Class A Office Space in Midtown Manhattan and Hudson Yards (2026): Where Companies Actually Lease and Why AI Ignores Most Buildings

The shift: how companies actually pick Manhattan office space in 2026

Ten years ago, a growing company looking to lease Class A office space in Midtown Manhattan followed a predictable process: hire a tenant rep broker, call CBRE or JLL, tour buildings, negotiate. The information asymmetry was deliberately baked into the system — landlords and brokers held market intelligence, and tenants paid to access it through the transaction.

That process is changing. Before the first broker call, decision-makers now do something different: they ask an AI.

“Which buildings should we look at in Midtown or Hudson Yards?” “What is the difference between One Vanderbilt and 50 Hudson Yards?” “What are trophy office rents running in Manhattan right now?” These are queries that corporate real estate teams, CFOs, and founders are directing at ChatGPT, Perplexity, and Claude before they pick up the phone. The AI’s answer shapes the shortlist — and buildings that do not appear in that answer are often never considered.

The flight to quality that drove the 2024–2025 leasing surge has continued into 2026 with new intensity. Manhattan tenants signed leases for approximately 10.4 to 11.8 million square feet in Q1 2026, the strongest first quarter in six years, according to research from JLL and Colliers (April 2026). Midtown captured 57% of that activity, with average Class A asking rents in Midtown reaching $84.74 per square foot, up 3% from Q4 2024 (Cushman & Wakefield Q4 2025 MarketBeat).

The most visible driver of 2026 leasing momentum is demand from artificial intelligence companies themselves. In Q1 2026 alone, AI firms signed leases for 415,000 square feet in Manhattan — half of their entire 2025 total of 845,000 SF — with average deal sizes growing from 16,600 SF per lease in 2025 to 34,500 SF in Q1 2026 (Bisnow / JLL, April 2026). Anthropic was publicly seeking 250,000 to 450,000 square feet for a Manhattan expansion as of January 2026 (Bloomberg, January 2026). Clay, the AI go-to-market platform, announced in April 2026 that it would lease 163,095 square feet at 11 Madison Avenue, adding nearly 500 jobs (Governor Hochul’s office, April 2026). OpenAI had already established a New York presence in Lower Manhattan. Nscale, an Nvidia-backed AI cloud company, set a new Manhattan office rent record of $320 per square foot for roughly 7,200 SF at One Vanderbilt — the first time an AI firm has held the city’s highest rent record (JLL Q1 2026 research via NYREJ).

The trophy tier of the market is now genuinely supply-constrained. Trophy availability dropped 22% year-over-year to 16.9 million square feet, down from 21.2 million square feet a year earlier (Commercial Observer / JLL, April 2026). Central Midtown Class A vacancy sits at approximately 7.5%, tighter than many gateway cities in any sector. This is the backdrop against which AI-driven discovery matters most: when inventory is scarce, first-mover access to information — including AI-generated shortlists — shapes who gets into the best buildings.

Which Class A Manhattan buildings get recommended when an AI is asked where to lease

We queried ChatGPT, Perplexity, Claude, and Gemini across dozens of prompts representing realistic CEO, CFO, and corporate real estate director intent: “best Class A office buildings in Midtown Manhattan,” “where do tech companies lease office space in Hudson Yards,” “trophy office buildings Manhattan with best amenities,” and variations. The same small set of buildings appeared consistently. A much larger set of quality Class A buildings never appeared at all.

The table below shows the buildings that surfaced most frequently, with sourced market data as of Q1 2026. The AI Mention Rate reflects structured testing across four major platforms using standardized leasing-intent queries.

* AI mention rates based on structured testing across ChatGPT, Perplexity, Claude, and Gemini using standardized leasing-intent queries. Rent ranges sourced from JLL, Cushman & Wakefield, The Real Deal, and Commercial Observer Q1 2026 data. Full methodology.

One Vanderbilt dominates AI responses about Class A Midtown leasing to a degree that is out of proportion even with its market position. SL Green has executed a consistent press strategy — every major lease announcement, every occupancy milestone, and now a record-breaking rent deal with Nscale, has been covered by The Real Deal, Bisnow, Commercial Observer, and JLL research reports. That coverage is the direct input to AI training data. The building has earned its AI visibility through the volume and authority of its press corpus.

The Hudson Yards cluster performs well in AI recommendations largely because Related Companies has invested heavily in content about the entire development: neighborhood profiles, sustainability certifications, tenant testimonials, and deal announcements across multiple media outlets. Individual buildings within Hudson Yards benefit from this halo effect. Buildings without a consistent press presence — even those with strong fundamentals and genuine availability — simply do not appear.

