AI and Data Trends in Commercial Real Estate (2026): What the Numbers Actually Say

TL;DR: Commercial real estate has finished the experimentation phase and is stuck at the transition to production. JLL found 88% of investors and owners piloting AI while only 5% have achieved all their program goals. Altus found roughly two thirds of CRE organizations experimenting and fewer than one in ten operationalized. The gap is not model quality. About half of CRE executives name data fragmentation as their top data problem, and a portfolio split across Yardi, MRI, RealPage, ARGUS, and CoStar cannot answer a portfolio-level question without a human reconciling it first. This is a review of the nine trends worth tracking in 2026, what the primary research actually says, and what has to be true underneath before any of it produces a number you would put in front of an investment committee.

Every commercial real estate conference this year has an AI track, and most of the sessions describe the same arc. A firm ran pilots. Some worked. Nothing scaled. Everyone in the room nods.

The research says the same thing in numbers. JLL's 2025 Global Real Estate Technology Survey found 88% of investors, owners, and landlords have started piloting AI, up from roughly 5% in 2023, with the average firm pursuing five use cases at once. In the same survey, 5% said they had achieved all the goals they set for their AI programs. Altus Group's work lands in the same place from the other direction: close to two thirds of CRE organizations are experimenting, fewer than one in ten have operationalized anything.

Those two numbers, 88% and 5%, describe the actual state of the industry better than any vendor deck. This piece walks through the nine trends behind them.

Key takeaways

• Piloting is universal, production is rare. JLL: 88% of investors and owners piloting, average of five concurrent use cases, 5% achieving all program goals. 87% of firms are raising real estate technology budgets to fund it.

• Data fragmentation is the named blocker. Altus Group found roughly 50% of CRE executives cite fragmentation as their top data problem, ahead of quality at 24% and governance at 23%.

• Data centers are the demand story. CBRE recorded 2,497.6 MW of North American absorption in 2025, up 38%, with vacancy at a record low near 1.6% and about 74% of construction preleased. Hyperscaler capital expenditure guidance for 2026 runs from roughly $630B to $725B.

• Capital is moving again, and so is the maturity wall. CBRE put 2025 US investment volume at $499B, up 22%. The Mortgage Bankers Association counts $875B of commercial mortgages maturing in 2026 against a $5.0 trillion total.

• Office distress is structural. Trepp's office CMBS delinquency rate reached an all time high of 11.66% in August 2025 and stood at 11.71% in March 2026, above the post crisis peak near 10.7%.

• Venture capital has re-rated the sector around AI. CRETI put 2025 proptech funding at $16.7B, up about 68%, with AI-native companies taking $4.5B.

• Regulators have arrived. The RealPage settlement with the Department of Justice, filed November 2025, is the first practical compliance template for algorithmic pricing, and RICS is consulting on formal AI in valuation guidance in 2026.

Trend 1: The pilot to P&L gap is the defining problem

The most useful finding in CRE technology research this cycle is the distance between activity and results.

JLL's survey covers both sides of the market. Among investors, owners, and landlords, 88% are piloting. Among corporate occupiers, 92% are. JLL catalogued 56 distinct AI use cases across the value chain, which explains the five-simultaneous-pilots average: there is no shortage of things to try. And 5% report hitting all of their goals.

Deloitte's 2026 Commercial Real Estate Outlook, built on a survey of more than 850 global CRE executives across 13 countries, shows the same pattern with different instruments. Only 22% report using industry-specific AI software platforms and 20% use publicly available large language models. Nineteen percent describe themselves as still early in adoption and 27% report implementation challenges outright.

The encouraging line in the Deloitte data is the direction of travel. Respondents reporting a transformative impact from AI rose from 1% to 7% year over year, and those reporting incremental operational improvement rose from 7% to 24%. Real gains are showing up. They are just concentrated in a small group.

What separates that group is rarely the model. JLL's October 2025 research on AI readiness found more than 60% of real estate investors are unprepared technically and strategically to adopt AI at full scale. That is a statement about internal foundations, not about what the technology can do.

Trend 2: Fragmentation is the top-cited data problem, by a wide margin

Ask CRE executives what is holding the technology back and they name their own data. Altus Group's research puts fragmentation first at around 50%, with data quality at 24% and governance at 23%.

Anyone who has assembled a portfolio report knows the shape of it. Property level financials live in Yardi or MRI. Residential operations may sit in RealPage or AppFolio. Valuation and cash flow models live in ARGUS. Market and comparable data comes from CoStar. Loan data comes from Trepp. Leases live in PDFs in a document management system, and the terms that actually govern the cash flows live inside those PDFs as prose.

