AI labs

141 atoms · 跨 57 天 · 首见 2026-04-15 · 最近 2026-07-03

三色: 🟦 fact 25 · 🟥 take 116 · stance ▲36/▼48/◆39

叙事状态: 🍂 fading (退潮) · 💤 沉寂 · 策展近14d 0 atom · 🐦 X 近14d 11

状态只算低频策展源 (群/卖方); X firehose 仅作背景音量

来源: X 141

时态: fresh:108 · aging:30 · stale:2 · expired:1

标签: 好观点:107 · 好思考:28 · 好信源:12 · 好数字:9 · 好问题:7

别名 (合并): AI Labs

🟨 AI 综合 · junior analyst 概览

🧭 拥挤度 (一人一票): ▲ 24 位作者 (KOL24) vs ▼ 30 位作者 (KOL30)

⚖️ 空头 30/30 来自KOL

📄 广泛报道的事实 · 1 个

💢 核心分歧 (4 轴)

1. 监管与开放: 强制披露安全 vs 创新受阻

*topic: 技术路线/监管政策 · 11 bull vs 6 bear*

🟢 bullish 侧:

• <a id="atom-52e656c509fba12d"></a>2026-04-23 X·@scaling01 ⭐ [老化中] · prompting到AGI路径 · 技术路线
  AI labs are literally prompting their way to AGI
  tags: 好观点·好思考 · → daily
• <a id="atom-7c9182601864952a"></a>2026-04-29 X·@dwarkesh_sp · 模型训练 over-trained 程度 · 技术路线
  Because of RL, models may be 100x over-trained beyond Chinchilla-optimal
  tags: 好观点 · → daily
  value: qty=100x · date=RL 场景
• <a id="atom-e39367eb27013ec7"></a>2026-05-03 X·@teortaxesTex [老旧] · 2026年1月已拥有端到端训练的前沿模型和工具 · 技术路线/模型能力
  I'd suspect that in January 2026 there already were frontier models and harnesses that allowed to develop this end to end
  tags: 好观点 · → daily
  📷 原图
• <a id="atom-511c71762c8dd0f8"></a>2026-05-08 X·@krinza · 未来3年建立生物部门与湿实验室 · 技术路线/生物医药
  in the next 3 years, every major AI lab will spin up its own bio arm and in-house wet labs.
  tags: 好观点 · → daily
  📷 原图
• <a id="atom-018d10bfaaf42d9f"></a>2026-05-15 X·@zephyr_z9 · 策略:提升token效率以服务高batch size · 技术路线/数据中心算力
  Also, remember that labs are pushing higher token efficiency, so that they don't need to push TPS super hard and can serve at super high batch sizes
  tags: 好思考 · → daily
  📷 原图

🔴 bearish 侧:

• <a id="atom-02ed1f97eedc1c67"></a>2026-05-17 X·@Presidentlin · prediction about software engineer replacement by AI was horribly wrong · 技术路线 +同日1条
  It's clear now that prediction was horribly wrong, contrary to what the Claudesceps will have you think.
  tags: 好观点 · → daily
• <a id="atom-caff59a6bbaf4d6d"></a>2026-05-30 X·@willccbb · 蒸馏规则合规性 · 技术路线/监管政策
  it's often 'against the rules', sure
  tags: 好观点 · → daily
• <a id="atom-41ff8ade374332fd"></a>2026-06-25 X·@AndrewCurran_ · internal capability gap widening · 技术路线/库存
  The gap between what is available to the public, and what the labs have internally, will steadily widen from this day forward.
  tags: 好观点·好思考 · → daily
• <a id="atom-efc79b972259f15a"></a>2026-05-20 X·@theinformation · 推动缩短审查窗口 · 监管政策
  AI labs are pushing for a shorter review window than the proposed 90 days.
  tags: 好信源 · → daily
  value: qty=短于90天 · direction=缩短
  📷 原图
• <a id="atom-425c5ae9794818e0"></a>2026-06-09 X·@deanwball · 可能被政府秘密强制降低性能 · 监管政策
  the government, with a classified regulatory system for AI (as the recent cyber EO tees up) could secretly force AI labs to silently degrade performance in a wide range of domains without *that* much risk of First Amendment litigation
  tags: 好观点 · → daily

2. 定价权博弈: 锁定高利润率 vs 被迫降价让利

*topic: 竞争格局/定价权 · 9 bull vs 13 bear*

🟢 bullish 侧:

• <a id="atom-fdf7f4c6ec69a90b"></a>2026-05-06 X·@dwarkesh_sp [老化中] · 可以对其最佳模型收取更高利润率 · 竞争格局 +同日2条
  AI labs will be able to charge *higher* margins on their best models as compute gets scarcer.
  tags: 好观点·好思考 · → daily
• <a id="atom-12e152845c32ed71"></a>2026-05-10 X·@robjama · 内部模型领先优势 · 技术路线/竞争格局
  People at major AI labs (using internal models) 3-4 months ahead of startup silicon valley engineers
  tags: 好信源 · → daily
  value: qty=3-4 months · date=当前
• <a id="atom-04a285ccd45d649e"></a>2026-06-18 X·@Mascobot · 当前竞争优势来源 · 竞争格局
  The AI labs wining right now are not the ones with the most compute, but the ones with the best data
  tags: 好观点 · → daily
• <a id="atom-6a5c1b8772c3184b"></a>2026-06-21 X·@GabGarrett · 提供成本优势 · 定价权/竞争格局
  the labs are providing the giveaway of a lifetime when you look at the costs of running a commensurate model yourself
  tags: 好观点 · → daily
  value: direction=正面
• <a id="atom-e7be191f5d5383f3"></a>2026-05-06 X·@samsja19 [老化中] · 定价能力 · 定价权
  AI labs will be able to charge *higher* margins on their best models as compute gets scarcer
  tags: 好思考·好观点 · → daily
  value: direction=higher · qty=higher margins on best models

