San Francisco AI venture capital data packaging: buying GitHub stars for 6 cents, annualizing two-week orders into full-year ARR

In Silicon Valley's startup legend, we are used to hearing stories about changing the world, faith in technology, and geek spirit. But if you truly peel back this glossy exterior and go deep into the circle of frontline founders in San Francisco (SF), you will find that an extremely absurd and even disturbing "data Great Leap Forward" is unfolding here.

Land a $3 million lab order within two weeks, multiply by 26, and publicly claim annual recurring revenue (ARR) of about $80 million. Schedule all of the year's orders to be signed within the same month, then take this best-performing month and directly multiply it by 12 as the full-year figure. Sign $5 million letters of intent (LOI) with a friend's startup and college classmates. You sign one, I sign one. This is also common in the circles of large model labs and compute centers. Spend 6 cents each to buy GitHub stars (Star) on the black market.

This industry insider let out a heavy sigh: "The previous company that swapped invoices with partners and inflated revenue to $220 million has now gone bankrupt. Sadly, it has now become entirely a game of Gresham's law (bad money drives out good); if you do not participate in fraud, you will be eliminated."

These words are like a clear incision, revealing the most prominent problem in the 2026 AI startup circle. When the carnival of technical metrics cannot conceal the anemia of commercial implementation, a top-down "data packaging movement" is becoming Silicon Valley's tacit survival rule.

Ruinously played ARR: the "magic" of two-week orders multiplied by 26

In the SaaS (Software as a Service) era, ARR (annual recurring revenue) was the core metric for measuring a company's health. But in today's AI large-model startup circle, ARR has become a piece of clay that can be kneaded at will.

"Land $3 million in orders within two weeks; multiply by 26 and it is $80 million ARR." This logic is utterly absurd commercially. A short-term, one-off test order is forcibly packaged as long-term stable subscription revenue. This is like a restaurant doing booming business in its first week of opening, and the owner immediately announcing that full-year net profit will increase 50-fold.

Why do founders do this? Because Wall Street and VCs only look at this. In top venture capital models, if your ARR cannot double within a quarter, you do not even qualify to meet partners. Under extremely anxious FOMO (fear of missing out), investors hold calculators looking for the next hundred-billion-dollar unicorn, while entrepreneurs desperately look for "number puzzle pieces" that fit the formula.

So, compressing a year's revenue into one month for recognition (creating a terrifying quarter-over-quarter growth curve), or annualizing two months of explosive revenue, has become everyone's tacit "financial makeup art." This "pulling up seedlings to help them grow" data not only distorts the company's true valuation, but also makes the entire industry's baseline extremely absurd. When everyone is boasting at 26x speed, companies that honestly record revenue based on actual deliveries look like snails about to go bankrupt.

Mutual order padding and LOI magic: a financial alchemy

The third point mentioned in the tweet is the extremely classic "round-tripping."

Startup A and startup B are founded by friends. A buys $5 million of compute or software services from B, and B simultaneously buys the same amount of services from A. Both sides sign contracts with no substantive transaction background, and real capital flows may not even be needed. Then the two companies take these contracts to tell investors: "Look how strong our commercialization capability is; major customers are rushing to pay!"

This is not only a packaging tactic aimed at VCs, but also a fulcrum for financial leverage.

The tweet specifically points out that this is also common in the circles of large model labs and compute centers (Neocloud). Many compute leasing companies, in order to obtain large loans from banks, will sign sky-high long-term framework agreements (LOI) with model companies. Model companies obtain paper compute guarantees, and compute companies use these contracts to pledge financing and expand debt leverage.

In this chain, orders are no longer for serving customers, but for generating a certificate that can be used as collateral. This explains why compute giants like CoreWeave, although their revenue and profit appear astonishing, carry heavy debts of tens of billions of dollars. When the chain breaks, the invoices everyone swapped become a pile of waste paper. That company that once inflated revenue to $220 million this way and ultimately went bankrupt and liquidated is the most ruthless mockery of this false prosperity.

A reputation at 6 cents each: open-source metrics with explicit price tags

If financial packaging is to attract capital, then buying GitHub stars is to influence judgments in the tech circle.

The number of GitHub Stars was originally an important metric for measuring an open-source project's popularity among developers and code quality. But according to this person's disclosure, 6 cents can buy one star.

On gray-market platforms, this industrial chain is extremely mature. A startup can spend a few hundred dollars to give a repository that has just gone online and has no practical value tens of thousands of Stars overnight. In the eyes of some investors without technical backgrounds, this represents "feverish pursuit in the geek circle."

Not only that, daily active users (DAU), API call volume, and even model rankings on various benchmark leaderboards can all be made to create false prosperity in a short time by hiring "cyber water armies" and using volume-boosting scripts. This is the most thorough deviation from the spirit of software engineering. When badges representing geek honor become explicitly priced commodities on an assembly line, the reputation of underlying technology is thoroughly rubbed into the ground.

The consequence of Gresham's law: who is driving bad money to drive out good?

The "Gresham's law" (bad money drives out good) that this entrepreneur finally lamented is the saddest and also the most real stroke in the entire tweet.

In this theater, when the people in the front row stand up, the people in the back row must step on chairs. If you do not pad orders, do not inflate ARR, and do not buy stars, your valuation will be lower than others', you will not be able to recruit the most expensive geniuses, and you will not be able to buy NVIDIA graphics cards.

Fundamentally, large models are an extremely brutal asset-heavy war of attrition. Expensive compute bills and top talent compensation force startups to raise financing desperately. And to obtain financing, they can only cater to VCs' strong preference for "super-high growth."

But behind this, are VCs really innocent victims? Not so. To some extent, VCs also participate in this cycle of data packaging. Many investment institutions also have internal KPI pressure; fund managers need to quickly invest in star projects and push up paper returns in order to raise the next fund from LPs (limited partners). So entrepreneurs use packaged data to seek high valuations, VCs use high-valuation projects to show returns to LPs, and LPs then count on the next round to take over. This is entirely a game of passing the parcel.

Conclusion: return to common sense, refuse to pay for bubbles

This "data Great Leap Forward" that the San Francisco tech circle is experiencing looks very much like the absurd years on the eve of the bursting of the 2000 internet bubble.

At that time, startups could double their stock price just by adding ".com" to their names; now AI startups can call themselves the next OpenAI just by multiplying two weeks of orders by 26.

For ordinary entrepreneurs and practitioners, the safer choice is: do not participate in this cycle, and do not become a co-participant in falsely reporting data.

The core financial model of truly great companies is always simple: real revenue, minus costs lower than revenue, leaving real profit. Whether OpenAI or DeepSeek, after stripping away all fancy packaging, what they compete on is still algorithm efficiency and real commercial monetization capability.

When the bubble bursts, you will find that the most reliable moat is never stars bought for 6 cents each, nor those contracts exchanged with each other, but the customers willing to pay for your real product in a bubble-free cold winter.