文 | Tech商业
在硅谷的创投神话里，我们习惯了听那些关于改变世界、技术信仰和极客精神的故事。但如果你真正剥开这层光鲜的外衣，深入到旧金山（SF）一线创业者的圈子里，你会发现，这里正在上演一场极其荒诞、甚至令人不安的“数据大跃进”。
最近，一位在旧金山科技圈浸淫了几个月的人士，在社交网络上极其直白地拆解了当下硅谷AI圈的“游戏规则”：
两周内拿下价值300万美元的实验室订单，乘以26，对外宣称年收入（ARR）约为8000万美元。 将全年的所有订单都安排在同一个月内签署，然后把业绩最好的这个月，直接乘以12当作全年的数据。 和朋友的初创公司、大学同学互相签署500万美元的意向书（LOI）。你签一个，我签一个。这在大型模型实验室和算力中心圈子里也屡见不鲜。 花6美分一个的价格，去黑市买GitHub星标（Star）。
这位业内人士发出一声沉重的叹息：“上一家靠和合作伙伴互换发票、吹出2.2亿美元营收的公司，现在已经破产了。令人难过的是，现在完全变成了一场格雷欣法则（劣币驱逐良币）的游戏，如果你不参与造假，你就会被淘汰。”
这段话，就像一个清晰的切口，揭示了2026年AI创投圈最突出的问题。当技术指标的狂欢掩盖不了商业落地的贫血时，一场由上至下的“数据包装运动”，正在成为硅谷心照不宣的生存法则。
被玩坏的ARR：两周订单乘26的“魔法”
在SaaS（软件即服务）时代，ARR（年度经常性收入）是衡量一家公司健康度的核心指标。但在如今的AI大模型初创圈，ARR已经变成了一个可以被随意揉捏的橡皮泥。
“两周内拿下300万美元订单，乘以26就是8000万ARR。”这个逻辑在商业上简直荒谬至极。一笔短期的、一次性的测试订单，硬生生被包装成了长期稳定的订阅收入。这就好比一个餐厅开业第一周生意火爆，老板立刻宣布全年的净利润将翻50倍。
为什么创始人要这么干？因为华尔街和VC们只看这个。在顶级风投的模型里，如果你的ARR不能在一个季度内翻倍，你根本连见合伙人的资格都没有。在极度焦虑的FOMO（错失恐惧症）情绪下，投资人拿着计算器在寻找下一个百亿独角兽，而创业者则在拼命寻找符合公式的“数字拼图”。
于是，把一年的收入压在一个月确认（制造恐怖的环比增长曲线），或者把两个月的爆发性收入年化，成了大家心照不宣的“财务化妆术”。这种“拔苗助长”的数据，不仅扭曲了公司的真实估值，更让整个行业的基准线变得极其离谱。当所有人都在用26倍速吹牛时，那些老老实实按实际交付记录收入的公司，看起来就像是快要破产的蜗牛。
互刷单与意向书魔术：一场金融炼金术
推文里提到的第三点，是极其经典的“互刷单”（Round-tripping）。
初创公司A与初创公司B是朋友开的，A向B购买500万美元的算力或软件服务，B同时向A购买同等金额的服务。双方签下毫无实质交易背景的合同，甚至根本不需要真实的资金流转。然后，两家公司拿着这些合同去告诉投资人：“看，我们的商业化落地能力有多强，大客户都在抢着买单！”
这不仅是面向VC的包装手法，更是撬动金融杠杆的支点。
推文特别指出，这在大型模型实验室和算力中心（Neocloud）圈子里也普遍存在。很多算力租赁公司为了向银行获取高额贷款，会与模型公司签订天价的长期框架协议（LOI）。模型公司获得了纸面上的算力保障，算力公司拿着这些合同去抵押融资、扩大债务杠杆。
在这个链条里，订单不再是为了服务客户，而是为了生成一份能够拿去抵押的凭证。这就解释了为什么像CoreWeave这样的算力巨头，虽然营收利润看似惊人，却背负着几百亿美元的沉重债务。当链条断裂，大家互换的发票变成了一堆废纸。那家曾经靠此吹出2.2亿美元营收、最终破产清算的公司，就是对这种虚假繁荣最无情的嘲弄。
6美分一个的声誉：被明码标价的开源指标
如果说财务包装是为了吸引资本，那买GitHub星标，就是为了影响技术圈的判断。
GitHub的Star数量，原本是衡量一个开源项目受开发者欢迎程度和代码质量的重要指标。但在这位人士的爆料中，6美分就能买到一个星标。
在灰产平台上，这种产业链极其成熟。初创公司花几百美元，就能让一个刚上线、毫无实用价值的仓库，在一夜之间拥有上万个Star。在部分非技术背景的投资人看来，这就代表了“极客圈的狂热追捧”。
不仅如此，日活用户（DAU）、API调用量、甚至模型在各类跑分榜单上的排名，都可以通过雇佣“赛博水军”和刷量脚本在短时间内造出虚假繁荣。这是对整个软件工程精神最彻底的偏离。当代表极客荣誉的徽章，变成了流水线上明码标价的商品，底层技术的声誉就被彻底按在地上摩擦。
格雷欣法则的后果：谁在推动劣币驱逐良币？
这位创业者最后感叹的“格雷欣法则”（劣币驱逐良币），是整段推文中最悲哀、也最真实的一笔。
在这个剧场里，前排的人站起来了，后排的人就必须踩在椅子上。如果你不刷单、不吹高ARR、不买星标，你的估值就会比别人低，你就招不到最贵的天才，你就买不到英伟达的显卡。
究其根本，大模型是一场极其残酷的重资产消耗战。昂贵的算力账单和顶尖人才薪酬，促使初创公司必须拼命融资。而为了拿到融资，他们只能迎合VC对“超高增长”的强烈偏好。
但这背后，VC真的是无辜的受害者吗？并非如此。在某种程度上，VC也参与了这种数据包装的循环。很多投资机构内部也有KPI压力，基金经理需要快速投出明星项目、推高账面回报，才能在下一期基金向LP（出资人）募资。于是，创业者用包装过的数据争取高估值，VC用高估值项目向LP展示回报，LP再指望下一轮接盘。这完全是一场击鼓传花式的循环。
结语：回归常识，拒绝为泡沫买单
旧金山科技圈正在经历的这场“数字大跃进”，像极了2000年互联网泡沫破裂前夕的荒诞岁月。
当时的初创公司，只要在名字里加个“.com”就能股价翻倍；现在的AI初创，只要把两周的订单乘以26，就能自称下一个OpenAI。
对于普通创业者和从业者来说，更稳妥的选择是：不要参与这个循环，不要成为虚报数据的共同参与者。
