼近出货一半，YMTC 升至全球第三 — Counterpoint（via @PolarisLog）。 推理（KV cache / 数据集）在拉高容量、低功耗 NAND；服务器 eSSD 有望在年内占 NAND bit 出货一半以上，消费级 NAND 价格也到了纪录高位。三星电子（005930.KS，268,000 韩元 +4.9%）产能紧、仍优先高毛利 DRAM。SK 海力士（000660.KS，1,593,000 韩元 +5.9%）靠 Solidigm，bit 出货环比 +40%。YMTC 吃到短缺（同比 +22% / 环比 +5%），267 层量产、Xtacking 300 层以上在研。出货量第三（Chosunbiz 份额约 14%），收入仍排第五，落后美光（MU，911.29 美元 +4.9%）与铠侠（285A.T，51,800 日元 +3.9%）——结构还偏消费。铠侠上季还是第三、服务器出货占比超 30%，随后因服务器 SSD 涨价客户放缓，增长被 YMTC 超过。YMTC 想在下半年加 eSSD 巩固第三；看到 2027，利润率看产品结构，不是出货量。

북극성

@PolarisLog

서버향 eSSD가 낸드 출하의 절반 가까이 차지, YMTC는 세계 3위로 올라섬
(Counterpoint Research)

- AI 산업의 중심이 학습에서 추론으로 옮겨가면서 KV 캐시와 데이터셋을 빠르고 저전력으로 저장할 낸드 수요가 크게 늘어남

- 초고용량 고성능 제품 중심으로 프리미엄화가 진행되면서 서버용 eSSD가 올해 안에 전체 낸드 비트 출하의 절반 이상을 차지할 전망임

…

Aug 12, 2026 · Open on X

Hon Hai (2317.TW, NT$262 −3.0%): Goldman sees the GPU/ASIC AI-server ramp after a 2Q OP beat. 2Q26 OP NT$94.8bn — 36% above Goldman, 18% above consensus. GM 6.1%, opex ratio 2.4%; NI NT$60bn (+35% YoY). Management is guiding high-double-digit QoQ growth in AI server rack shipments in 3Q26; Goldman models 3Q revenue at NT$3.04tn (+20% QoQ / +48% YoY). Company color: ~50% global AI server rack share, >40% in ASIC servers. 1H26 capex NT$80.9bn; FY26 capex up >30% YoY (racks / liquid cooling / testing), full-year OPM above last year’s 3.2%.

鸿海（2317.TW，262 新台币 −3.0%）：高盛看好 GPU/ASIC AI 服务器放量，二季度营业利润超预期。 2Q26 营业利润 948 亿新台币 — 较高盛高 36%、较共识高 18%。毛利率 6.1%、费用率 2.4%；净利润 600 亿（同比 +35%）。管理层指引 3Q26 AI 机架出货高双位数环比；高盛模型三季营收 3.04 万亿（环比 +20% / 同比 +48%）。公司口径：全球 AI 机架份额约 50%、ASIC 服务器 >40%。上半年资本开支 809 亿，全年资本开支同比增 >30%（机架 / 液冷 / 测试），全年营业利润率高于去年 3.2%。

Lenovo (0992.HK, HK$34.90 +20.2%) — FY1Q27: the tell is ISG / AI servers. Revenue $26.9bn (+43% YoY), ~20% ahead of Bloomberg. ISG nearly doubled to $8.5bn — ~39% above consensus, 9% above Morgan Stanley’s street-high. ISG OP $777m vs ~$242m est; OPM 9.1% (from 3.6% in Mar-Q, vs 4.1–4.5% expected). AI server backlog $5.4bn (+157% QoQ) on new cloud customers; Goldman’s pipeline number is $54bn, GB300 ramping at CSPs, Vera Rubin rack-scale in the queue. Dell (DELL, $484.50 +9.9%) and HPE (HPE, $58.79 +8.1%) bounced on the read-through. Adj. OP $1.9bn (+195% YoY) after warrant MTM. IDG/SSG also beat (IDG OPM 7.1%, AI PC share 25%), though PC units were still −2% YoY vs global −4%. What’s open is 2H PC (Intel/AMD already weaker; prior unit guide −15–20% y/y) and memory cost — not whether ISG printed.