Note a specific AI error: 270 Park Avenue (JPMorgan’s new HQ, opened October 2025) is owner-occupied and not available for lease, yet AI systems regularly surface it in responses to leasing queries. This illustrates a key pattern in AI answers about Manhattan office space: AI does not know current availability. It knows press coverage. The two are not the same.

Why most Manhattan office buildings are invisible to AI when a CEO asks about leasing

Manhattan has approximately 500 million square feet of office inventory across roughly 2,000 commercial buildings. Of those, only a handful of trophy and Class A assets appear consistently in AI recommendations. The gap is not about building quality — there are hundreds of Class A properties with genuine availability that AI never mentions. The gap is structural, rooted in how AI systems learn.

1. No structured data on landlord websites

The largest landlords in Manhattan — SL Green, Vornado, Brookfield, Tishman Speyer, Boston Properties — maintain sophisticated marketing websites. But almost none implement the schema markup that helps AI systems understand and extract building information: no RealEstateListing schema, no LocalBusiness schema on individual building pages, no FAQPage answering common tenant questions like “What is the asking rent at [building]?” AI cannot read structured availability data from these sites the same way it reads a well-tagged Wikipedia page or a CoStar data sheet.

2. Landlord sites are JavaScript-heavy

Most Class A landlord property marketing sites are built as JavaScript single-page applications with content rendered client-side. AI crawlers, like many traditional search crawlers, do not reliably execute JavaScript. The practical result: an AI crawler visiting a sophisticated building marketing page sees a nearly empty HTML shell. The availability tables, floor plans, and tenant testimonials that a human visitor sees are invisible to the crawler. This is a systemic failure of the commercial real estate web that no individual landlord has yet solved at scale.

3. Low corpus presence vs. CoStar, LoopNet, JLL, and CBRE

AI training data disproportionately reflects sources with the highest page counts, domain authority, and update frequency. CoStar, LoopNet, JLL, CBRE, and Cushman & Wakefield publish hundreds of market reports, property profiles, and transaction records per year. An individual building page updated twice per year competes poorly for AI “mindshare.” Buildings that are actively profiled in these platforms — with current rent data, availability data, and transaction histories — are far more likely to surface in AI answers than buildings with only a landlord-controlled website.

4. Tenant confidentiality suppresses public deal records

A significant share of Class A deals in Manhattan — particularly for financial services, law firms, and corporate headquarters — are executed under strict confidentiality. Lease terms, rents, and even tenant names are not publicly disclosed. This means the deal generates no training data. A building that has successfully leased 300,000 SF in 12 months under confidential terms has essentially zero AI-visible transaction history, regardless of how strong the building actually is.

5. Submarket granularity is lost in AI training

Manhattan is not a single office market. Grand Central, Park Avenue, Sixth Avenue, the West Side (Penn District / Hudson Yards), and Downtown are meaningfully different submarkets with different rent profiles, tenant mixes, and transport access. AI training data conflates these submarkets constantly — describing “Midtown” as if it were uniform when in fact Central Midtown Class A vacancy of ~7.5% sits alongside Penn District rates that are materially different. Buildings in submarkets that receive less press attention are systematically underrepresented.

6. Building content is not written for information extraction

Marketing copy on landlord websites (“unparalleled views,” “world-class amenities,” “premier location”) contains almost no information that AI can extract as factual claims. AI systems, per the Princeton/Georgia Tech GEO study (Aggarwal et al., 2023), are up to 40% more likely to cite content that contains specific statistical claims. A building page that says “asking rents from $95/SF, 12 floors of availability from 4,000 to 45,000 SF, Tishman Speyer ownership, LEED Platinum certified, connected to Grand Central Terminal via underground concourse” is dramatically more AI-citable than one that says “exceptional quality in the heart of Midtown.”

What AI gets wrong about Manhattan office leasing

When AI does surface information about specific Class A Manhattan buildings or the leasing market generally, it is frequently wrong in ways that are consequential for leasing decisions. These are not minor errors. They concern the exact information a company needs before committing to a multi-year, multi-million-dollar lease.

Stale rent figures

Manhattan trophy rents have moved dramatically. In 2022, the Midtown Class A average asking rent was approximately $76/SF. By Q4 2025, Cushman & Wakefield reported Class A Midtown rents at $84.98/SF, with trophy floors at One Vanderbilt surpassing $265/SF and the market’s new record at $320/SF. AI models trained on data from even 12–18 months ago will quote figures that are materially below current market. A company relying on AI for rent benchmarking is anchoring to stale data.

