🐦 推特Tae KimQwen · 实盘不做把握 3/10不做原因:标的无直接敞口
北京时间·🕒 发布 02:48:24时区推断·⟳ 抓取 02:48:44·⚡ 决策 09-04 02:51:56 +3m12s·53.6 天前
OpenAI发布GPT-6 Astra,声称达到AGI水平,使用10万+GPU集群训练
GPT-6发布是非上市公司产品发布,训练用10万+GPU集群已建成(后视事实),不构成新增订单或收入催化。NVDA作为$5.5万亿市值超大市值标的属软催化(无新金额/指引变化);A股供应链标的与本次发布无新增需求因果关系(硬件已采购部署完毕),且主要候选处于下行趋势或已大幅上涨。推理侧扩容是渐进预期,非突发增量。
📄 原文Tae Kim@firstadopter↗ 原文
GPT-6 is here! “For me personally, I do think we’re there [GPT-6 is AGI]” – Greg Brockman OpenAI: “The world’s best com…
不做 NVIDIA(NVDA.US)
$5.54万亿超大市值,GPT-6训练集群已建成(后视),无新增订单/收入/指引变化,属软催化;今日已涨+2.4%实时重定价中,无剩余edge
不做 中际旭创(300308.SZ)
训练集群已建成无新增光模块需求;股价处于明确下行趋势(20日-14.9%,13周轨迹100→71),做多逆风
不做 新易盛(300502.SZ)
同上,训练已完成无新增需求,股价下行趋势(20日-8.8%,13周轨迹100→70)
不做 英维克(002837.SZ)
液冷20日已涨+29.4%,今日再涨+3.4%;训练集群冷却已部署,GPT-6发布不构成新增液冷订单催化
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
OpenAI发布GPT-6 Astra,核心卖点为10万+GPU训练和AGI叙事。但训练集群已建成,硬件采购为过去完成时,不构成NVDA或供应链的新增订单。NVDA今日已涨2.4%(5日+7%),$5.5万亿市值下无金额增量的产品发布属软催化。A股光模块(中际旭创、新易盛)处于明确下行趋势,液冷(英维克)已涨29%,均无新增需求催化支撑做多。事件本质是叙事/情绪催化而非需求冲击,无正期望交易机会。
关键理由
- GPT-6确认AI训练规模持续扩大,但训练集群已建成,硬件需求为既成事实非新增催化
- 推理侧扩容(全用户+AWS)是渐进过程,已被市场定价
- AGI叙事可能驱动短期情绪,但不改变任何公司的收入/订单/指引
主要风险
- 若市场将AGI叙事解读为新一轮capex周期起点,AI板块可能出现超预期情绪反弹
- 推理规模若远超预期(如computer use驱动企业大规模部署),可能产生增量需求
反向证据
- 做多前查反向 · 来源 query_modeling_kb · 日期 2026-09-03 · 影响 invalidates
modeling_kb显示OpenAI-NVIDIA 10GW合作于2025年9月已公布,训练集群为既有部署;无新增订单或收入指引变化
检索:GPT-6发布对NVDA/A股算力链是否有新增订单或收入影响 - 做多前查反向 · 来源 price_snapshot · 日期 2026-09-03 · 影响 invalidates
中际旭创20日-14.9%、新易盛-8.8%均处下行趋势,英维克+29.4%已充分上涨;无证据表明GPT-6发布构成这些标的的新增需求催化
检索:A股光模块/液冷近期走势与GPT-6因果关系
先例如何影响了这次决策
对本次决策的影响:先例缺失不改变结论:本事件的核心问题是因果敞口而非幅度校准,训练集群已建成使供应链无新增需求
方向:无高相似先例可引用
涨幅量级:无法用先例校准幅度
形态:无
工具执行结果 · 成功 7 · 降级 0 · 失败 0 · 取消 0
- search_processed_events 成功 stock_tools HTTP · 385 ms
{"result": "search_processed_events:agent query=`gpt-6 100000 gpu cluster训练 | gpt-6 astra openai发布 | openai gpt-6 agi模型`;结构化索引覆盖完整 72 小时;旧记录兼容扫描近 72.0/72 小时、7594 条本臂决策(完整),返回 0 条候选。未命中只表示这些 agent query 表达式没有召回候选,不能据此证明消息首发或不是重复;按当前事件继续正常研判。"} - fetch_decision_experience 成功 stock_tools HTTP · 21 ms
{ - kb_search 成功 stock_tools HTTP · 1,847 ms
{ - query_modeling_kb 成功 stock_tools HTTP · 49 ms
{"result": "modeling_kb 命中 12 条 (库快照 2026-07-27 · AI 算力链 19 页 · 本次命中证据最新 2026-04-30 · 其中 ⟨预测⟩ 3 条) —— 这是**定期重生成的快照**,不是活数据;比快照更晚的事实以你手上的信号/kb_search 为准。\n\n• [A] OpenAI · 表 · 主要客户英伟达 GPU/机柜部署规模(按日历年) · CY2026E = ≥10 GW / 数百万 GPU(签约) GPU 颗 / GW / $ ⟨预测⟩ ⚠️源间数字有冲突[1][2][3][4][5][6][7][8][9][10][11]\n - stock_quote 成功 stock_tools HTTP · 252 ms
{ - price_snapshot 成功 stock_tools HTTP · 295 ms
{ - search_case_library 成功 stock_tools HTTP · 475 ms
