Financing balance in two markets decreases by 140.42 billion yuan 两市融资余额减少140.42亿元

When juxtaposed with the bold declarations of our neighbor Mr. Huang (Jensen Huang), the drama becomes palpable. On one side, leveraged retail investors are “running” (liquidating positions); on the other, the global warlord of AI compute is locking eyes with SK Hynix, pledging to continue investing billions over the coming years to secure every last top-tier HBM (High Bandwidth Memory) chip available. He said the collaboration would last “more than two years”—translation: the AI fire will burn for at least two more years, my GPUs will sell themselves, but the real worry is having enough “food” (chips) to keep the kitchen running.

One is shrinking risk exposure; the other is frantically locking down future “food” supplies. Together, these two events paint the most authentic, fractured picture of the current AI capital narrative: at the top, the arms race has entered a phase of “stockpiling” and “ally-building,” where lithography machines and high-end memory chips have become harder currency than gold. Meanwhile, the downstream capital markets—especially the A-share market, driven by expectations and liquidity—are already showing signs of fatigue toward this long, expensive “future story.” The 14 billion yuan net outflow in margin financing doesn’t mean people have stopped believing in AI; rather, the belief is there, but wallets are waving the white flag first. After all, you can’t eat promises, but margin interest has to be paid back in real cash.

Interestingly, another trending topic—“After using AI, my company seems even poorer”—hangs on the hot search list. This is the perfect darkly humorous footnote. Nvidia and its ilk are raking in profits, and upstream suppliers like SK Hynix have orders pouring in. Yet the broad application-layer companies—those investing real money to actually “use” AI—find themselves trapped in an “AI tax” pit. Sky-high API fees, exorbitant compute costs, business processes needing complete overhaul… efficiency gains are nowhere in sight, but the cost hole has already ripped open. This isn’t “empowerment”; it’s “enpowerment”—a heavy financial burden strapped on.

So you see, the AI world today presents a bizarre “inverted pyramid” prosperity: the top-tier infrastructure and chip suppliers are gobbling up the lion’s share of industry profits and attention, with every Jensen Huang pronouncement acting like a market sedative. The middle layer of large model companies is burning through cash in a cutthroat frenzy. Meanwhile, the broadest layer—the industry application layer, which should be where value is actually realized—is bearing cost pressures and sending cautious signals like the drop in margin financing balances. Capital is becoming “smarter” and more “short-sighted”—it would rather chase the certain, imminent billions in orders Huang speaks of than keep funding distant AI application scenarios that require endless cash to experiment with.

The 600% surge in night market stall equipment, “Little Sam’s Club” becoming the top mall trend… these hot topics, Jensen Huang’s speeches, and market data may seem entirely unrelated, yet they all tell the same story of an economic season: high-end industries are betting everything on the future, while mass consumer spending is seeking value-for-money exits. AI is the vast uncharted territory of stars and seas, but for most, the fare for the journey hasn’t even been scraped together yet. When the dream of “AGI (Artificial General Intelligence)” meets reality, the first question is: Who’s going to pay this month’s API bill?

A股的融资余额一天没了140多亿，这消息一出，整个盘面就像被抽掉了一管血。上证那边是重灾区，跑了113个亿，深证也意思了一下。近三万亿的融资盘，听着吓人，但数字本身是冰冷的，它背后是无数个账户里，那些试图用杠杆撬动财富的神经，在某个瞬间集体松动了。

这和隔壁老黄（黄仁勋）的豪言壮语放一起看，戏剧性就来了。这边厢，加杠杆的股民在“润”；那边厢，全球AI算力的军火头子，在跟SK海力士确认眼神，表示未来几年还要继续砸下数十亿美元，把最顶级的HBM（高带宽内存）芯片包圆。他说合作要“超过两年”，这话翻译过来就是：AI这股火，至少还得烧两年，我家的显卡不愁卖，但愁的是没足够的“粮食”（芯片）下锅。

一个在收缩风险敞口，一个在疯狂锁定未来的“粮食”供应。这两件事摆在一起，勾勒出当下AI资本叙事最真实的撕裂图景：顶层的军备竞赛进入“拼库存”、“拼盟友”的卡脖子阶段，光刻机、高端存储芯片成了比黄金还硬的通货；而下游的资本市场，特别是依赖预期和流动性驱动的A股，却已经开始对漫长而昂贵的“未来故事”感到一丝疲态。140亿的融资净流出，不是说大家不信AI了，而是信归信，钱包先认怂了。毕竟，画饼不能当饭吃，融资利息却是要实打实还的。

有意思的是，热搜榜上挂着的另一条——“用了AI之后，公司好像更穷了”。这简直是绝妙的黑色幽默注脚。英伟达们赚得盆满钵满，SK海力士这类上游供应商订单接到手软，但广大的应用层公司，那些真金白银投入去“用”AI的企业，却发现自己陷入了“AI税”的陷阱。高昂的API调用费、天价的算力成本、需要重写的业务流程……效率的提升还没见着，成本的窟窿先撕开了。这哪是“赋能”，简直是“附能”——附加上了沉重的财务负担。

所以你看，现在的AI世界，呈现一种诡异的“倒金字塔”繁荣：顶端的基础设施和芯片供应商，吃掉了行业绝大部分的利润和关注度，黄仁勋的每次发言都像市场定心丸；中间层的大模型厂商在卷生卷死，疯狂烧钱；而最广阔的、本应是价值落地所在的行业应用层，却在承受成本压力，并反馈出像融资余额下降这样的谨慎信号。资本市场的钱，正变得更“聪明”，也更“短视”——它们宁愿去追逐黄仁勋口中“数十亿美元”这样确定的、即将到来的订单，也不愿为遥远的、需要用无数真金白银去试错的AI应用场景持续买单了。

夜市摆摊设备暴涨600%，“小山姆”成商场顶流……这些热搜和黄仁勋的讲话、资本市场的数据，看似风马牛不相及，却共同诉说着同一个经济季节的故事：高端产业在孤注一掷地豪赌未来，而大众消费在寻找性价比的出口。AI是星辰大海，但眼下，大多数人的船票钱都还没凑齐。当“AGI（通用人工智能）”的梦想照进现实，第一个问题是：这个月的API账单，谁来付？