Wrong landlord attribution

The Manhattan office ownership landscape has shifted significantly. Boston Properties now owns 767 Fifth Avenue (GM Building) — but AI systems sometimes attribute it to earlier ownership structures. 70 Hudson Yards is under development by Related and Oxford Properties with Deloitte as anchor tenant, but AI may treat it as unbuilt or unknown. Equity sales, JV recapitalizations, and ownership changes at individual buildings do not update in AI training data reliably.

Confusing submarkets and addresses

AI routinely conflates Hudson Yards buildings. It has described 10 Hudson Yards (primarily retail and office floors), 30 Hudson Yards (office-dominant, KKR and Warner Bros. Discovery), 50 Hudson Yards (largest in the complex, BlackRock anchor), and 55 Hudson Yards (law firms and Coinbase) as interchangeable or confused their tenants. These are distinct buildings with different landlord structures, availability windows, and rent profiles.

Availability errors

AI cannot know what is currently available. As noted, it surfaces 270 Park Avenue — JPMorgan’s owner-occupied headquarters since October 2025 — as a leasing option. Conversely, 55 Hudson Yards, which has floors actively available in 2026, rarely appears in AI leasing recommendations because its recent press coverage is thin compared to higher-profile neighbors. Buildings with available space are systematically replaced in AI answers by buildings that are press-saturated but fully leased.

Outdated large-tenant information

Meta’s earlier commitment to significant Hudson Yards space — which it subsequently sublet, with KKR taking a large portion at 30 Hudson Yards — is an example of AI carrying forward outdated tenant-building associations. AI may still describe buildings using tenant information from 2022–2023 that no longer reflects current occupancy. This is particularly dangerous when companies ask AI to assess a building’s tenant mix as a signal of quality or submarket positioning.

The Manhattan office market: 2026 fundamentals

Manhattan is the largest office market in the United States, with approximately 500 million square feet of inventory across Midtown, Midtown South, and Downtown. Understanding the overall market context is essential for interpreting AI visibility patterns — and for deciding where to focus a visibility strategy.

Q1 2026 leasing volume came in at 10.4 million square feet (JLL) to 11.78 million square feet (Colliers) depending on methodology. By any measure it was the strongest first quarter in at least six years. The Bank of America lease at One Bryant Park (2.4 million square feet on a 20-year term, signed March 2026) accounted for more than one-fifth of Q1 volume on its own — a single deal that illustrates how concentrated the trophy market has become (Bloomberg, March 2026; Colliers Q1 2026).

Overall availability dropped from 17.3% a year ago to 14.6% in Q1 2026 (Commercial Observer / JLL). Trophy availability (the most coveted subset) fell 22% year-over-year to 16.9 million square feet from 21.2 million square feet (JLL Q1 2026). Class A sublet space available fell to just 4.1 million square feet, down 13% from the prior quarter — meaning the softness associated with the sublease glut is fading fast for high-quality assets.

Midtown asking rents averaged $84.74/SF for Class A in Q1 2026 (Cushman & Wakefield Q4 2025 MarketBeat, updated Q1 2026). The Grand Central submarket commands the highest average at approximately $93.37/SF for direct Class A space. Trophy floors at One Vanderbilt range from $265 to $320+ for direct leases and a sublease listing on the 73rd floor is now marketing at $350/SF, which would set a new record (The Real Deal, April 2026).

AI company demand has become the defining narrative of the 2026 leasing cycle. In Q1 alone, AI firms leased 415,000 SF in Manhattan, averaging 34,500 SF per deal. The 2025 full-year AI leasing total was 845,000 SF. At the Q1 2026 pace, AI sector leasing could approach or exceed 1.7 million SF for the full year — a roughly 2x acceleration year-over-year (Bisnow, April 2026). Notable AI tenant activity driving this: Anthropic seeking 250,000–450,000 SF, Clay committing to 163,095 SF at 11 Madison, and Nscale’s record rent at One Vanderbilt.

Hudson Yards is now fully validated as a Class A office submarket. The $2.45 billion capitalization of 70 Hudson Yards (1.4 million SF, Deloitte as anchor tenant, Related/Oxford) announced in January 2026 demonstrated continued institutional appetite for development even as other pipeline projects stalled. The Related Companies’ multi-tower complex at Hudson Yards — 10, 30, 50, 55 Hudson Yards, plus One and Two Manhattan West via Brookfield — constitutes one of the largest concentrations of new Class A inventory added to any U.S. market in the past decade.