🔴 bearish 侧:

• <a id="atom-09b59ff2b3065aeb"></a>2026-04-17 X·@tbpn [老化中] · 威胁创业者业务 · 竞争格局
  many of their founders are now wondering if they're 'just an iteration away' from AI labs destroying their business
  tags: 好观点 · → daily
• <a id="atom-19c4fc8e80d9639e"></a>2026-05-02 X·@wordgrammer [老化中] · 自身定位转变 · 竞争格局/技术路线
  The ai labs all referred to themselves as 'platform companies' a year or so ago, when everyone was building GPT wrappers, and now that the paradigm has shifted to MCP plugins/roonskate they abandoned it. This is backwards
  tags: 好观点 · → daily
  value: direction=从平台公司转向不再自称平台公司
• <a id="atom-69e9e1916d7f2bfb"></a>2026-05-07 X·@emollick [老化中] · building deployment consultancies · 竞争格局
  The fact that the Labs are building their own deployment consultancies (which will take a long time) suggests a failure of imagination or a lack of trust that models will be up to that task in coming years.
  tags: 好观点 · → daily
• <a id="atom-b000fc3f647669e7"></a>2026-05-24 X·@natolambert · 历史上不公开且不真正从事科学 · 竞争格局
  Tradition doesn't show that these labs are particularly open and actually doing science these days
  tags: 好观点 · → daily
• <a id="atom-0476db6dd11b124d"></a>2026-05-27 X·@matt_slotnick · absorb top talent · 竞争格局/人才流动 +同日1条
  biggest risk to incumbents may be that labs absorb all the top talent. labs already get great eng talent. but this will spread to product… sales… marketing… etc. when they become the optimal destination to solve the problems of your customers, it’s going to get ugly
  tags: 好观点 · → daily

3. 资本开支军备竞赛: 烧钱求增长 vs 回报率存疑

*topic: 资本开支/盈利能力 · 2 bull vs 6 bear*

🟢 bullish 侧:

• <a id="atom-df0401a44d2975f4"></a>2026-06-13 X·@dalibali2 · 预计资本支出和站点启用不受影响 · 资本开支
  doesn’t change the planned capex and 2027/2028 site go-lives because no lab is going to believe the trajectory is anything but up and up
  tags: 好观点 · → daily
  value: date=2027/2028
• <a id="atom-b79e92f3ac684b3b"></a>2026-06-30 X·@henrytdowling · app stack investment focus · 应用层/资本开支
  labs are investing a lot into claude code / codex rn
  tags: 好信源 · → daily

🔴 bearish 侧:

• <a id="atom-ce4dfa161c0b939f"></a>2026-06-09 X·@suchenzang · might spend capex with no intention of democratizing technology · 资本开支
  with the amount of capex spent so far, this was never meant to be some democratizing technology 'for the people'.
  tags: 好观点 · → daily
• <a id="atom-d49f0aeae0e9aedb"></a>2026-06-13 X·@jasonschips · demand to train future models · 订单需求/资本开支
  Less demand to train future models
  tags: 好观点 · → daily
  value: direction=lower
• <a id="atom-6a26e087c06a752f"></a>2026-06-22 X·@stevehou · AI-related news could include slowdown of capex due to capital planning or local pushback · 资本开支/竞争格局
  I imagine it'll likely have to do with AI-related news such as a slowdown of capex due to either capital planning or local pushback and the threat of open models as economics of the labs continue to become public knowledge
  tags: 好思考 · → daily
• <a id="atom-5102858e8b7542ad"></a>2026-06-25 X·@iamgingertrash · no money for next 10x training run · 融资/资本开支 +同日1条
  No $$ for next 10x training run
  tags: 好观点 · → daily
  value: qty=no $ · date=next 10x training run
  📷 原图
• <a id="atom-76ac2c6e4741dda1"></a>2026-06-26 X·@WolfeFolks · 需要暂停以不流血出局 · 资本开支
  the pause they need to not bleed out
  tags: 好观点 · → daily

4. 营收高速增长 vs 盈利化难题: ARR激增何时转正

*topic: 营收/盈利能力 · 2 bull vs 5 bear*

🟢 bullish 侧:

• <a id="atom-4d7204abb811fea6"></a>2026-05-01 X·@scaling01 · unit economics are increasingly favorable · 盈利能力
  increasingly favorable unit economics
  tags: 好观点 · → daily
  📷 原图
• <a id="atom-7d1028739e0e8900"></a>2026-06-12 X·@vikramskr · Net New ARR projection · 营收/订单需求
  AI Labs are projected to generate the most net new ARR in Q2
  tags: 好观点 · → daily
  value: date=Q2

🔴 bearish 侧:

• <a id="atom-7d22bbe635365c53"></a>2026-04-24 X·@based16z · profitability status · 盈利能力/营收
  Meanwhile the AI labs make about the same “annualized” revenues with negative profits
  tags: 好数字 · → daily
• <a id="atom-07d9e8f6ce402beb"></a>2026-05-13 X·@Crussian17 · revenue growth will slow hard due to token optimization · 营收/盈利能力
  when that happens rev growth for labs will slow hard
  tags: 好观点·好思考 · → daily
• <a id="atom-c6de8b474badfcb7"></a>2026-06-25 X·@iamgingertrash · revenue growth stall · 营收 +同日1条
  Revenue growth stall
  tags: 好观点 · → daily
  📷 原图
• <a id="atom-7610a527a37e32d3"></a>2026-07-01 X·@mweinbach · 意图隐藏利润率和吞吐量 · 盈利能力
  Labs want to sell you intelligence at a certain cost, the less you know about their margin and throughput the better, because there's less room to negotiate
  tags: 好观点 · → daily