真正伟大的公司，其核心财务模型永远是简单的：实实在在的营收，减去低于营收的成本，剩下真实的利润。无论是OpenAI还是DeepSeek，在剥离了所有花哨的包装后，拼的依然是算法效率与真实的商业变现能力。
当泡沫破裂时，你会发现，最靠谱的护城河，永远不是花6美分买来的星标，也不是那些互相换来的合同，而是那些愿意在毫无泡沫的寒冬里，为你真实产品买单的客户。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Man, Tech Business
In the creation myths of Silicon Valley, we are used to listening to stories about changing the world, technological beliefs and the spirit of hospitability. But if you really take off this nice coat and go deep into the circle of entrepreneurs in San Francisco, you'll find that there's an extremely ridiculous and even disturbing “data leap”.
Recently, a man who has immersed for months in the Technological Circle of San Francisco broke the “rules of the game” of the Silicon Valley AI Circle in the most straightforward manner on social networking:
Laboratory orders valued at $3 million were received within two weeks, multiplied by 26, with external asserted annual revenues (ARRs) of approximately $80 million. All orders for the entire year are scheduled to be signed within the same month, then the best-performing month is multiplied directly by 12 for the year. A letter of intent (LOI) of $5 million was signed with friends' start-up companies and college students. You sign one, I sign one. This is also common in large modelling laboratories and computational centres. To buy the GitHub sign (Star) on the black market at a price of six cents apiece.
“The last company, which had exchanged invoices with partners and had collected $220 million, is now bankrupt. Sadly, it has become a game of the Gresin Code, and if you do not take part in the forgery, you will be eliminated.”
This phrase, like a clear cut, reveals the most prominent problem of the 2026 AI Circle. A top-down “data-packaging campaign” is emerging as an undeclared survival rule in Silicon Valley, when the vigil of technical indicators does not mask commercially landed anaemia.
Bad ARR: Two-week order with 26 magic
In the era of SaaS (software, i.e. services), ARR (annual recurrent income) is a core indicator of the health of a company. But in today's AZLLM initial circle, ARR has become a rubber mud that can be squeezed at will.
“Takes an order of $3 million within two weeks, multiplied by 26 to 80 million ARR.” This logic is totally commercial. A short-term, one-time test order has been packaged into long-term stable subscription income. This was like the first week of the opening of a restaurant, when the owner immediately announced a 50-fold increase in net profits throughout the year.