{"result": "search_case_library「NVDA.US」无历史决策命中(我们过去没在这些标的上决策过,或历史尚未回填)。\n◆ NVDA.US:相关历史决策 3231 条 · 其中该标的 ENTER 32 条(1.0%) · 已结算 32 笔 · 胜率 50% · 中位 -0.12% · 均值 +0.14%\n↑ 胜率/收益口径:ENTER 当日之后 T+1 开盘入场、次日开盘出场,做空腿已按方向取反(所以正数=这个方向做对了)。统计的只有★那只票自己的腿。"}
原文全文
展开原文 · 1,231 字
GPT-6 is here!
“For me personally, I do think we’re there [GPT-6 is AGI]” – Greg Brockman
OpenAI: “The world’s best computer use model”
“Astra also saturates ARC-AGI-3 with a 99.9% score and ExploitBench with a 100% score.”
First OpenAI model to be trained on 100,000+ GPU cluster.
“GPT‑6 Astra is rolling out today to a limited set of organizations and over the coming days will become available to all ChatGPT Plus, Pro, Business, and Enterprise users, as well as through the OpenAI API and AWS.”
“We put Astra on real work across Epic and it managed C++ at the scale of Unreal Engine, worked out complex problems faster than anything we've tested, and showed real nuance to support non-engineering teams work multimodally. We're seeing early signs of a model that unlocks more creative exploration with less handholding.” - Daniel Vogel, Chief Operating Officer, Epic Games
“Astra can handle much broader, much more ambitious work than its predecessors. I can ask it to figure out how to improve a metric without first breaking that goal into tasks, and it’ll quickly find a way forward, so I end up spending less time telling it what to do and more time working through ideas.” - Sarah Sachs, AI Engineering Lead, Notion
“For me personally, I do think we’re there [GPT-6 is AGI]” – Greg Brockman
OpenAI: “The world’s best computer use model”
“Astra also saturates ARC-AGI-3 with a 99.9% score and ExploitBench with a 100% score.”
First OpenAI model to be trained on 100,000+ GPU cluster.
“GPT‑6 Astra is rolling out today to a limited set of organizations and over the coming days will become available to all ChatGPT Plus, Pro, Business, and Enterprise users, as well as through the OpenAI API and AWS.”
“We put Astra on real work across Epic and it managed C++ at the scale of Unreal Engine, worked out complex problems faster than anything we've tested, and showed real nuance to support non-engineering teams work multimodally. We're seeing early signs of a model that unlocks more creative exploration with less handholding.” - Daniel Vogel, Chief Operating Officer, Epic Games
“Astra can handle much broader, much more ambitious work than its predecessors. I can ask it to figure out how to improve a metric without first breaking that goal into tasks, and it’ll quickly find a way forward, so I end up spending less time telling it what to do and more time working through ideas.” - Sarah Sachs, AI Engineering Lead, Notion
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。