联想（0992.HK，34.90 港元 +20.2%）— FY1Q27：看点在 ISG / AI 服务器。 营收 269 亿美元（同比 +43%），较 Bloomberg 高约 20%。ISG 近翻倍至 85 亿美元 — 超共识约 39%、高出摩根士丹利街边最高预期约 9%。ISG 营业利润 7.77 亿美元 vs 预期约 2.42 亿；营业利润率 9.1%（上季 3.6%，预期 4.1–4.5%）。AI 服务器 backlog 54 亿美元（环比 +157%），主因新云客户；高盛口径的 pipeline 是 540 亿美元，CSP 在推 GB300，Vera Rubin 机架在排队 — Dell（DELL，484.50 美元 +9.9%）与 HPE（HPE，58.79 美元 +8.1%）的反弹就是这条 read-through。剔权证市值波动后调整营业利润 19 亿美元（同比 +195%）。IDG/SSG 也超预期（IDG 营业利润率 7.1%，AI PC 份额 25%），PC 出货仍同比 −2%（全球 −4%）。接下来是下半年 PC（Intel/AMD 已偏弱；此前出货指引同比 −15–20%）和存储成本 — 不是 ISG 有没有打出来。

Memory long cycle — SK Securities (via @PolarisLog). AI compute is framed less as a fast-depreciating gadget and more as infrastructure that gets redeployed across generations and keeps throwing off cash. CoreWeave (CRWV, $107.73 +19.3%) is the proof point: six-year-old A100 capacity remains under long contracts into 2029, with used-GPU prices holding up—not just because of scarcity, but because customers are buying performance/$ for the workload. Frontier training gets the newest GPUs; H100/A100 move into general inference, fine-tunes, and batch. If economic life stretches past ~3 years (Amazon has said servers/networking often pay back inside three), the residual years are incremental profit rather than payback. That means new GPU sales and “second life” revenue can run together. It also argues against a simple ASP peak-out read on memory: SK remains overweight the complex, pointing to structural earnings power, longer visibility, and shareholder returns at Samsung Electronics (005930.KS, ₩268,000 +4.9%) and SK hynix (000660.KS, ₩1,593,000 +5.9%)—combined OP estimates of ₩641tn / ₩977tn for 2026/27.

存储长周期 — SK 证券（via @PolarisLog）。 AI 算力被写成可按世代再部署、反复产生现金流的基础设施，而不是很快贬值的消耗品。CoreWeave（CRWV，107.73 美元 +19.3%）是证据：上市约 6 年的 A100 资源仍签到 2029，二手 GPU 价格也站得住——不只是缺货，而是客户按 workload 买性能/成本。最新 GPU 做前沿训练/高性能推理，H100/A100 下沉到通用推理、微调与 batch。若经济寿命超过约 3 年（亚马逊称服务器/网络设备常在三年内回本），之后更多是增量收益而非回本期——新 GPU 销售与「第二生命」收入可以并行。这也反驳简单的存储 ASP peak-out：SK 维持超配，强调三星电子（005930.KS，268,000 韩元 +4.9%）与 SK 海力士（000660.KS，1,593,000 韩元 +5.9%）的结构性盈利与更长可见度，并关注股东回报；合计营业利润预期 2026/27 约 641 万亿 / 977 万亿韩元。

북극성

@PolarisLog

반도체, 긴 사이클
(SK증권 한동희)

- AI 컴퓨트는 빠르게 가치가 사라지는 소모품이 아니라 세대별로 재배치되며 반복적으로 현금흐름을 만드는 인프라 자산에 가까움

- CoreWeave 실적을 보면 출시된 지 6년 지난 A100 기반 자원도 2029년까지 장기 계약이 이어지고 있고 구형 GPU 가격도 의미 있는 수준을 유지함

