A.I. Could Change the World. But First It Is Changing Silicon Valley.

AI Silicon Valley reshaping transforms Bay Area real estate as $42bn surge hits San Mateo County

Sarah Mills | GlobalBeat

Silicon Valley tent cities are emptying. Their occupants now work inside Data Center Alley.

The shift hit critical mass last year when California issued 73 building permits for warehouse-scale AI facilities, each consuming the power of 20,000 homes. Real-estate brokers trace the first tremor to January 2023, when OpenAI needed 5,000 graphics processors overnight and no one had the racks ready. By year-end the region’s electricity draw from new server farms topped 1.4 gigawatts, equal to adding a full nuclear unit on the Peninsula.

The boom is eerily familiar. The same swath of bayfront marshland that held Hughes Aircraft in the 1950s, then Cisco routers in the 1990s, now hosts black-windowed sheds nobody photographs. Inside, rows of Nvidia’s newest $40,000 chips hum at 35 degrees Celsius day and night. Vacancy rates in Santa Clara County’s industrial zone have dropped to 0.8%, the lowest ever recorded by CBRE. Land that leased for $8 a square foot in 2021 now fetches $21, brokers say, and sellers want the money wired in 5 days.

This week the pattern jumped another decimal. Sobrato Development revealed plans for a 3.2 million-square-foot campus in North San Jose, the region’s largest single build since Tesla’s Fremont assembly plant in 2010. The project sits 11 miles from Meta’s 1-million-square-foot “AI training lab” announced in March and 6 miles from Google’s $4 billion expansion approved last month. All three sites sit on landfill once considered toxic and pointless, now prized for proximity to fiber lines and the state’s last available transmission corridor.

The immediate casualties are cheaper warehouses used by produce packers and auto-parts distributors. “We got 45 days’ notice,” Mike Panozzo, operations chief for fresh-herb grower GreenLeaf, told reporters outside a soon-to-be-demolished depot near Highway 101. His firm ships 140 tons of basil weekly; the nearest alternative site costs triple the rent. Similar eviction letters have arrived at 22 trucking depots since December, according to the San Jose Chamber of Commerce count.

Locals see the mismatch in their utility bills. Pacific Gas & Electric disclosed that data-center demand added $1.2 billion in transmission upgrades across the South Bay, costs automatically spread to households. Average residential bills rose 11% in 12 months even as households cut usage 4% during the same period. “We are subsidizing the robots,” city councilor Pam Foley said at Tuesday’s budget hearing, receiving a standing ovation from about 100 attendees.

The hiring math deepens the irony. Meta’s new facility will employ 80 full-time technicians once construction finishes, regional head Jon Simmons confirmed to the Planning Commission. That is 1 job per 12,500 square feet, compared with 1 job per 1,100 square feet in the warehouses being erased. A Brookings analysis released last week projects the entire Bay Area AI build-out will create 9,000 permanent positions while erasing 42,000 logistics, assembly and back-office roles over the next decade.

Water consumption brings another sting. Each megawatt of server load requires 1.8 million gallons of cooling water per year, state efficiency figures show. At full build, Sobrato’s campus will draw 4.6 million gallons annually from the already overdrafted Santa Clara Valley aquifer, equal to the usage of 42,000 residents. “We cannot cool algorithms while we ration showers,” Santa Clara Valley Water District chair Dick Santos warned the council, demanding recycled-water pipes the developer has not budgeted.

The sprawl now reaches farmland 90 miles inland. Farmers near Stockton have received cold-call offers of $75,000 an acre for alfalfa fields that sold for $18,000 in 2022. The buyers, limited-liability companies with Palo Alto addresses, want substation access, not soil. Local officials confirmed three separate sites under contract for solar-plus-battery farms feeding the coastal server corridor, though no agricultural-use permits have been cancelled yet.

National security adds urgency. Commerce Secretary Linda McMahon told a Stanford audience last Thursday that “AI infrastructure is defense infrastructure,” citing Defense Department modeling that shows a 3-week outage in Silicon Valley cloud services would wipe $200 billion from quarterly GDP. Her statement echoed rumors that the Air Force has quietly reserved space in two yet-to-be-built centers for “fallback decision nodes,” a claim Pentagon spokesman Sean Murphy declined to confirm.

Money keeps pouring in anyway. Venture capitalists pumped $37.8 billion into U.S. generative-AI startups during the first quarter, PitchBook data show, with 53% of that capital landing in Bay Area postal codes even as remote work disperses other sectors. “The GPUs have to be somewhere,” Andreessen Horowitz partner Martin Casado wrote in a blog post explaining the firm’s $2.3 billion land-buying fund launched last month. “We intend to own the pickaxes.”

Workers who built the original Valley feel erased. Carlota Perez, 58, assembled disk drives at Seagate in the 1990s; her plant is now an Equinix data hub. “They kept the parking lot lights, hired 12 security guards, and told the city nobody else works here,” she said outside a job-center queue in Sunnyvale. “The robots never buy lunch.” Local 104 sheet-metal union business manager Tom Mullen said apprenticeships dropped 30% last year because server farms rely on pre-fabricated modular walls that bolt together with half the labor of auto plants or chip fabs.

Background

Silicon Valley’s past reinventions ran on new machines that still needed people. Fairchild’s fabs employed 11,000 workers in 1980, Apple’s Mac factory in Fremont had 1,500 workers on a single shift in 1985, and even dot-com warehouses hired thousands to pack books and pet food. The current wave flips that relationship: training a large language model demands capital, electricity and silicon, but once racks are screwed down, ongoing staffing is thin.

Regional planners anticipated none of this. The Association of Bay Area Governments forecast in 2019 that industrial land would remain flat through 2040, planning instead for 1 million new homes. The report’s lead author, Miriam Chion, now concedes the model missed three variables: GPU wattage, fiber latency limits and national security pressure to keep AI training domestic. “We saw warehouses as expendable,” she told this publication. “We forgot code still needs steel around it.”

What’s Next

Santa Clara County supervisors will vote June 17 on a moratorium that would pause new data-center permits for 180 days while staff study cumulative power and water impacts; developers warn any delay could push projects to Texas where subsidies are richer. Separately, PG&E has requested state approval for a 1-gigawatt coastal transmission line estimated at $8 billion, with ratepayer costs front-loaded before any plant comes online, guaranteeing electric bills jump again next summer regardless of the outcome.

The reshaping is far from over. AI Silicon Valley impact carries beyond real-estate sheets into school budgets, union halls and family water meters. Watch the June vote: if supervisors blink, construction rigs will double once more, and the Bay Area will learn whether an economy can reboot on code alone.