A meme titled "The SF Social Contract" made the rounds on X last week, turning San Francisco’s AI startup economy into a blunt flowchart about where venture money actually goes. The image was created by the X account @chiefofautism. BlockTempo took the meme’s numbers and labels and added its own interpretation around them.
The article’s core claim is simple: venture firms fund startups, startups send large chunks of that money to model vendors, API providers, cloud platforms, and GPU infrastructure, and the rest leaks into Bay Area rent, high-cost lifestyles, and acqui-hire deals by large tech companies. The meme is framed as satire, but the write-up argues that the chain it sketches is not far removed from reality.
First stop: compute, models, and API burn
BlockTempo uses a fictional 29-year-old San Francisco engineer named Arjun as the stand-in for a Bay Area AI founder. By day he codes with Cursor. By night he watches AWS and GPU bills climb. On weekends, he explains in coffee shops that his startup is not just another AI wrapper.
In the meme, once Arjun gets funded, the first arrows point to OpenAI, Anthropic, AWS, and GPU cloud providers, labeled "model & API burn." The article says API and inference costs consume 40% to 50% of revenue for AI startups. Put next to the funding picture, that creates the awkward loop at the center of the piece: venture capital goes into startups, and startups send a large share of operating income straight back to the companies that dominate foundation models and compute access.
BlockTempo says U.S. venture investors deployed $412.7 billion in the first half of 2026, a record for the period, with 86% going to AI. OpenAI and Anthropic alone accounted for $217 billion, according to the article.
The loop does not end there. The article says Anthropic’s annualized compute cost around mid-2026 was roughly $4 billion to $4.5 billion, equal to about 60% of annualized revenue. OpenAI’s ratio was described as above 75%. Even the companies selling model access and infrastructure are still burdened by the cost of compute.
Landlords as the next layer of beneficiaries
The second layer in the meme is the landlord class. BlockTempo describes them through a fictional couple, Bob and Susan, both 67, who bought a Victorian row house decades ago for what would now look like a down payment. They charge $6,800 a month per room, spend time skiing in Tahoe or wine tasting in Napa, and text tenants from a vacation home in Palm Springs to remind them to move the trash bins.
The article ties that caricature to California’s Proposition 13, which took effect in 1978. As described in the piece, the law caps property tax at 1% of assessed value, limits annual assessment increases to 2%, and triggers reassessment only when ownership changes or new construction occurs. The result, according to the article, is that homeowners who bought in the 1980s often face an effective tax rate of just 0.2% to 0.4% of market value. On the same street, with the same layout, a longtime owner can pay only a tenth of what a recent buyer pays in property taxes.
BlockTempo sums that up in one line: landlords are collecting 2026 rent on a 1980s tax base.
For current rent levels, the article cites Zumper’s latest report. San Francisco’s median rent for a one-bedroom apartment is listed at $4,180 a month, up 22.9% year over year, second in the U.S. behind New York at $4,560. The median for a two-bedroom apartment reached $6,000 in July 2026.
The meme’s next character: the Bay Area Asian girlfriend stereotype
A second image posted 12 minutes later breaks one path into six panels. Its main character is Mei, 27, presented as someone raised on a familiar high-achievement track associated in the article with Asian immigrant families: piano lessons, Kumon, SAT preparation, then Berkeley or UCLA, followed by a tech career in product, design, or AI research.
The portrayal leans heavily on stereotype, but BlockTempo uses it to talk about a specific Bay Area professional lifestyle: a $9 matcha latte in hand, Alo workout wear on the way to Pilates, evenings spent at AI events in Hayes Valley, part networking, part job search, part dating market.
The meme gives Mei total compensation of $220,000, plus possible equity. The article says that for mid-level AI product managers with three to seven years of experience in the Bay Area, total compensation generally falls between $150,000 and $220,000. At the top of that range, she may still live in an older apartment without in-unit laundry or parking.
The point here is narrower than a labor-market thesis. In the article’s framework, a high salary does not cancel out the pressure created by San Francisco rent and day-to-day costs. It becomes one more station in the same money loop.
The end of the loop: startup exits through acqui-hire structures
In the meme, one of Arjun’s three life goals is to get acquired by a major tech company. Another is to get into Y Combinator and raise $3 million. A third is to jump to OpenAI before his own startup runs out of money. BlockTempo focuses on the path labeled "acqui-hire," with arrows pointing to Google, Meta, and Stripe.
The article says that from March 2024 to January 2026, Google, Microsoft, Amazon, and Meta spent more than $20 billion combined to bring in AI startup founding teams. Technically, it says, they did not acquire the companies themselves.
BlockTempo lists several examples. Google used roughly $2.4 billion in licensing fees to bring in Windsurf co-founder Varun Mohan and the startup’s core team, without buying the company or taking equity. Meta invested $14 billion in Scale AI while also hiring away co-founder Alexander Wang. Microsoft’s arrangement with Inflection involved licensing the model and bringing over about 70 people, including Mustafa Suleyman.
The article calls this structure a form of reverse merger hiring built to avoid antitrust scrutiny. It says the U.S. Department of Justice antitrust division has publicly described such deals as a red flag. In these arrangements, the people who receive the main economic upside are founders and the most senior research staff, while the rest return to large-company employment.
Why the joke lands
By the article’s telling, the whole chain works like this: venture firms fund engineers, engineers build startups, startups pay GPU providers, landlords, and girlfriends, landlords and girlfriends spend money in Napa wineries, and everyone collectively feeds LinkedIn.
Arjun feels the money went to GPUs and rent. Mei feels her salary disappears too elegantly. Bob and Susan collect rent and drink wine in Napa. OpenAI and Anthropic, despite capturing more than 40% of global startup funding according to the article, are still seeing more than 60% of revenue eaten by compute in their own stack.
That is the closing note of the piece. Every participant in this chain believes they are the one who did not really make out ahead.
Two background references at the end of the article
BlockTempo also mentions a recent public remark from OpenAI CEO Sam Altman that two years of college is enough and staying until graduation is a waste of time. At the same time, the article says many Bay Area Asian engineers are still following the familiar parental upgrade path of piano, Kumon, SAT, and STEM preparation, with the long-term goal of elite schools and jobs in venture capital or major tech.
Those details are not developed into a separate argument. They remain part of the same San Francisco AI narrative the article is sketching: money, education, compute costs, rent structures, and talent flows folding back into one another inside a single ecosystem.