…

Aug 13, 2026 · Open on X

Samsung / SK hynix — Daishin (via @DrNHJ): early returns, LTAs, and a 2027 cycle that still extends. Four things should drive the bounce: early shareholder returns, a re-rating of long-term agreements, HBM, and another demand check into 2027. Daishin sees annual returns of ₩150tn+ at Samsung Electronics (005930.KS, ₩268,000 +4.9%) and ₩80tn+ at SK hynix (000660.KS, ₩1,593,000 +5.9%)—max ₩200tn / ₩100tn—split into two tranches (an early tranche in Aug–Sep 2026, the rest in early-mid 2027). LTAs: rolling extensions look underpriced. Prepayments are already >20% of the contract (a year-plus on a 5-year deal), so cancellation risk is overstated and trough earnings should sit above the last cycle. Price bands don’t cap margins. Nvidia’s Rubin (incl. Ultra) HBM spec tweak is read as a shortage decision, not a smaller 2027 TAM—watch mix and profitability. September is the next demand check: server DRAM demand +50%+ in 2027, with shortage intensity wider than 2026. The basket also includes Micron (MU, $911.29 +4.9%), Sandisk (SNDK, $1,344.29 +5.8%), and Kioxia (285A.T, ¥51,800 +3.9%).

三星 / SK 海力士 — 代证（via @DrNHJ）：提前股东回报、长约重估、2027 周期仍在拉长。 四点支撑反弹：提前股东回报、长约（LTA）重估、HBM，以及 2027 需求再确认。代证预计三星电子（005930.KS，268,000 韩元 +4.9%）年度股东回报 150 万亿韩元以上、SK 海力士（000660.KS，1,593,000 韩元 +5.9%）80 万亿以上（上限约 200 万亿 / 100 万亿），分两笔：2026 年 8–9 月先发一笔，余下 2027 年初中。长约：滚动续约被低估；预付款已超合同 20%（五年约相当于一年以上），违约风险被说重了，底部盈利应高于上一轮。价格带也未必封死利润率上限。Nvidia Rubin（含 Ultra）HBM 规格调整被读成短缺下的决定，不是 2027 TAM 变小 — 看结构和利润率。9 月是下一轮需求确认：服务器 DRAM 2027 需求同比 +50% 以上，短缺强度大于 2026。篮子里还有美光（MU，911.29 美元 +4.9%）、Sandisk（SNDK，1,344.29 美元 +5.8%）与铠侠（285A.T，51,800 日元 +3.9%）。

루팡

@DrNHJ

$SKHY, SK hynix, Samsung, $MU, $SNDK, Kioxia

삼성전자 , 투자의견 Buy, 목표주가 560,000원

SK하이닉스, 투자의견 Buy, 목표주가 3,200,000원

…

Aug 13, 2026 · Open on X

Unitree — ITHome. Cumulative humanoid production has reached ~18k units.

宇树 — IT之家。 人形机器人累计产量约 1.8 万台。

Tech components / MLCC + FC-BGA — Metroseoul. AI servers are still pulling high-spec parts: MLCC plants running hot, FC-BGA demand broadening from AI servers into general-purpose servers and switch ICs. SEMCO (009150.KS, ₩1,503,000 +12.6%) and LG Innotek (011070.KS, ₩615,000 +1.8%) are adding capacity and locking LTAs.

科技零部件 / MLCC + FC-BGA — Metroseoul。 AI 服务器还在拉高规格元件：MLCC 厂开得满，FC-BGA 需求从 AI 服务器扩到通用服务器与 switch IC。SEMCO（009150.KS，1,503,000 韩元 +12.6%）与 LG Innotek（011070.KS，615,000 韩元 +1.8%）在扩产并锁长约。

SEMCO (009150.KS, ₩1,503,000 +12.6%) glass substrate — The Elec. Mass-production schedule is slipping. Even if equipment is ordered this year, manufacturing lead times mean line setup is unlikely before 2H27 — MP 2028 or later.

SEMCO（009150.KS，1,503,000 韩元 +12.6%）玻璃基板 — The Elec。 量产节奏在往后拖。即便今年下设备订单，产线搭建也难早于 27H2（制造交期）— 量产或到 2028 年及以后。