Each system is competent at its job. None was designed to answer a question that crosses all of them, and most portfolio-level questions cross all of them. "What is our true exposure to this tenant across the portfolio" touches leases, property financials, and entity records that spell the tenant's name three different ways. "Why did NOI miss" touches the general ledger, the rent roll, the budget model, and the operational reality that neither one recorded.

Larger firms have a second version of the problem. Multiple instances of the same platform, acquired with the portfolios they came with, running different charts of accounts. Standardizing those retroactively is a remediation project measured in months, which is why so many firms quietly keep doing it in spreadsheets.

We wrote about the mechanics of this specific problem in why Yardi, MRI, and ARGUS disagree, including how entity resolution and a canonical metric definition resolve it without replacing any of the underlying systems. The broader architecture is covered in AI agents, workflows, and knowledge graphs in commercial real estate.

Trend 3: Underwriting is being compressed from hours to minutes

The clearest product category to emerge in CRE AI is document extraction feeding underwriting.

The workflow it targets is well understood. A broker sends an offering memorandum, a trailing twelve month statement, and a rent roll. An analyst spends two to four hours retyping that into a model, normalizing the categories, and sanity checking it against comparable properties. Most of those hours produce a pass.

Platforms including AcquiOS and NOAL now convert those same documents into a validated model in minutes. AcquiOS checks extracted figures against market comparables. NOAL, launched out of Vessel's venture studio, reports that teams evaluate a deal in 10 to 15 minutes rather than two to four hours and underwrite five to ten times more deals with the same staff. Those specific figures are vendor reported rather than independently studied, so treat them as directional. The direction is not in doubt.

Cambio raised $18M at roughly a $100M valuation in January 2026 for AI asset management software aimed at institutional investors, which tells you where the capital thinks this goes next: from screening into ongoing management.

The strategic consequence matters more than the hours saved. If screening capacity increases by a factor of five, the constraint on acquisitions moves from analyst throughput to deal flow and judgment. Firms that only automate extraction get faster at the same funnel. Firms that also widen the top of the funnel change what they can buy. Our underwriting acceleration demo shows what a broker pro forma looks like when it is checked against reality rather than retyped, using fictional data.

There is a caution worth stating plainly. Extraction accuracy on a clean institutional rent roll is very different from extraction accuracy on a scanned rent roll from a 1980s property with handwritten annotations. The value of these tools depends on validation, not just extraction, and on what happens to the exceptions.

Trend 4: Agentic AI has reached property operations

The 2025 story was assistants that answered questions. The 2026 story is agents assigned standing jobs.

AppFolio reported that Realm-X AI actions grew roughly sevenfold year over year in its first quarter of fiscal 2026, with about 500% quarter over quarter growth in adoption of what it calls Performers, agents that own leasing, maintenance, and communication tasks end to end. The company reports that 98% of its customers were using at least one AI-native capability by 2026. These are company-reported figures from earnings coverage, so read them with the usual discount, but the underlying shift is visible across the sector.

EliseAI raised a $250M Series E in August 2025 led by Andreessen Horowitz, more than doubling its valuation to $2.2B after passing $100M in annual recurring revenue. Its platform covers leasing communication, tours, lease audits, maintenance, and rent collection, which is a list of operational jobs rather than a list of features.

Lease abstraction is the cleanest example of the economics. Manual abstraction of a complex commercial lease takes roughly three to eight hours. Realcomm reports deployments cutting per-lease review time by about 85%, from roughly two hours to seventeen minutes, and outsourcing costs by 50% to 90%. Again, industry media and vendor figures rather than controlled study, but abstraction is a task where the before and after is easy to measure honestly, and firms that have done it generally confirm the shape.

Abstraction is also where operations and data strategy meet. A lease abstracted into a database stops being a document and becomes a set of facts a model can compute against: escalation schedules, renewal options, recovery structures, co-tenancy clauses. That is the difference between an AI that summarizes your lease and an AI that can tell you what happens to portfolio NOI if a specific anchor tenant exercises a termination right.

Trend 5: Data centers are the demand story, and they are not slowing

CBRE recorded 2,497.6 MW of net absorption in North American data centers in 2025, up 38% from 1,809.5 MW in 2024. Total capacity grew about 36% to 9,432 MW, and vacancy still fell to a record low near 1.6%. Roughly 74.3% of capacity under construction is already preleased, overwhelmingly to cloud and AI companies. Northern Virginia alone absorbed 1,102 MW, more than double its 2024 figure. First quarter 2026 absorption across the top four markets rose 34% year over year.

Deloitte's outlook found data centers retook the top spot among property types in respondent preference, with the entire new construction pipeline pre-committed across nine major global markets.

