AI Is Leasing Offices and Cutting the Jobs That Fill Them

CBRE released its annual Scoring Tech Talent report on August 18, 2026. The headline finding is that the number of AI-skilled tech talent workers across the U.S. and Canada increased by 45 percent year over year to 751,000 as of mid-2026. In the U.S., CBRE reports, AI-related roles accounted for 31 percent of available tech talent jobs as of June 2026, up from 11 percent when overall tech postings last peaked in mid-2022. In the San Francisco Bay Area, AI’s share rose to 57 percent from 20 percent over the same periods. The fastest growing AI roles over the year were data scientist, which added 29,000 jobs, and computer and information systems manager, which added 24,600. CBRE says the Bay Area and New York each added more than 20,000 AI jobs since mid-2025 through both new jobs and the conversion of existing jobs to AI-skilled roles, which is a distinction the rest of this brief turns on.

The report sets that against a much flatter base. Total U.S. tech talent employment grew by 1.8 percent, or 108,760 jobs, in 2025, and total Canadian tech talent employment grew by 7.6 percent, or 91,300 jobs. CBRE counts 6.3 million tech talent workers in the U.S. and 1.3 million in Canada, 4.0 percent and 7.0 percent of each country’s total workforce in 2025. In the same report CBRE records, citing Challenger, Gray & Christmas as of June 2026, that job cuts employers attributed directly to AI rose to 101,743, or 22.9 percent of all cuts, so far this year, up from 54,836, or 4.5 percent, for all of last year. Challenger’s own July report, published after CBRE’s cutoff, updates the figure: through July, AI has been cited in 112,713 job cut announcements, roughly 24 percent of all cuts, the fifth consecutive month AI led all stated reasons. The two records do not conflict. They are the same series read one month apart, and the later one is higher.

The market table moved too. CBRE reports that New York became the largest tech talent market in 2025, surpassing the San Francisco Bay Area: the New York Metro workforce grew by 30,640 to 394,300 between 2022 and 2025, while the Bay Area contracted by 23,900 to 375,730. The Bay Area nonetheless still ranks first in CBRE’s indexed rankings, ahead of Seattle, Toronto, New York Metro, Austin and Washington, D.C., because the index weights 13 metrics of depth, vitality and attractiveness rather than headcount alone. CBRE also prices the seat: the total annual labor and real estate cost for a typical 500-person tech company occupying 60,000 square feet ranged from $36 million in Quebec City to $91 million in the San Francisco Bay Area.

Two pieces of arithmetic are worth doing on CBRE’s own figures, because they bound what the growth headline can mean for space. First, 751,000 divided by 1.45 computes to about 518,000 a year earlier, so the 45 percent growth represents roughly 233,000 AI-skilled workers added across both countries. Second, CBRE’s own 500-person, 60,000 square foot benchmark computes to 120 square feet per employee. Applied to 233,000 workers, that computes to roughly 28 million square feet, but only if every one of those workers were a net new hire taking a net new seat in space that did not previously exist. CBRE says plainly that the growth came through both new jobs and the conversion of existing jobs to AI-skilled roles, and a reclassified employee occupies the same desk they occupied last year. That is PSV arithmetic on CBRE’s published figures, not a CBRE projection, and its purpose is to establish a ceiling rather than a forecast. On the leasing side CBRE has published measured figures separately: in a May 12, 2026 release the firm reported that tech accounted for 16.8 percent of all U.S. office leasing in 2025 and 22.7 percent in the first quarter of 2026, and that AI companies have leased 21 million square feet in San Francisco and Silicon Valley and 9.4 million square feet in Manhattan, Boston and Seattle since 2019.

The gateway office recovery has a narrow base. Tech at 22.7 percent of first-quarter U.S. office leasing is a real demand signal, and CBRE Tech Insights Center executive director Colin Yasukochi has said that AI investment “has moved from research and experimentation to large-scale deployment” and that CBRE expects more U.S. office markets to benefit as the tech growth cycle expands. That is CBRE’s forward view and it may well be right. What the Scoring Tech Talent numbers add is the shape of the cohort behind it. AI is 31 percent of open U.S. tech roles and 57 percent of them in the Bay Area, while the tech talent occupation base as a whole grew 1.8 percent last year. Demand is concentrated in a fast-growing slice of a slow-growing category, in a handful of markets, funded by a capital cycle CBRE itself measures at $578 billion of U.S. AI venture capital since 2020 with roughly 80 percent of it going to the Bay Area. Concentrated demand is still demand. It is just a different risk profile than broad-based absorption, and it should be underwritten as one.

The second fact is the one that does not usually make the leasing memo. The same technology driving the leasing is now the leading stated reason for layoffs in the country, five months running, and the sector doing the leasing is also the sector doing the cutting. Challenger reports that technology led all sectors in July 2026 with 9,867 announced cuts, for 149,023 in the year to date. Set that against CBRE’s 108,760 net U.S. tech talent jobs added in all of 2025 and the contrast is stark, but the two are not the same measure and must not be netted: Challenger counts announced cuts by employer sector across the whole economy, CBRE counts employment in roughly twenty tech occupations across all industries, and announced cuts are announcements rather than separations. The honest reading is directional, not arithmetic. A tenant can be hiring AI-skilled staff and shrinking total headcount in the same quarter, and several are. For a landlord that means the tenant’s AI story and the tenant’s space requirement can move in opposite directions at renewal.

There is also a rank-versus-direction trap in the market table. The Bay Area holds the top spot in CBRE’s index while its tech talent workforce contracted by 23,900 between 2022 and 2025, and New York Metro, which added 30,640 over the same window and now has the larger workforce, sits fourth. An index built on depth, vitality and attractiveness is measuring accumulated stock and pipeline, not the direction of the last three years. Either metric can be the right one to underwrite against. Using the ranking as a proxy for momentum, or the headcount shift as a proxy for quality of demand, is where the mistake gets made. Broader context is worth carrying too: Challenger reports 477,033 total announced cuts through July 2026, down 41 percent from 806,383 in the first seven months of 2025, and July hiring plans of 16,095, up 47 percent from June. The labor market is not falling apart. It is reallocating, and office demand follows the reallocation rather than the aggregate.

The job is to convert a national report into a tenant-level exposure view, and it is a research and reconciliation task, which is what these tools are actually good at. Inputs: the rent roll for the office portfolio with tenant legal names, square footage, expiration dates and options; the CBRE Scoring Tech Talent 2026 market tables for every market the portfolio touches; the Challenger monthly job cut reports for the year to date; and each tech tenant’s own public record, meaning press releases, 10-K and 10-Q risk factors and headcount disclosures, and posted job listings. Output: one table, one row per tech tenant, carrying square footage, expiration, the tenant’s stated AI hiring posture with a citation, any announced reduction with a citation and a date, the market’s CBRE AI posting share, and a flag for tenants where the two signals disagree.

The gate matters more than the model. Every cell in that table carries a source link and a date, and any cell the model cannot source stays empty rather than being filled with an inference. Arithmetic is done in the spreadsheet, not by the model, because a model asked to compute rollover exposure across 40 leases will produce a plausible number and no audit trail. The asset management lead reviews every flagged row before the table goes anywhere, and the head of asset management signs the version that reaches the investment committee. Refresh it monthly when Challenger publishes and annually when CBRE does. PSV has not run this specific build for a client and is not claiming a result from it. It is the shape the work takes, and the value in it is that the exposure question gets asked with citations attached instead of being answered from a leasing anecdote.