🐦 推特Deedy DasCodex · 实盘不做把握 8/10
北京时间·🕒 发布 11:51:15时区推断·⟳ 抓取 11:51:35·⚡ 决策 07-28 11:53:16 +1m41s·91.2 天前
Deedy Das转述中国国产浸没式DUV光刻机量产并称将供给中芯、华虹和长鑫
推文内容与系统近72小时内已处理的“中国国产浸没式DUV光刻机量产报道”属于同一具体催化，重复来源不再出卡。
📄 原文Deedy Das@deedydas↗ 原文
China today made its biggest move deep in the AI stack: it began producing its own chip-making machine and homegrown AS…
不做 芯源微（688037.SH）
同一国产浸没式DUV量产催化已在11:26被系统处理，本条为社交转述且无新增事实，不重复出卡。
不做 张江高科（600895.SH）
命中近72小时同一具体催化去重规则，且本条未新增一手确认、订单金额或交付节奏。
不做 ASML（ASML.US）
方向上偏负但不是新增公司级实质利空，且已由此前同催化覆盖，不重复开空或登记风险。
展开详情(成交明细 / 研究档案 / 原文全文 / 决策记录链)
研究档案
关键理由
- 推文内容与系统近72小时内已处理的“中国国产浸没式DUV光刻机量产报道”属于同一具体催化，重复来源不再出卡。
- 该事件方向上利好国产光刻机及零部件主题、利空ASML对华DUV服务预期，但本条没有新增交付数量、客户确认或供应链名单变化。
- 内部KB显示7月28日已有券商围绕The Information报道映射永新光学、炬光科技等国产光刻机部件标的，当前推文更多是二次传播。
主要风险
- 若后续出现宇量昇、SiCarrier、SMIC、华虹或长鑫的一手确认并披露真实订单/交付数量，可作为新增事实重新研判。
- 社交转述容易放大“量产”含义，实际仍有12-18个月调机、良率和日本关键零部件依赖等兑现风险。
工具调用 · 历史记录,结果未知(3 种)
- search_processed_events · 结果未知
- kb_search · 结果未知
- query_modeling_kb · 结果未知
原文全文
展开原文 · 1,733 字
China today made its biggest move deep in the AI stack: it began producing its own chip-making machine and homegrown ASML.
Yuliangsheng, tied to Huawei and SiCarrier, is now mass-producing immersion DUV lithography machines, the tools that make 7nm chips. their first deliveries are this year, to SMIC, the makers of Huawei’s Ascend chips, Hua Hong and CXMT, memory manufacturers.
The China AI compute supply chain can be thought of as ASML : Yuliangsheng -> TSMC : SMIC -> Nvidia : Huawei.
SMIC makes 7nm node chips using immersion DUV systems on ASML tools bought before the various sanctions. If the current MATCH Act passes, they can’t even be serviced. China desperately needs Yuliangsheng to ramp capacity to increase the domestic compute production, even on yesteryears nodes.
The current machines are equivalents of ASML NXT:1980Di, a 28nm tool that reaches 7nm with multi-patterning, which is 10-15yrs behind frontier. For comparison, the new Vera Rubins are on an EUV 3nm node. And China’s still ramping capacity, projecting 20 machines in 2027 vs ~130 DUV/yr and ~45 EUV/yr from ASML.
This also has tons of caveats. They rely on critical parts from Japan. New scanners need 12-18mos of tuning per fab. Yields take a long long time to get right. But China tends to be quite good at manufacturing at scale.
Even though they’re far behind on frontier tech, this is a significant milestone to China owning its homegrown AI supply chain. Maybe not in 5yrs, but maybe 10, they’re going to mass produce compute that could be quite threatening. With largely better manufacturing abilities and fewer regulatory hurdles to deploy data centers at scale with ample power, this poses a real risk to the balance of global compute.
Yuliangsheng, tied to Huawei and SiCarrier, is now mass-producing immersion DUV lithography machines, the tools that make 7nm chips. their first deliveries are this year, to SMIC, the makers of Huawei’s Ascend chips, Hua Hong and CXMT, memory manufacturers.
The China AI compute supply chain can be thought of as ASML : Yuliangsheng -> TSMC : SMIC -> Nvidia : Huawei.
SMIC makes 7nm node chips using immersion DUV systems on ASML tools bought before the various sanctions. If the current MATCH Act passes, they can’t even be serviced. China desperately needs Yuliangsheng to ramp capacity to increase the domestic compute production, even on yesteryears nodes.
The current machines are equivalents of ASML NXT:1980Di, a 28nm tool that reaches 7nm with multi-patterning, which is 10-15yrs behind frontier. For comparison, the new Vera Rubins are on an EUV 3nm node. And China’s still ramping capacity, projecting 20 machines in 2027 vs ~130 DUV/yr and ~45 EUV/yr from ASML.
This also has tons of caveats. They rely on critical parts from Japan. New scanners need 12-18mos of tuning per fab. Yields take a long long time to get right. But China tends to be quite good at manufacturing at scale.
Even though they’re far behind on frontier tech, this is a significant milestone to China owning its homegrown AI supply chain. Maybe not in 5yrs, but maybe 10, they’re going to mass produce compute that could be quite threatening. With largely better manufacturing abilities and fewer regulatory hurdles to deploy data centers at scale with ample power, this poses a real risk to the balance of global compute.
决策记录链 · 只追加
每次领取产生一条记录,重新决策追加一条并以 supersedes 指向被替代的记录(ADR 0014)。
这条决策早于决策账本(2026-10-08 起逐条追加),没有追加记录。