一个AUM从200亿美元跌到100亿、刚把公开持仓清得差不多的AI对冲基金，本周又砸出4亿美元。
据TechCrunch报道，Situational Awareness本周向Source Foundry投资4亿美元——一家由斯坦福研究人员创立的芯片制造创业公司，目标是让芯片制造“更快、更便宜”。加上此前的投资，该基金对Source Foundry的总投入已达到5亿美元。
一边是AI基建股暴跌、基金被迫割肉回撤，一边是对芯片制造环节的重仓加码。这笔反差感拉满的交易，把AI资本的最新流向摆到了台面上：钱正在从'卖铲子的'，流向'造铲子的人'。
最狂的AI对冲基金：从All in到逆势All in
Situational Awareness这个名字，在AI圈几乎是“疯狂”的代名词。创始人Leopold Aschenbrenner是前OpenAI研究员，二十多岁，2024年创立基金时没有任何交易经验，却靠重仓AI基础设施股票在早期拿到了惊人的回报。他写过的长文《Situational Awareness》至今仍是AI投资圈的一份“圣经”式文本。
但今年夏天，风向变了。AI基础设施股票大幅回撤，基金遭遇严重亏损；7月底，它把大部分公开持仓卖给了Ken Griffin的Citadel，只保留了Anthropic的股份。AUM从200亿美元缩水到100亿美元——对普通基金这是灾难，对这家基金，更像是一次主动换仓。
因为就在“割肉”之后不到两周，它又对Source Foundry追加了4亿美元。这种“刚清仓就重仓”的节奏，与其说是抄底，不如说是它一直以来的打法：极度集中、极度自信、只押自己看得懂的叙事。
为什么是芯片制造
这笔投资踩中的，是AI算力供应链里最稀缺、也最被忽视的一环。过去两年，AI资本疯狂涌向GPU、数据中心、电力——英伟达吃肉，云厂商喝汤，创业公司啃骨头。但芯片制造（foundry）本身，始终是少数寡头的游戏：晶圆代工门槛极高，新玩家几乎进不来。
Source Foundry试图打破的就是这个格局：让芯片制造更快、更便宜，把造芯的门槛降下来。如果这条路走得通，它改变的不只是一家公司的估值——而是整个AI算力供给的游戏规则。对一个坚信“算力是新时代石油”的基金来说，投制造环节，等于投石油时代的炼油厂。
当然，这笔交易也带着鲜明的赌性：Source Foundry还是stealth状态，技术细节没有公开；4亿美元的估值依据，更多是创始团队的出身和赛道想象力。在AI基建股的暴跌声里下重注，要么是看到了别人没看到的东西，要么是信仰足够深。
对行业观察者来说，比单笔交易更有意义的，是资金流向的指示作用：当最懂AI的一批人（前OpenAI研究员、AI安全圈的核心人物）开始把真金白银押在“芯片制造”上，说明算力焦虑已经从“买不到卡”，升级到了“卡不够造”。算力基建的下半场，可能不在云端，而在晶圆厂里。
明天见。
A hedge fund whose AUM fell from $20 billion to $10 billion — and just sold off most of its public portfolio — just threw another $400 million this week.
Per TechCrunch, Situational Awareness invested $400 million in Source Foundry, a stealth chip-manufacturing startup founded by Stanford researchers aiming to make chipmaking "faster and cheaper." Combined with prior investments, the fund's total stake in Source Foundry now reaches $500 million.
On one side: AI infrastructure stocks crashing, the fund forced into painful drawdowns. On the other: doubling down on chip manufacturing. This contradiction-heavy deal puts AI capital's latest direction on the table: money is flowing from the shovel-sellers to the shovel-makers.
The Wildest AI Hedge Fund: From All-In to Counter-Trend All-In
In AI circles, the name Situational Awareness is nearly synonymous with "crazy." Founder Leopold Aschenbrenner is a former OpenAI researcher in his mid-twenties with zero trading experience when he launched the fund in 2024 — yet heavy bets on AI infrastructure stocks produced spectacular early returns. His long essay "Situational Awareness" remains a scripture-like text for AI investors.
But this summer, the wind turned. AI infrastructure stocks routed, the fund suffered steep losses; in late July, it sold most of its public portfolio to Ken Griffin's Citadel, keeping only its Anthropic shares. AUM fell from $20 billion to $10 billion — a disaster for a normal fund. For this one, it looked more like an active rotation.
Because less than two weeks after that "capitulation," it dropped another $400 million on Source Foundry. This rhythm of "sell everything, then go big" isn't bottom-fishing — it's the playbook it always ran: extremely concentrated, extremely confident, only betting narratives it believes it understands.
Why Chip Manufacturing
This investment hits the scarcest, most overlooked link in the AI compute supply chain. For two years, AI capital has flooded GPUs, data centers, and power — NVIDIA eats, clouds sip, startups chew bones. But chip manufacturing itself remains a game of a few oligarchs: foundry barriers are enormous, new entrants almost never get in.
Source Foundry's bet is breaking that: make chipmaking faster and cheaper, lower the barrier to building chips. If it works, it changes more than one startup's valuation — it rewrites the rules of AI compute supply. For a fund convinced that compute is the new oil, investing in manufacturing is investing in the refineries of the oil age.
Of course, the trade carries unmistakable gambling flavor: Source Foundry is still in stealth, technical details undisclosed; the $400 million pricing rests largely on the founders' pedigree and the track's imagination. Placing heavy bets amid the AI-infrastructure rout means either you see something others don't, or your conviction runs very deep.
For industry watchers, the signal matters more than the single deal: when the people who best understand AI — ex-OpenAI researchers, core figures in AI safety circles — start parking real money in "chip manufacturing," compute anxiety has upgraded from "can't buy cards" to "there aren't enough cards being made." The second half of compute infrastructure may not be in the cloud. It may be inside the fab.
See you tomorrow.
当最懂AI的一批人把真金白银押在'芯片制造'上，算力焦虑已经从'买不到卡'，升级到了'卡不够造'。
—— Dawn Vision 判断
When the people who best understand AI park real money in chip manufacturing, compute anxiety has upgraded from 'can't buy cards' to 'there aren't enough cards being made.'
— Dawn Vision analysis
算力基建 · Situational Awareness · Source Foundry · 芯片制造 · Aschenbrenner · AI对冲基金 · 晶圆代工 · 融资 · 算力焦虑
Compute Infrastructure · Situational Awareness · Source Foundry · Chip Manufacturing · Aschenbrenner · AI Hedge Fund · Foundry · Funding · Compute Anxiety
Sources · 信源 Sources
本文基于 Dawn Vision 认知引擎处理的公开信息整理，素材来源：TechCrunch、华尔街日报、TechCrunch后续报道。
This article is based on public information processed by Dawn Vision. Sources: TechCrunch, The Wall Street Journal, TechCrunch follow-ups.