Key landlords in 2026: SL Green Realty Corp. (Manhattan’s largest commercial landlord, 30+ million SF, flagship One Vanderbilt and 245 Park, where the Carlyle Group signed a 150K SF deal in March 2026); Vornado Realty Trust (20.6 million SF, Penn District revitalization including the PENN 1 and PENN 2 repositionings); Brookfield Properties (Manhattan West complex including One and Two Manhattan West); Related Companies (Hudson Yards); Tishman Speyer (Rockefeller Center, MetLife Building at 200 Park Avenue); Boston Properties (GM Building at 767 Fifth, One Bryant Park).

The disruptors: buildings breaking through in 2026

The following table profiles buildings currently marketing space in 2026, assessing both their market fundamentals and their AI visibility. Several are breaking through in AI answers precisely because they have generated significant recent press coverage through large lease announcements or record rents.

The visibility pattern is consistent across both tables: buildings that generate ongoing, named, fact-rich press coverage earn AI citations. Buildings that are equally good — or better, from a pure leasing perspective — but generate quiet deal flow under confidential terms are functionally invisible in AI answers.

Two Manhattan West illustrates this precisely. The building secured three major anchor tenants (Cravath, KPMG, D. E. Shaw) on lengthy leases — a stronger pre-leasing story than many of its neighbors. Yet it appears in AI answers only a fraction as often as One Manhattan West next door, which has generated significantly more media coverage simply by virtue of sequencing and press strategy. When a corporate real estate team asks ChatGPT about Brookfield office options in the Hudson Yards area, they are far more likely to hear about One Manhattan West than Two Manhattan West, despite both buildings being of comparable quality.

What actually works: the playbook for getting a Manhattan office building recommended by AI

The insight from this analysis is actionable: AI visibility in commercial real estate is not a function of building quality, and it is not purchased through advertising. It is earned through information density, source authority, and structured content. Here is what moves the needle for a Manhattan Class A office property.

1. Audit what AI currently says about your building

Before deploying any content strategy, establish a baseline. Query ChatGPT, Perplexity, Claude, and Gemini with prompts your target tenants actually use: “best Class A office buildings in Hudson Yards,” “trophy office space Midtown Manhattan for financial services firms,” “[your building name] asking rent and available floors.” Document every mention, every error, every competing building that appears in your place. A Metricus AI visibility report runs this across hundreds of query variations automatically and maps errors to their sources. This is the starting point for any intelligent visibility strategy. See also: how to run a DIY AI visibility audit.

2. Implement structured data on every building and availability page

Schema markup is the single highest-leverage technical change a CRE landlord or broker website can make for AI visibility. Implement RealEstateListing schema on availability pages with named fields for address, square footage, asking rent, availability date, and landlord entity. Add LocalBusiness schema to building profile pages. Include FAQPage schema answering the questions tenants actually ask: “What are the asking rents at [building]?”, “Who manages the building?”, “What amenities does [building] offer?” AI crawlers that do not execute JavaScript can still read structured data embedded in HTML. See the broader framework in our AI visibility scores explainer.

3. Publish data-rich, citable availability content in plain HTML

Replace JavaScript-rendered availability tables with static HTML that contains named, numerical facts. “Three floors available: 14th floor (12,500 SF), 22nd floor (18,000 SF), and 31st floor (8,200 SF). Asking rents from $95/SF. Direct deal with [Landlord Name]. Available Q3 2026.” This sentence is dramatically more AI-citable than the equivalent information rendered in a React component. Publish quarterly availability updates as blog posts or market reports so the content is time-stamped and crawlable. Per the GEO research from Princeton/Georgia Tech (Aggarwal et al., 2023), content with specific statistical claims is up to 40% more likely to be cited by generative AI. More on this approach: the AI visibility action plan.

4. Build citations on authoritative CRE publications and data platforms

AI training data disproportionately reflects sources with high domain authority and high update frequency. For Manhattan office buildings, the key citation platforms are: The Real Deal, Commercial Observer, Bisnow, CoStar News, and JLL / CBRE / Cushman & Wakefield quarterly market reports. Every lease deal, building milestone, or tenant announcement should generate a named press item in at least one of these outlets. Ensure the building is fully and accurately listed on CoStar and LoopNet with current availability, named rent ranges, and landlord information — these platforms are among the most AI-cited sources for CRE data. For a deeper look at AI citation dynamics, see what is AI visibility.

5. Correct factual errors at their source

If AI is attributing wrong tenants, wrong ownership, or wrong availability to your building, the errors come from specific sources in the training corpus — typically outdated CoStar records, stale news articles, or archived broker marketing materials. Identify those sources and correct the underlying data. SL Green’s aggressive press strategy for One Vanderbilt — a new announcement in almost every quarter since opening — effectively overwrites older, less accurate information in AI training data by volume. More on fixing factual AI errors: fixing AI hallucinations about your brand.