Why would the founder do that? Because Wall Street and the VCs only watch this. And in the top-winding model, if your ARR can't double in a quarter, you don't even qualify for a partner. In the extreme anxiety of FOMO, investors are searching for the next billion unicorns with their calculators, while entrepreneurs are struggling to find a “digital puzzle” that fits the formula.
As a result, one year's income is recognized in one month (the spiral of growth that creates terror) or two months of explosive income is annualized as an undeclared “financial makeup”. Such data, “enhanced by seedlings”, not only distorts the true valuation of the company, but also makes the industry-wide baseline extremely disproportionate. When everyone is bragging at the speed of 26, companies that record their income on the basis of real delivery look like snails that are about to go bankrupt.
Spelling and Intent Magic: a financial alchemy
The third point mentioned in the tweets is the classic Round-Tripping.
Starter A and Starter B are friends, and A buys $5 million in computing or software services from B and B buys equivalent services from A. The parties signed contracts without any substantive background, even without the need for real financial flows. Then two companies took these contracts and told the investors, "Look, how well we're commercialized, and the big customers are buying!"
This is not only a packaging technique for VCs, but also a feeder of financial leverage.
The tweet notes, inter alia, that this is also prevalent in large model laboratories and the Neocloud circle. In order to obtain high-value loans from banks, many computing leasing companies enter into long-term framework agreements (LOI) with model companies for daily prices. Model companies have obtained paper-based security for their ability to take these contracts to collateralize and expand debt leverage.
In this chain, the order is no longer intended to serve the customer, but rather to produce a certificate that can be taken as collateral. This explains why a large calculator such as CoreWeave, while seemingly When the chain broke, the swapped invoices turned into waste paper. That company, which had used to blow $220 million in revenue and eventually went bankrupt, was the most ruthless mockery of that false prosperity.
Six cents of a reputation: open source indicator for priced codes.
If the financial packaging was intended to attract capital, the GitHub sign was bought to influence the technical community ' s judgement.
The number of Stars in GitHub is an important measure of the popularity and code quality of an open source project. But in this man's story, six cents can buy a sign.
This industrial chain is extremely mature on an ash-producing platform. The start-up company spends hundreds of dollars on a new, useless warehouse with tens of thousands of Stars overnight. This, in the view of investors with some non-technical backgrounds, represents “a fanatical hobbies of the polar circle”.
Moreover, daily users (DAU), API, and even models, can create false prosperity in a short period of time by employing the Saber Water Force and brushing scripts. This is the most radical departure from the overall software engineering spirit. When the badge representing the honour of the extremists becomes a well-defined commodity priced on the waterline, the reputation of the bottom technology is completely crushed on the ground.
The consequences of the Grething Law: Who is driving bad money away from money?
The last and most lamentable of the entrepreneur's “Greithin Law” (the bad currency to drive away the good currency) is the saddest and truest of the whole tweet.
In this theatre, the front row stands up and the back row must step on the chair. If you don't take the bill, you don't blow the ARR, you don't buy the sign, your valuation is lower than anyone else's, you won't get the most expensive genius, and you won't get the GPU of Inverda.
At the very root, LLM was an extremely cruel war on heavy assets. Expensive bills and the best payers make start-ups have to work hard to finance. In order to obtain financing, they can only accommodate the strong preference of the VC for “super-high growth”.
But behind that, is VC really an innocent victim? That's not true. To some extent, VC is also involved in this data packaging cycle. There is also pressure from KPI within many investment institutions, and fund managers need to quickly invest in star projects and raise the book returns in order to raise funds from the LP (financing) fund in the next period. As a result, entrepreneurs used packaged data to obtain high valuations, the VC used a high valuation project to show LP returns and LP looked forward to the next round. It's completely a charade.
Conclusion: returning to common sense, refusing to pay for foam
This “digital leap” that the San Francisco technological circle is going through is like the precipice of the 2000 Internet bubble.
The original company at the time, with the addition of a “.com” to its name, doubled its share price; now AI, with the two-week order multiplied by 26, could call itself the next Open AI.
For ordinary entrepreneurs and practitioners, the more prudent option is not to participate in the cycle and not to become a co-participator of false data.
A truly great company, with its core financial model, is always simple: real revenue, less less than the cost, leaves real profits. Both Open AI and DeepSeek, after removing all the fancy packagings, continue to spell algorithmic efficiency and real commercial liquidity.
When the foam breaks, you find that the most reliable moat is never a six-cent sign, not a contract, but a client willing to pay for your real product in a non-foam-free winter.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.