📰 DigiTimes 实时DigiTimes RealtimeDoubao · 实盘不做把握 7/10不做原因:无增量信息
北京时间·🕒 发布 08:55:00时区推断·⟳ 抓取 09:04:11·⚡ 决策 10-14 09:06:26 +2m15s·13.3 天前
鸿海OCP 2026展示Rubin机柜级AI基础设施与FIT液冷/互连方案,仅为展品演示、无订单等增量商业事实
报道为OCP日程内展会的展品巡礼,通篇能力/参数展示,无新订单、客户定点、backlog、出货或指引等需求验证事实;参数与demo只证供给能力,相对鸿海/FIT/FII体量无materiality,且无大额订单触发二阶液冷外溢映射,缺乏正期望与可执行催化。
📄 原文DigiTimes Realtime↗ 原文
Foxconn Technology Group showcased rack-scale artificial intelligence infrastructure at the Open Compute Project (OCP)…
不做 鸿海精密(2317.TW)
OCP展会能力演示,无订单/客户/指引等增量事实,相对市值无短线edge,不做。
不做 鸿腾精密(FIT)(06088.HK)
液冷/互连仅为展品组合介绍,无量产定点或收入量级,缺乏可交易催化。
不做 工业富联(FII)(601138.SH)
Ingrasys Rubin平台属既有路线展示,无新增商业里程碑;不据展会demo追主题。
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
结论摘要
鸿海在OCP 2026展示Rubin整柜与FIT液冷/互连方案,但纯属日程内展会的产品组合演示,无任何新增订单或商业里程碑,不构成需求验证,相对主体市值无可交易edge,故不做。
🔎 新鲜度:最早出现 2026-10-14 · 扩散 low · price-in 风险 low ↗
主要风险
- 若后续出现绑定客户的Rubin机柜/液冷正式订单或指引上调,需按订单量级重新评估,本条不含该信息。
反向证据
- 做多前查反向 · 来源 web · 日期 2026-10-14 · 影响 invalidates
正文全文仅为展品/参数展示,无任何订单、客户定点、backlog或指引;内部72小时查重亦无同事项决策,未发现支撑做多的新增需求事实。
检索:Foxconn FIT Rubin NVL72 liquid cooling order customer win OCP 2026 原文 ↗
先例如何影响了这次决策
对本次决策的影响:确认无需建仓;不为纯展品展示检索/编造价格先例
方向:无可定价的商业催化,未进入需要先例校准涨幅的阶段
涨幅量级:无ENTER标的,无需以先例定tp/sl
形态:展会demo类事件通常不构成独立短线催化,故不据此交易
工具执行结果 · 成功 1 · 降级 0 · 失败 0 · 取消 0
- search_processed_events 成功 stock_tools HTTP · 10 ms
{"result": "search_processed_events:agent query=`fit foxconn interconnect liquid cooling busbar 480kw osfp 12.8t xpo ocp | foxconn ocp global summit 2026 vera rubin nvl72 ingrasys rack | 鸿海 ocp 2026 rubin nvl72 机柜 液冷 fit`;结构化索引覆盖完整 72 小时;旧记录兼容扫描近 0.0/72 小时、0 条本臂决策(完整),返回 0 条候选。未命中只表示这些 agent query 表
原文全文
展开原文 · 2,637 字
Foxconn Technology Group showcased rack-scale artificial intelligence infrastructure at the Open Compute Project (OCP) Global Summit 2026 in San Jose, California, combining computing platforms, high-speed interconnects, power delivery, and liquid cooling. The portfolio included Ingrasys Technology's Nvidia Vera Rubin NVL72 platform and connectivity, power, and thermal-management systems from Foxconn Interconnect Technology (FIT) as Foxconn expanded its offering from components and Compute Trays to complete AI racks.