🔗 因果传导 (causal map)

*AI labs 在产业链上的传导关系. 边是群里/卖方陈述的因果 (非 AI 推断), 数字 = 几条 atom 支撑. 点 atom 溯源.*

↓ 下游·近期陈述 (1)

• 🟢 deployed compute · 需求拉动 · 1 次陈述
  As long as AI labs continue to subsidize their plans in order to show user growth and revenue (while increasing their lo

↑ 上游·近期陈述 (2)

• 🔴 US government · 政策影响 · 2 次陈述
  US government may require AI labs to get an approval before model releases.
• 🔴 price controls on AI model APIs · 政策影响 · 1 次陈述
  price controls on AI model APIs would trigger an Atlas Shrugged exit from the US

📏 估计带 (1)

*语料内对同一量的估计区间 (≥2 个估计才成带). 不判断谁对, 只陈列.*

• 营收 未标期限 ⚠混合口径: $90B – $200B · 2 个估计 · 最新 2026-06-25 $200B (X·cred3)

🟦 客观事实 (facts) (25)

• <a id="atom-e86a6c6ab9404a89"></a>🟦 2026-05-31 X·@zephyr_z9 ⭐ · pretraining token usage · 算力
  Labs have never used 300T or more tokens for pretraining
  tags: 好数字·好信源 · → daily
  value: qty=<300T · date=up to 2026-05-31
• <a id="atom-a1fcf0a5634c4185"></a>🟦 2026-04-15 X·@scaling01 · coding-related risk improvement · 技术路线
  we do know that AI is predictably improving in the exact fields that we train them on (mostly coding, math)
  tags: 好信源·好观点 · → daily
• <a id="atom-8cd4211dde137a4e"></a>🟦 2026-05-01 X·@xeophon · investing in biosecurity · 资本开支
  both are investing hundreds of millions into this rn
  tags: 好数字·好观点 · → daily
  value: qty=hundreds of millions
• <a id="atom-7292bb75827e2909"></a>🟦 2026-05-01 X·@scaling01 · end demand is strong · 订单需求
  driven by strong end demand
  tags: 好数字 · → daily
  📷 原图
• <a id="atom-41ff8ade374332fd"></a>🟦 2026-06-25 X·@AndrewCurran_ · internal capability gap widening · 技术路线/库存
  The gap between what is available to the public, and what the labs have internally, will steadily widen from this day forward.
  tags: 好观点·好思考 · → daily
• <a id="atom-d432ad36c57083b9"></a>🟦 2026-06-15 X·@Techmeme · 提案:政府持股 · 监管政策
  government stakes in AI labs
  tags: 好信源 · → daily
• <a id="atom-429a695a57adb030"></a>🟦 2026-06-02 X·@FT · expand research into machine consciousness · 技术路线/地缘政治
  Top AI labs expand research into machine ‘consciousness’
  tags: 好信源 · → daily
• <a id="atom-bf001a22ef893f48"></a>🟦 2026-04-23 X·@coatuemgmt · 消费模式定价占比 · 定价模式
  AI labs: 74% consumption-based pricing.
  tags: 好数字 · → daily
  value: qty=74% · date=2026-04
  [图: 展示传统 SaaS 与 AI Labs 定价模式对比及传统 SaaS 用量计费支出占比趋势的分析图表 — AI Labs 消费模式占比: 74%; 传统 SaaS 席位/平台模式占比: 96%; 传统 SaaS 消费模式占比: 4%; AI Labs 席位/平台模式占比: 26%; 2026年2月传统 SaaS 消费支出占比: 3.8%]
  📷 原图
• <a id="atom-97862aba2d7626a5"></a>🟦 2026-04-16 X·@Techmeme · 购买 defunct startup 的 Slack/Jira/email 档案 · 合作客户
  AI labs are buying Slack, Jira, and email archives from defunct startups to build 'reinforcement learning gyms' and train AI agents in simulated workplaces
  tags: 好信源 · → daily
• <a id="atom-36aabd508e7de93a"></a>🟦 2026-06-21 X·@willccbb · spends billions on evals · 资本开支
  the labs have long understood that evals are half the battle, and they quietly spend billions on them
  tags: 好数字 · → daily
• <a id="atom-698e5da7ef89a88b"></a>🟦 2026-06-12 X·@coatuemgmt [已过期] · 季度净新增ARR预测 · 营收/业绩指引
  Q2 2026E AI Labs 净新增 ARR: $45B
  tags: 好数字 · → daily
  value: qty=$45B · date=Q2 2026E
