AI Is Tightening Prime Office Space — But Not the Way CBRE's Headline Suggests

When the world's largest commercial real estate brokerage tells you AI is saving the office market, it deserves one extra moment of scrutiny. CBRE published the second installment of its AI-and-real-estate series this week, and the headline claim is exactly what a company that earns fees on office leases wants you to believe: artificial intelligence is a "net tailwind" that is pushing the office market into a K-shaped recovery, where genuinely valuable prime buildings tighten while the rest decay.

The numbers do real work underneath. CBRE forecast the national office vacancy rate to fall from 18.3% today to 14.5% by 2031, and it walks through the mechanism: only 5% of office-using jobs are "highly vulnerable" to displacement, the jobs AI creates are collaborative and in-person, and tech has become the biggest leasing force, taking 22.7% of U.S. leasing in the first quarter of 2026. On its face, that is a bullish enough story that "AI office demand" has become a mantra for the sector.

Here is the part the press release does not emphasize. The claim is "strengthening prime space" — and it means the market is splitting in two. Independent data agrees the split is real, and it is dramatic. CBRE's own tracking puts prime vacancy at around 12.7%, falling about 80 basis points a quarter, while the gap between premium and everything else has widened to the largest on record. Cushman & Wakefield, the competitor that does the same data work, puts national vacancy at roughly 19%. Fewer vacancies at the top, sky-high vacancies everywhere else. That is the K, confirmed by a rival with no stake in CBRE's fees.

But the K is doing more of the work than the word "demand" admits. A big share of the tightening is not AI renting space — it is space being taken out of the market. Roughly 80% of U.S. office markets have seen inventory shrink since 2024 as obsolete B/C buildings are converted to housing or demolished. Prime vacancy falls faster when you literally delete the buildings that would otherwise sit empty. "AI demand" is only one leg of the improvement; the other leg is subtraction.

And on the demand side, the AI leg itself is smaller than the headline implies. Early in the cycle, CBRE counted 21 million square feet leased by AI companies in San Francisco and Silicon Valley since 2019. Striking, until you remember the economics of the tenants doing the renting. AI is a capital-heavy, headcount-light business; GPUs do not need desks. The marquee deals — OpenAI's building, the big Manhattan AI leases — move billions in compute capital for companies that count their employees in the thousands, not hundreds of thousands. Real estate demand scales with people, and AI scales with compute. A handful of trophy tenants in two or three cities is what "AI office demand" actually is, which is why CBRE's own numbers show tech leasing concentrated overwhelmingly in San Francisco and Manhattan, and why roughly 80% of AI venture capital sits in the Bay Area. The concentration is the tell: this is a narrow flywheel, not a broad recovery.

Now bring it back to the stock that carried the headline, because that is what "AI strengthens prime space" is supposed to be bullish for.

CBRE does not collect rent. It collects fees. It is a broker and adviser: it earns on leasing transactions, property sales, appraisals, and facilities management. It is the middleman of the office market, not a landlord betting on the buildings themselves. So the model's answer to "AI rescues offices" reaches CBRE only through the slice that turns into transactions — and the transactions it earns on are disproportionately the prime ones it just celebrated. The billions of square feet of obsolete second-tier space sitting empty across the country generates no fee for anyone.

That is the lens for reading the actual earnings. CBRE's numbers are genuinely good — and that is the uncomfortable part. In the second quarter of 2026, revenue rose 16% to $11.2 billion, core EPS of $1.56 beat consensus, and the company raised its full-year core EPS guidance to a range of $7.80–$7.90, growth of about 23%. The AI-office narrative is not hurting the business; the fundamentals improved while the story got written.

Which leaves the stock itself doing something odd. CBRE shares are down roughly 14% for the year even as the numbers got better. The company trades at about 30 times trailing earnings, against roughly 16 times for close competitor JLL; that premium has been there for years, justified by scale and a higher-margin mix. A down year on top of that premium is the market declining to pay up again for a story it has already heard. Whatever the AI tailwind adds to fee volumes, it is not enough to make a mid-30s multiple look cheap — and the sector's exposure to interest rates and to year after year of high secondary vacancy keeps a lid on what investors will spend for the growth.

The honest read does not need to be forced into false neutrality. The underlying claim has real support: prime space is genuinely tightening, the K-shape is confirmed by the competitor's own data, and both supply shrinkage and a genuine if narrow AI leasing wave are real. What is marketing is the scope. "AI strengthening prime space" sounds like a tide lifting the whole sector, when it is really a modest, concentrated pool of money chasing a shrinking stock of premium buildings.

For a shareholder, none of this is the doom the bears want or the boom the press release sells. It is a genuine tailwind for CBRE's transaction-fee business wrapped in a valuation that already assumes it. The last question an investor faces is not whether AI demand is real — it clearly is — but whether a roughly 30-times-earnings fee business has the operating leverage to make that premium feel earned, year after year, in an office market that is only recovering in its most expensive corners.