Ingrasys featured the Vera Rubin NVL72 platform for large-scale model training and AI inference. The platform's inference performance per watt reaches up to 10 times that of Nvidia's Blackwell platform, according to the company. Ingrasys also presented the Nvidia Groq 3 LPX Rack and Compute Tray for agentic AI inference, along with a Vera CPU Rack Compute Tray using high-density liquid cooling. The design can accommodate up to 256 Vera CPUs in one rack and support 22,500 sandboxes for agentic AI, reinforcement learning, and data-processing workloads.
The exhibition also included the HGX Rubin NVL8 platform and an Nvidia MGX 4U architecture equipped with Nvidia RTX PRO 6000 Blackwell Server Edition GPUs. Foxconn presented additional MGX support spanning server chassis, racks, power delivery, and liquid cooling. Its ultra-dense interconnect (UDI) and ZQD technologies were featured on the Nvidia MGX Ecosystem Wall as solutions for high-density AI rack architectures.
FIT added connectivity and infrastructure technologies to the portfolio. Its 1.6T octal small form factor pluggable (OSFP) platform supports direct-attach copper, active copper, active retimer, and active electrical cables, as well as a 32-port OSFP liquid-cooling solution. Its 12.8T XPO, or eXtra Dense Pluggable Optics, platform combines electrical and optical interconnects with direct liquid cooling to address bandwidth, power, and thermal requirements while increasing port density and energy efficiency. FIT's portfolio supports PAM4, PCIe 6.0, and PCIe 7.0, along with 224G and 448G SerDes.
For rack power and cooling, FIT displayed a liquid-cooled busbar system rated for up to 480 kilowatts, high-current alternating-current whips supporting up to 400 amps, and OCP-compliant UQD and UQDB liquid-cooling quick disconnects. Foxconn said it was continuing work with Intel on rack-scale agentic AI systems based on Intel Xeon platforms. The group plans to combine Ingrasys computing platforms with FIT's components as AI infrastructure moves toward higher density, increased power consumption, and broader use of liquid cooling.
Ingrasys featured the Vera Rubin NVL72 platform for large-scale model training and AI inference. The platform's inference performance per watt reaches up to 10 times that of Nvidia's Blackwell platform, according to the company. Ingrasys also presented the Nvidia Groq 3 LPX Rack and Compute Tray for agentic AI inference, along with a Vera CPU Rack Compute Tray using high-density liquid cooling. The design can accommodate up to 256 Vera CPUs in one rack and support 22,500 sandboxes for agentic AI, reinforcement learning, and data-processing workloads.
The exhibition also included the HGX Rubin NVL8 platform and an Nvidia MGX 4U architecture equipped with Nvidia RTX PRO 6000 Blackwell Server Edition GPUs. Foxconn presented additional MGX support spanning server chassis, racks, power delivery, and liquid cooling. Its ultra-dense interconnect (UDI) and ZQD technologies were featured on the Nvidia MGX Ecosystem Wall as solutions for high-density AI rack architectures.
FIT added connectivity and infrastructure technologies to the portfolio. Its 1.6T octal small form factor pluggable (OSFP) platform supports direct-attach copper, active copper, active retimer, and active electrical cables, as well as a 32-port OSFP liquid-cooling solution. Its 12.8T XPO, or eXtra Dense Pluggable Optics, platform combines electrical and optical interconnects with direct liquid cooling to address bandwidth, power, and thermal requirements while increasing port density and energy efficiency. FIT's portfolio supports PAM4, PCIe 6.0, and PCIe 7.0, along with 224G and 448G SerDes.
For rack power and cooling, FIT displayed a liquid-cooled busbar system rated for up to 480 kilowatts, high-current alternating-current whips supporting up to 400 amps, and OCP-compliant UQD and UQDB liquid-cooling quick disconnects. Foxconn said it was continuing work with Intel on rack-scale agentic AI systems based on Intel Xeon platforms. The group plans to combine Ingrasys computing platforms with FIT's components as AI infrastructure moves toward higher density, increased power consumption, and broader use of liquid cooling.
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)。
98cf3fe7a0· 2026-10-14 09:06 · doubao / doubao-seed-evolving-latest-version · 策略包2226518bfdee当前