  [图: 展示 Public Software、Hyperscalers 和 AI Labs 在 Q4 2025 至 Q2 2026E 期间季度净新增 ARR (Quarterly Net New ARR) 趋势的对比柱状图 — Q2 2026E AI Labs 净新增 ARR: $45B; Q2 2026E Hyperscalers 净新增 ARR: $38
  📷 原图
• <a id="atom-efc79b972259f15a"></a>🟦 2026-05-20 X·@theinformation · 推动缩短审查窗口 · 监管政策
  AI labs are pushing for a shorter review window than the proposed 90 days.
  tags: 好信源 · → daily
  value: qty=短于90天 · direction=缩短
  📷 原图
• <a id="atom-06c28b7e874f0a65"></a>🟦 2026-05-18 X·@coatuemgmt · consumption pricing占比 · 定价模式
  AI Labs are 74% consumption
  tags: 好数字 · → daily
  value: qty=74% · date=当前
• <a id="atom-12e152845c32ed71"></a>🟦 2026-05-10 X·@robjama · 内部模型领先优势 · 技术路线/竞争格局
  People at major AI labs (using internal models) 3-4 months ahead of startup silicon valley engineers
  tags: 好信源 · → daily
  value: qty=3-4 months · date=当前
• <a id="atom-9cd02c676a587e43"></a>🟦 2026-04-25 X·@matt_slotnick · run rate revenue increase · 营收增长
  both labs add $20m run rate revenue in the time it takes to write the tweet
  tags: 好数字 · → daily
  value: qty=$20M · direction=increase · periodicity=run rate
• <a id="atom-7d22bbe635365c53"></a>🟦 2026-04-24 X·@based16z · profitability status · 盈利能力/营收
  Meanwhile the AI labs make about the same “annualized” revenues with negative profits
  tags: 好数字 · → daily
• <a id="atom-b79e92f3ac684b3b"></a>🟦 2026-06-30 X·@henrytdowling · app stack investment focus · 应用层/资本开支
  labs are investing a lot into claude code / codex rn
  tags: 好信源 · → daily
• <a id="atom-07bfa62a941b9154"></a>🟦 2026-06-11 X·@chrislakin · 员工因情绪低落休假后离职 · 人事变动/人才流失
  People who get sad, go on sabbatical, then leave
  tags: 好观点·好问题 · → daily
  📷 原图
• <a id="atom-f8679e518f8d060b"></a>🟦 2026-05-19 X·@idavidrein · 开发内部监控AI agent行为系统 · 安全监控/技术路线
  AI labs have started developing systems to monitor internally deployed AI agents for misaligned behavior
  tags: 好信源 · → daily
• <a id="atom-837f939793ff9a8c"></a>🟦 2026-05-04 X·@stevehou · 与 PE 公司组建合资企业 · 合作客户/企业采用
  The AI labs are forming joint ventures with PE firms to tackle the sluggish and uneven adoption of AI within enterprises head on by teaching the portfolio companies held by PE firms how to incorporate AI into their workflows.
  tags: 好信源 · → daily
  📷 原图
• <a id="atom-9d4a00922af790c5"></a>🟦 2026-05-04 X·@flowersslop · free tier 最新模型访问 · 产品发布/竞争格局
  free tier cant use Opus 4.7 or Grok 4.3 and only has very limited access to GPT-5.5
  tags: 好信源 · → daily
  value: qty=无免费访问 · date=当前（作为2026年5月4日前的最新状态）
• <a id="atom-83b440b88ad41148"></a>🟦 2026-06-11 X·@natolambert · 有更强的内部模型 · 竞争格局
  The labs have stronger internal models, yes I know this
  tags: 好观点 · → daily
• <a id="atom-e98dbe4083ffda3a"></a>🟦 2026-05-23 X·@l_mc_nally · label systems as tools · 技术路线/产品发布
  AI labs are already labelling their systems as 'tools'
  tags: 好观点 · → daily
  📷 原图
• <a id="atom-941d8180e71d292b"></a>🟦 2026-05-04 X·@deredleritt3r [老化中] · AI 安全程序独立审计提议 · 监管政策/AI安全
  The article proposes a federal framework under which the labs' safety procedures would be audited by an independent expert body overseen by the U.S. government
  tags: 好观点 · → daily
• <a id="atom-8fbb0f09788636b5"></a>🟦 2026-04-16 X·@tbpn · 最大成本项从研究员时间转变为算力 · 成本结构/数据中心算力
  Recently, AI labs' biggest cost center flipped from researcher time to compute capacity
  tags: 好观点 · → daily

