San Francisco's Housing Rights Committee fields 200 calls a week on its rental assistance hotline. Its clinic is at capacity. More than 150 voicemails pile up every weekend. This is the downstream reality of the AI boom — not the gleaming offices of Anthropic and OpenAI in SoMa and Mission Bay, but the grinding displacement of the people who were already here. The numbers are stark. A one-bedroom now averages $4,400 a month, up more than 25% from last year and nearly three times the national average. Two-bedrooms run $6,000. A centrally located four-bedroom for a family easily exceeds $10,000. Even six-figure earners struggle. For everyone else — service workers, immigrants, the elderly — the math is existential. Frederick Jones, a 47-year-old Ghirardelli factory worker, spends half his income on the smallest room in a three-bedroom house after losing his apartment to a condo conversion. The extraction mechanism is precise. Roughly two-thirds of San Francisco's residents rent, and the city has rent control on more than 160,000 of its 250,000-plus units. But landlords are exploiting the Ellis Act to evict entire buildings, sell vacant properties at premiums exceeding $1 million over asking to AI-industry buyers, and convert rent-controlled apartments into luxury condos. Eviction notices are up 44% citywide year over year, per the San Francisco rent board. On Powell Street in North Beach, a mile from Anthropic's headquarters, new owners of a seven-unit building occupied by decades-long tenants are attempting to merge units into four upscale condos. The tactics go beyond legal loopholes. Landlords are engineering pretextual evictions — switching rent collection methods without adequate notice, then issuing nonpayment notices when confused tenants pay the old way. A Tenderloin building saw mass eviction notices after its new owners failed to inform residents of a payment change. Others deploy "renovictions" — unnecessary, loud remodels designed to drive tenants out — or simply refuse to address infestations, broken appliances, and trash. The Housing Rights Committee reports a 23% increase in landlord disputes related to repairs. In Chinatown, an investor purchased a rent-controlled building of mostly older Cantonese-speaking residents and immediately began issuing eviction notices, claiming long-term tenants were hoarders. Under the federal immigration crackdown, some landlords are weaponizing deportation fears as leverage. The human cost is medical-grade. UCSF's Margot Kushel describes patients delaying preventive care, turning to stimulants to manage multiple jobs, stopping medications because they cannot afford a $10 co-pay, and starving themselves. One elderly patient had not eaten in three days, unable to pay bus fare to appointments because every dollar went to rent. Displaced families double up — four or five people to a bedroom, ten workers in a studio — with cascading effects on mental health, children's education, and physical safety. Mayor Daniel Lurie declared a "rent emergency" in September, proposing funding for eviction legal aid, larger relocation payments, and caps on rent increases for newly vacant rent-controlled units. Housing advocates are unimpressed. "This is San Francisco on repeat," said Quintin Mecke of the SF Council of Community Housing Organizations. "We treat housing like a commodity rather than a home that people have a right to, and we don't learn any lessons." The crisis is also hollowing out the support infrastructure itself: interfaith agencies offering services to displaced residents are closing because their own landlords are raising rents or refusing lease renewals. The structural pattern is not new, but the scale of AI capital intensifies it. The generative wealth created by the AI industry stays overwhelmingly within that industry — high salaries, IPO windfalls, inflated property values — while the costs are externalized onto the city's most vulnerable residents through a housing market that functions as a one-way value extraction engine. Until the supply side of this equation changes fundamentally, each boom will replay the same script with higher stakes.