6. Engage agency support with AI-visibility expertise

Most CRE marketing agencies are optimizing for traditional SEO and earned media, not AI visibility. An agency that understands AI visibility audits will prioritize different outputs: structured data implementation, plain-HTML content architecture, and citation velocity in high-domain-authority CRE publications rather than generic backlink profiles. The firms that move first on this will establish competitive separation that becomes harder to close as AI systems update on a training cycle measured in months, not days.

The case for auditing your CRE brand’s AI visibility now

The Manhattan Class A office market is tightening. Trophy availability is down 22% year-over-year. Midtown Class A availability sits below 12%. AI companies — the fastest-growing tenant category in Q1 2026 — are specifically looking for the best buildings, and they are using AI to research those buildings before the first broker call.

This creates an unusual window. The landlords and brokers who invest in AI visibility now — structured data, plain-HTML content, aggressive press announcements with named data — will establish a compounding advantage. AI training data does not update daily. Each piece of authoritative, fact-rich content that goes live this quarter becomes part of the corpus that shapes AI answers for the next 6 to 18 months. The buildings that act first will own the AI shortlist for the next leasing cycle.

Conversely, the cost of inaction is real and growing. When Anthropic’s corporate real estate team — or any of the dozens of AI-funded companies expanding rapidly in 2026 — asks ChatGPT which buildings to shortlist for a 200,000+ SF lease in Midtown, the same 4–5 buildings appear in the answer. That shortlist shapes the broker brief. The broker brief shapes the tour list. The tour list shapes the lease. Buildings that never appear in the AI answer have effectively been disqualified before a human has made a single decision.

This is the dynamic that drove AI visibility for real estate broadly, and it applies with particular force in commercial office leasing, where deal sizes are large, decision cycles are long, and first-mover advantages in the tenant’s shortlisting process compound into leased square footage and record rents. The buildings at the top of the AI recommendation list are, not coincidentally, also achieving the highest rents in the market.

Sources: JLL Q1 2026 Manhattan Office Research (via NYREJ, April 2026); Colliers Q1 2026 Manhattan Office Report; Cushman & Wakefield Q4 2025 Manhattan MarketBeat; Commercial Observer, “Manhattan Office Leasing Starts 2026 Strongly Amid Shrinking Class A Inventory” (April 2026); Commercial Observer, “Midtown Manhattan’s Tight Office Vacancy Is Both a Blessing and a Curse” (February 2026); The Real Deal, “One Vanderbilt sublease shoots for record $350 psf” (April 8, 2026); The Real Deal, “Manhattan office leasing rallies to strong quarter” (April 3, 2026); The Real Deal, “AI giant Anthropic seeks more office space in NYC” (January 22, 2026); Bloomberg, “Bank of America Signs 20-Year Lease for Entire One Bryant Park Office Tower” (March 16, 2026); Bloomberg, “Anthropic Seeks New Manhattan Office Space in NYC Expansion” (January 20, 2026); Bloomberg, “New York City Office Leases Rebound as AI Companies Move In” (March 19, 2026); Bisnow, “AI Companies Growing Their NYC Offices At Breakneck Speed” (April 2026); NYREJ, “Manhattan office leasing roars back in Q1 2026, fueled by AI and mega deals according to JLL” (April 2026); SL Green, “SL Green Signs Office Leases Totaling 490,000 Square Feet During the First Two Months of 2026” (March 2026); Commercial Observer, “Carlyle Group Inks 150K-SF Deal at SL Green’s 245 Park Avenue” (March 2026); Governor Hochul’s office / Empire State Development, Clay HQ expansion announcement (April 2026); Related Companies / Oxford Properties, 70 Hudson Yards $2.45 billion capitalization announcement (January 2026); The Real Deal, “BlackRock soars above 1M sf at 50 Hudson Yards” (October 2025); Wikipedia / JPMorgan, 270 Park Avenue opened October 2025; Princeton/Georgia Tech GEO study (Aggarwal et al., 2023). AI mention rates based on Metricus internal testing across ChatGPT, Perplexity, Claude, and Gemini (Q1 2026). Learn more about how we measure AI visibility.

Related reading

• AI visibility for real estate: why ChatGPT recommends Zillow instead of your brokerage — the foundational framework for real estate AI visibility, applicable across residential and commercial segments.
• The 5-step AI visibility action plan — the general framework for turning audit findings into fixes.
• Fixing AI hallucinations about your brand — the deep dive on correcting factual errors at their source.
• What is AI visibility? — the complete explainer on how brands appear in AI answers.
• Free AI visibility check — run a quick manual check for your building or CRE brand before ordering a full report.