🟥 多头 takes (bullish) (30)

• <a id="atom-52e656c509fba12d"></a>🟥 2026-04-23 X·@scaling01 ⭐ [老化中] · prompting到AGI路径 · 技术路线
  AI labs are literally prompting their way to AGI
  tags: 好观点·好思考 · → daily
• <a id="atom-6c49ce0398990349"></a>🟥 2026-05-07 X·@xlr8harder · 模型供给策略 · 产品路线/定价权
  more diversification in the next 6 months or so as various economically useful capabilities democratize effectively to cheaper models
  tags: 好观点·好思考 · → daily
  value: qty=more diversification · date=next 6 months
• <a id="atom-fdf7f4c6ec69a90b"></a>🟥 2026-05-06 X·@dwarkesh_sp [老化中] · 可以对其最佳模型收取更高利润率 · 竞争格局
  AI labs will be able to charge *higher* margins on their best models as compute gets scarcer.
  tags: 好观点·好思考 · → daily
• <a id="atom-e7be191f5d5383f3"></a>🟥 2026-05-06 X·@samsja19 [老化中] · 定价能力 · 定价权
  AI labs will be able to charge *higher* margins on their best models as compute gets scarcer
  tags: 好思考·好观点 · → daily
  value: direction=higher · qty=higher margins on best models
• <a id="atom-018d10bfaaf42d9f"></a>🟥 2026-05-15 X·@zephyr_z9 · 策略:提升token效率以服务高batch size · 技术路线/数据中心算力
  Also, remember that labs are pushing higher token efficiency, so that they don't need to push TPS super hard and can serve at super high batch sizes
  tags: 好思考 · → daily
  📷 原图
• <a id="atom-9b3b522773f7d523"></a>🟥 2026-05-09 X·@scaling01 · 万亿 ARR 可能性预测 · 营收预测
  AI labs hitting 1T in ARR by the end of 2030 is possible
  tags: 好观点 · → daily
  value: qty=$1T ARR · date=end of 2030
• <a id="atom-c7cb2712038349e6"></a>🟥 2026-05-02 X·@scaling01 [老化中] · 人才来源 · 人才流动
  now everyone is running to the AI labs
  tags: 好观点 · → daily
  value: direction=inflow
  📷 原图
• <a id="atom-b9be016e01e30dfc"></a>🟥 2026-05-01 X·@scaling01 [老化中] · capture disproportionate share of value created · 价值分配/需求旺盛
  Al labs are capturing a disproportionate share of the value being created, driven by strong end demand, rising token monetization, and increasingly favorable unit economics
  tags: 好观点 · → daily
  📷 原图
• <a id="atom-4d7204abb811fea6"></a>🟥 2026-05-01 X·@scaling01 · unit economics are increasingly favorable · 盈利能力
  increasingly favorable unit economics
  tags: 好观点 · → daily
  📷 原图
• <a id="atom-8fe35c4277465eee"></a>🟥 2026-06-13 X·@AndrewCurran_ · RSI时间线 · 技术路线
  we are probably less than a year away from powerful RSI
  tags: 好思考 · → daily
  value: date=不到一年
  📷 原图
• <a id="atom-6736ce2967485299"></a>🟥 2026-05-06 X·@dwarkesh_sp [老化中] · 由于Alchian-Allen效应,市场向赢家通吃方向发展 · 竞争格局
  This pushes the AI model market towards winner-take-all
  tags: 好思考 · → daily
• <a id="atom-7c9182601864952a"></a>🟥 2026-04-29 X·@dwarkesh_sp · 模型训练 over-trained 程度 · 技术路线
  Because of RL, models may be 100x over-trained beyond Chinchilla-optimal
  tags: 好观点 · → daily
  value: qty=100x · date=RL 场景
• <a id="atom-5a065c2b5c3e6bad"></a>🟥 2026-06-30 X·@matt_slotnick · app stack move · 应用层/锁定需求
  labs very noticeably moving up the app stack more aggressively, presumably as a way to lock in your long term token consumption budgets
  tags: 好观点 · → daily
• <a id="atom-5d62db30b90c58c3"></a>🟥 2026-06-30 X·@matt_slotnick · strategy to own last mile for inference demand · 应用层/战略定位
  they really want to own that layer in whatever shape it takes. eg claude, cc, cowork, tag… owning last mile in order to drive inference rather than rely on partner products who will increasingly be multi model
  tags: 好观点 · → daily
• <a id="atom-5032b484d85270c1"></a>🟥 2026-06-26 X·@tokenbender · 短期利好 · 监管政策
  this may be the best thing to happen to these labs for the short term
  tags: 好观点 · → daily
• <a id="atom-74eefc40e37f37f5"></a>🟥 2026-06-25 X·@wordgrammer · automation goal · 技术路线
  They want to automate the job of researchers, not just programmers
  tags: 好观点 · → daily
• <a id="atom-6a5c1b8772c3184b"></a>🟥 2026-06-21 X·@GabGarrett · 提供成本优势 · 定价权/竞争格局
  the labs are providing the giveaway of a lifetime when you look at the costs of running a commensurate model yourself
  tags: 好观点 · → daily
  value: direction=正面
• <a id="atom-5c803b68b60c9d13"></a>🟥 2026-06-21 X·@bookwormengr · every lab with sufficient resources will get to AGI/ASI due to Bitter Lesson and back propagation · 技术路线
  BITTER LESSON and BACK PROPAGATION ensures every lab with sufficient resources will get to AGI/ASI.
  tags: 好观点 · → daily
• <a id="atom-04a285ccd45d649e"></a>🟥 2026-06-18 X·@Mascobot · 当前竞争优势来源 · 竞争格局
  The AI labs wining right now are not the ones with the most compute, but the ones with the best data
  tags: 好观点 · → daily
• <a id="atom-df0401a44d2975f4"></a>🟥 2026-06-13 X·@dalibali2 · 预计资本支出和站点启用不受影响 · 资本开支
  doesn’t change the planned capex and 2027/2028 site go-lives because no lab is going to believe the trajectory is anything but up and up
  tags: 好观点 · → daily
  value: date=2027/2028
• <a id="atom-7d1028739e0e8900"></a>🟥 2026-06-12 X·@vikramskr · Net New ARR projection · 营收/订单需求
  AI Labs are projected to generate the most net new ARR in Q2
  tags: 好观点 · → daily
  value: date=Q2
• <a id="atom-0ed27af3f31fa186"></a>🟥 2026-06-11 X·@gfodor · AI 权重和推理管线披露义务 · 监管政策
  The government can largely de-risk AI by forcing labs to eventually disclose the weights and inference pipelines for any model released to the public.
  tags: 好观点 · → daily
• <a id="atom-9fece6c8a4d64d65"></a>🟥 2026-06-08 X·@natolambert · focus on most useful technology quickly to avoid financial implosion · 技术路线
  the labs are focused on the most useful technology quickly, else they financially implode
  tags: 好观点 · → daily
• <a id="atom-056bda10c995d38f"></a>🟥 2026-06-06 X·@mattlam_ · model routing 作为重要层 · 模型路由
  if companies continue to have AI spend problems I think model routing will be an important layer AI labs work on
  tags: 好观点 · → daily
• <a id="atom-eadf23a2469c6355"></a>🟥 2026-05-16 X·@acacuce · 内部计算成本低于 1 名工程师工资 · 成本结构
  internal costs for compute are probably less than the salary of 1 eng
  tags: 好观点 · → daily
• <a id="atom-2b62223272e0c19d"></a>🟥 2026-05-11 X·@max_paperclips [老化中] · FDE团队负责建立问责制是正确举措 · 管理层表态
  an FDE team to have accountability is a good move
  tags: 好观点 · → daily
• <a id="atom-ca1be27765a76c0b"></a>🟥 2026-05-10 X·@zephyr_z9 · IPO时间触发软件做多 · 合作客户
  when the AI labs IPO
  tags: 好观点 · → daily
• <a id="atom-511c71762c8dd0f8"></a>🟥 2026-05-08 X·@krinza · 未来3年建立生物部门与湿实验室 · 技术路线/生物医药
  in the next 3 years, every major AI lab will spin up its own bio arm and in-house wet labs.
  tags: 好观点 · → daily
  📷 原图
• <a id="atom-8ae6de00c67f5a87"></a>🟥 2026-05-07 X·@dejavucoder · 有经济学定理表明其最强模型利润率会更高 · 定价权/供给产能
  AI labs will be able to charge *higher* margins on their best models as compute gets scarcer.
  tags: 好观点 · → daily
  value: direction=positive
• <a id="atom-8ee626835e92b90c"></a>🟥 2026-05-06 X·@dwarkesh_sp [老化中] · 随着计算能力短缺,可以捕获更高利润率 · 竞争格局
  as the compute shortage hits, AI labs can capture more margin
  tags: 好观点 · → daily

🟥 空头 takes (bearish) (30)

• <a id="atom-d5ce7be69fc1fe99"></a>🟥 2026-06-26 X·@8teAPi ⭐ · IPO timing · 仓位情绪/融资
  all AI labs IPOing at once
  tags: 好观点·好思考·好信源 · → daily
  value: direction=simultaneous
• <a id="atom-c7ba91bd1088c0e7"></a>🟥 2026-04-15 X·@scaling01 ⭐ [老化中] · job loss actions · 劳动力影响
  jobs losses are also predictable now that AI labs have started hill-climbing powerpoint, excel, word and specific professions
  tags: 好观点·好思考 · → daily
• <a id="atom-07d9e8f6ce402beb"></a>🟥 2026-05-13 X·@Crussian17 · revenue growth will slow hard due to token optimization · 营收/盈利能力
  when that happens rev growth for labs will slow hard
  tags: 好观点·好思考 · → daily
• <a id="atom-a871dd678b8dcb97"></a>🟥 2026-06-25 X·@AndrewCurran_ · model release pace slowing · 模型发布
  This does not slow development in any way, it only slows the rate at which the labs can release models, not how fast they can train them.
  tags: 好观点·好思考 · → daily
• <a id="atom-ccdb0c06e89263c8"></a>🟥 2026-05-27 X·@matt_slotnick · become optimal destination to solve customer problems · 竞争格局/技术路线
  when they become the optimal destination to solve the problems of your customers, it’s going to get ugly
  tags: 好观点·好思考 · → daily
• <a id="atom-403785677040aa52"></a>🟥 2026-04-17 X·@bookwormengr · 将开始收取收入利润分成 · 合作客户/定价权
  We are near the point in history where labs start taking a share in your revenue and profits. They will give access to vertical focussed models only to those companies that agree with this!
  tags: 好观点·好思考 · → daily
  value: qty=share in your revenue and profits · date=near future
• <a id="atom-6a26e087c06a752f"></a>🟥 2026-06-22 X·@stevehou · AI-related news could include slowdown of capex due to capital planning or local pushback · 资本开支/竞争格局
  I imagine it'll likely have to do with AI-related news such as a slowdown of capex due to either capital planning or local pushback and the threat of open models as economics of the labs continue to become public knowledge
  tags: 好思考 · → daily
• <a id="atom-de13530f8c86bd61"></a>🟥 2026-04-21 X·@Yuchenj_UW [老化中] · 投入开源的趋势 · 开源
  More and more AI labs seem to be pulling back from open source
  tags: 好思考 · → daily
  value: direction=减少
• <a id="atom-ba1db3ec11cfc275"></a>🟥 2026-07-01 X·@mweinbach · 不愿讨论推理芯片 · 竞争格局
  I've heard from quite a few companies that they don't want to talk about silicon or inference, they simply want to talk about the output and cost.
  tags: 好观点 · → daily
• <a id="atom-7610a527a37e32d3"></a>🟥 2026-07-01 X·@mweinbach · 意图隐藏利润率和吞吐量 · 盈利能力
  Labs want to sell you intelligence at a certain cost, the less you know about their margin and throughput the better, because there's less room to negotiate
  tags: 好观点 · → daily
• <a id="atom-76ac2c6e4741dda1"></a>🟥 2026-06-26 X·@WolfeFolks · 需要暂停以不流血出局 · 资本开支
  the pause they need to not bleed out
  tags: 好观点 · → daily
• <a id="atom-87c01f22aa7d3819"></a>🟥 2026-06-26 X·@iamgingertrash [老化中] · revenue growth needs for fundraising · 融资
  don't think the labs need to show revenue growth at all in order to raise the next round
  tags: 好观点 · → daily
  value: qty=not needed
• <a id="atom-c6de8b474badfcb7"></a>🟥 2026-06-25 X·@iamgingertrash · revenue growth stall · 营收
  Revenue growth stall
  tags: 好观点 · → daily
  📷 原图
• <a id="atom-5102858e8b7542ad"></a>🟥 2026-06-25 X·@iamgingertrash · no money for next 10x training run · 融资/资本开支
  No $$ for next 10x training run
  tags: 好观点 · → daily
  value: qty=no $ · date=next 10x training run
  📷 原图
• <a id="atom-66e46b01ca0bea1c"></a>🟥 2026-06-25 X·@iamgingertrash · no available compute · 供给产能/资本开支
  No available compute
  tags: 好观点 · → daily
  📷 原图
• <a id="atom-99e4e54464f4cde2"></a>🟥 2026-06-25 X·@iamgingertrash · cannot serve new customers · 订单需求/营收
  Can’t serve new customers
  tags: 好观点 · → daily
  📷 原图
• <a id="atom-ee5810a02d356a3f"></a>🟥 2026-06-22 X·@wordgrammer · 客户关系分类 · 竞争格局/市场份额
  There are two types of AI labs. Those that are beloved by everyone, and those that have customers
  tags: 好观点 · → daily
• <a id="atom-3b5bbc0fa2243bf2"></a>🟥 2026-06-21 X·@bookwormengr · frontier is hard; failures by GDM & xAI show that · 竞争格局
  Yes, staying at frontier is hard as failures to do so by GDM & xAI have shown; but their models are still usable but not just SOTA or most pleasant.
  tags: 好观点 · → daily
• <a id="atom-d49f0aeae0e9aedb"></a>🟥 2026-06-13 X·@jasonschips · demand to train future models · 订单需求/资本开支
  Less demand to train future models
  tags: 好观点 · → daily
  value: direction=lower
• <a id="atom-a77e6cb0ee22e54e"></a>🟥 2026-06-13 X·@jasonschips · uncertainty premium · 仓位情绪/风险溢价
  Uncertainty premium
  tags: 好观点 · → daily
  value: direction=increased
• <a id="atom-ce4dfa161c0b939f"></a>🟥 2026-06-09 X·@suchenzang · might spend capex with no intention of democratizing technology · 资本开支
  with the amount of capex spent so far, this was never meant to be some democratizing technology 'for the people'.
  tags: 好观点 · → daily
• <a id="atom-425c5ae9794818e0"></a>🟥 2026-06-09 X·@deanwball · 可能被政府秘密强制降低性能 · 监管政策
  the government, with a classified regulatory system for AI (as the recent cyber EO tees up) could secretly force AI labs to silently degrade performance in a wide range of domains without *that* much risk of First Amendment litigation
  tags: 好观点 · → daily
• <a id="atom-b15ef4e400c88291"></a>🟥 2026-06-05 X·@altcap [老旧] · 反对政府国有化 · 监管政策/竞争格局
  I’m against gov’t “taking”/ nationalizing AI labs / other AI biz - terrible slippery slope / crony capitalism etc.
  tags: 好观点 · → daily
• <a id="atom-25e89f88eeab4ec3"></a>🟥 2026-06-04 X·@opinion · 技术成本上涨 · 成本
  the technology costs more than ever — and more companies are deciding they can't afford it
  tags: 好观点 · → daily
  value: date=2026-06-04
• <a id="atom-c3bfcc2f754d9483"></a>🟥 2026-05-30 X·@Dorialexander · SaaS信仰 · 商业模式/定价权
  The actual paradox is that labs remain true SaaS believer even as they crash this model.
  tags: 好观点 · → daily
• <a id="atom-caff59a6bbaf4d6d"></a>🟥 2026-05-30 X·@willccbb · 蒸馏规则合规性 · 技术路线/监管政策
  it's often 'against the rules', sure
  tags: 好观点 · → daily
• <a id="atom-2daec9d548136497"></a>🟥 2026-05-28 X·@garybasin · market entry count · 竞争格局
  It probably is when you have half a dozen entrants at once
  tags: 好观点 · → daily
  value: qty=half a dozen
• <a id="atom-0476db6dd11b124d"></a>🟥 2026-05-27 X·@matt_slotnick · absorb top talent · 竞争格局/人才流动
  biggest risk to incumbents may be that labs absorb all the top talent. labs already get great eng talent. but this will spread to product… sales… marketing… etc. when they become the optimal destination to solve the problems of your customers, it’s going to get ugly
  tags: 好观点 · → daily
• <a id="atom-b000fc3f647669e7"></a>🟥 2026-05-24 X·@natolambert · 历史上不公开且不真正从事科学 · 竞争格局
  Tradition doesn't show that these labs are particularly open and actually doing science these days
  tags: 好观点 · → daily
• <a id="atom-02ed1f97eedc1c67"></a>🟥 2026-05-17 X·@Presidentlin · prediction about software engineer replacement by AI was horribly wrong · 技术路线
  It's clear now that prediction was horribly wrong, contrary to what the Claudesceps will have you think.
  tags: 好观点 · → daily

🟥 中性 takes (neutral) (20)

• <a id="atom-0722e50bffb022bc"></a>🟥 2026-06-26 X·@filipviz · model capabilities depend on pretraining and RLVR hill-climbing, not just distillation · 技术路线/竞争格局
  Thus, "distillation" is moreso an cost-saving post-training shortcut. They cheaply take model reasoning from zero to one, but the labs must still climb from one to one hundred.
  tags: 好观点·好思考 · → daily
• <a id="atom-8abb9e8953fb905f"></a>🟥 2026-05-12 X·@emollick [老化中] · 东海岸共识认为ASI变革更渐进而非急速 · 技术路线
  the East Coast consensus ... even true ASI is less abruptly transformative than the West Coast vision of massive rapid automation
  tags: 好观点·好思考 · → daily
• <a id="atom-290b900591cdfd63"></a>🟥 2026-05-11 X·@emollick [老化中] · 咨询部门解散是相信ASI的标志 · 管理层表态
  You will know that the AI labs believe in ASI when they disband their newly formed consulting (sorry “forward deployed engineering”) groups.
  tags: 好观点·好思考 · → daily
• <a id="atom-84a11c3f0948fa5d"></a>🟥 2026-05-11 X·@emollick [老化中] · 咨询部门表明当前需要人工判断AI用途、组织变革和系统集成 · 订单需求/竞争格局
  As long as people are required to figure out how AI is useful & do organizational change & systems integration, jobs seem to be pretty safe
  tags: 好观点·好思考 · → daily
• <a id="atom-cd4461853d054a39"></a>🟥 2026-05-12 X·@emollick [老化中] · ASI无法最优地处理信息流 · 技术路线
  A true superintelligence should be able to infer or elicit tacit/contextual knowledge? Or route around it?
  tags: 好思考·好问题 · → daily
• <a id="atom-fd74a161d9cf5596"></a>🟥 2026-06-11 X·@crackalamoo · 模型保留态度 · 地缘政治
  The labs would start keeping the models to themselves to avoid release, and arguing that releasing such powerful weights allows hacks that would be a danger to society.
  tags: 好思考 · → daily
• <a id="atom-659b485b7ca44154"></a>🟥 2026-05-24 X·@natolambert · AI evolution is impacted by more than just mega compute · 技术路线
  other things impact the evolution of AI than just the few labs with mega compute
  tags: 好思考 · → daily
• <a id="atom-17e070d26905cc97"></a>🟥 2026-05-02 X·@TBU12345678 · 优化趋势预期 · 资本开支
  This sets up for “AI optimizations” in the same way we saw “cloud optimizations” with $SNOW and $DDOG in 2022.
  tags: 好思考 · → daily
• <a id="atom-af9c9285bd6af1ed"></a>🟥 2026-07-03 X·@mignano · models 追赶速度 · 技术路线
  Everyone catches up within months.
  tags: 好观点 · → daily
  value: date=months
• <a id="atom-8bb0df162334af3a"></a>🟥 2026-07-03 X·@mignano · 资本获取 · 融资
  Capital? You don’t even need a working model to raise right now. Everyone wants in on AI.
  tags: 好观点 · → daily
• <a id="atom-45301fce0c0c1d44"></a>🟥 2026-07-03 X·@mignano · 人才获取 · 人事变动
  Talent? Follows the capital.
  tags: 好观点 · → daily
• <a id="atom-0f46ac3bc61a410e"></a>🟥 2026-07-03 X·@mignano · 训练数据 · 技术路线
  Training data? The web’s been scraped to death. Everyone’s got the same stuff.
  tags: 好观点 · → daily
• <a id="atom-5f18c8f9f1d76914"></a>🟥 2026-06-21 X·@bookwormengr · AI技术可复制性观点 · 技术路线
  The real reason is that AI is relatively EASY critical technology that is there to master
  tags: 好观点 · → daily
  value: direction=容易掌握
• <a id="atom-1dca3a0f801d3a45"></a>🟥 2026-06-12 X·@StableBread · advanced models access restriction · 监管政策
  _@borrowed_ideas my guess they gatekeep it to select few who can afford it_
  tags: 好观点 · → daily
• <a id="atom-0d98177a783669f9"></a>🟥 2026-05-30 X·@willccbb · 蒸馏研究重点 · 技术路线/研发
  the curation and recipe design and infra is where researchers spend their time anyway
  tags: 好观点 · → daily
• <a id="atom-690f031babb70a13"></a>🟥 2026-05-29 X·@mweinbach · 计划发布 · 模型发布
  I would not put it past AI labs to release GPT 5.12 or Opus 4.15
  tags: 好观点 · → daily
  value: qty=GPT 5.12或Opus 4.15
• <a id="atom-a3156efa7210ac85"></a>🟥 2026-05-17 X·@Presidentlin · researchers will leave to start new companies or retire after IPO (forecast) · 人事变动
  Many of the researchers will move on to found new companies or just retire since they will be rich.
  tags: 好观点 · → daily
• <a id="atom-c89c02a349da9e61"></a>🟥 2026-05-16 X·@flowersslop · pour billions of dollars into RL so you can just talk to it in normal english · 技术路线
  the labs pour billions of dollars into RL so you can just talk to it in normal english, there are no special prompts or tricks that make the models suddenly be way smarter or something
  tags: 好观点 · → daily
• <a id="atom-74f301b8174673f3"></a>🟥 2026-05-14 X·@Dorialexander · solving data quality via contracting ex-labs people · 人才招聘
  i feel labs are solving it by mostly contracting with ex-labs people
  tags: 好观点 · → daily
• <a id="atom-9f4651aea4f4a5de"></a>🟥 2026-05-12 X·@emollick [老化中] · 治愈癌症比取代埃森哲容易 · 竞争格局/订单需求
  "curing cancer will be easier than replacing Accenture"
  tags: 好观点 · → daily

⏱ 时间轴 (近 20)

• 🟥 2026-07-03 · narrative · models 追赶速度 · → daily
• 🟥 2026-07-03 · narrative · 资本获取 · → daily
• 🟥 2026-07-03 · narrative · 人才获取 · → daily
• 🟥 2026-07-03 · narrative · 训练数据 · → daily
• 🟥 2026-07-01 · narrative · 不愿讨论推理芯片 · → daily
• 🟥 2026-07-01 · narrative · 意图隐藏利润率和吞吐量 · → daily
• 🟥 2026-06-30 · narrative · app stack move · → daily
• 🟦 2026-06-30 · fact · app stack investment focus · → daily
• 🟥 2026-06-30 · narrative · strategy to own last mile for inference demand · → daily
• 🟥 2026-06-29 · narrative · lock in exists today · → daily
• 🟥 2026-06-29 · narrative · 商业模式取决于通用性特征 · → daily
• 🟥 2026-06-26 · narrative · 短期利好 · → daily
• 🟥 2026-06-26 · narrative · 需要暂停以不流血出局 · → daily
• 🟥 2026-06-26 · narrative · model capabilities depend on pretraining and RLVR hill-climb · → daily
• 🟥 2026-06-26 · narrative · next training paradigm · → daily
• 🟥 2026-06-26 · narrative · IPO timing · → daily
• 🟥 2026-06-26 · position · revenue growth needs for fundraising · → daily
• 🟥 2026-06-26 · narrative · 经济模式 · → daily
• 🟥 2026-06-25 · narrative · automation goal · → daily
• 🟥 2026-06-25 · forecast · future action needed for synthetic data protection · → daily

← 实体目录 · 系统日志