Behind that is capital expenditure at a scale the property industry has not previously had to model. Amazon, Microsoft, Alphabet, Meta, and Oracle are guiding to something in the range of $630B to $725B in combined 2026 capex, against a record of roughly $388B in 2025, with about three quarters of it AI-related. Futurum Group frames 2026 as a $690B infrastructure sprint.

Two cautions belong in any serious analysis. First, this demand is concentrated in a handful of counterparties, which is a different risk profile than a diversified tenant roster. Second, the binding constraint has moved from land to power, which means underwriting a data center is now partly an energy interconnection question and partly a regulatory one. Neither of those risks is visible in a traditional CRE data stack.

Trend 6: AI is starting to show up in conventional space demand

Cushman & Wakefield published research in May 2026 projecting that AI adoption will generate roughly 330 million square feet of additional US commercial real estate demand over the next decade, lifting expected total net absorption from a pre-AI forecast of 2.7 billion square feet to just over 3.0 billion through 2035. That is a 12.2% upside revision attributed to AI.

This is the newest and least tested thesis on the list, and it runs against the more common assumption that AI reduces headcount and therefore reduces space. It is worth tracking rather than acting on, and it comes from a named primary source rather than the conference circuit, which is why it belongs in a trends review at all.

Trend 7: Capital is moving again into a large maturity wall

CBRE put full year 2025 US commercial real estate investment volume at $499B, up 22% over 2024 and well ahead of forecasts near 10%. Fourth quarter volume rose 29% year over year to $171.6B, with private investors the largest buyer cohort at $92B in the quarter.

Meeting that recovery is a substantial refinancing requirement. The Mortgage Bankers Association counts $875B, or 17%, of the $5.0 trillion in outstanding US commercial mortgages maturing in 2026, down 9% from the $957B scheduled in 2025. Part of that volume is the product of extensions granted from 2023 through 2025. MSCI separately estimates roughly $600B of mortgages have been pushed past their original maturity dates.

Office remains the acute case. Trepp's office CMBS delinquency rate hit an all time high of 11.66% in August 2025, spiked again in January 2026 when One New York Plaza transferred to special servicing, and stood at 11.71% in March 2026. That is above the post financial crisis peak of roughly 10.7% in late 2012. Trepp characterizes the current stress as structural rather than cyclical, which is a meaningfully different underwriting statement.

The connection to data strategy is direct. In a market where capital is available but selective, and where a large slice of the debt stack has to be refinanced at repriced rates, the difference between firms is the quality and speed of their deal selection and their visibility into their own downside. Both are data problems before they are AI problems. A firm that needs three weeks to assemble a defensible portfolio view is not going to win a competitive process or catch a covenant problem early.

Trend 8: Venture capital has re-rated proptech around AI

CRETI put 2025 proptech venture funding at $16.7B, up roughly 68% year over year and above the 2019 pre-pandemic peak near $14B. AI-native companies captured $4.5B of that, with their share growing 42% year over year, close to double the growth of SaaS peers. Rounds of $100M or more accounted for $11.2B. January 2026 alone saw roughly $1.7B invested, up 176% against January 2025.

Bisnow's February 2026 review noted that the four newest proptech unicorns are all AI-centered: EliseAI at $2.2B, Bedrock Robotics at $1.75B after a $270M Series B, Juniper Square at $1.1B after a $130M Series D, and Vantaca at $1.25B after a $300M growth investment. Together they raised about $950M in a year.

For an operator the practical read is about vendor risk and category maturity. Capital at that level means the tools will improve quickly and consolidate quickly. It also means a portion of what is being sold today is ahead of what it can deliver, and that procurement should weight data portability and integration openness heavily, because the vendor set in 2029 will not look like the one in 2026.

Trend 9: The regulators arrived before the industry finished experimenting

Two developments in the last year set the governance baseline for algorithmic tools in real estate.

The Department of Justice settlement with RealPage, filed in November 2025 in the Middle District of North Carolina and subject to court approval, resolves the government's algorithmic price fixing case. RealPage admitted no liability but agreed that its rent recommendation tools may not use nonpublic, competitively sensitive data from competing landlords, limiting runtime pricing to a landlord's own data plus public information. It accepted a court-appointed monitor with access to source code, model training documentation, and runtime logic, and agreed to cooperate in the DOJ's ongoing case against landlord users. Law firm analyses have described it as the first de facto compliance blueprint for algorithmic pricing.

The operative lesson is not "avoid algorithms." It is that when an algorithm influences a commercial decision, you may have to explain what data went into it and demonstrate that explanation to a third party. Systems that cannot produce lineage on demand are a liability, not just an inconvenience.

On the valuation side, the 2025 RICS Red Book update permits AI in valuations only with appropriate governance and human oversight, and requires that automated valuation models be overseen by a qualified professional who retains responsibility. RICS is issuing its first global practice guidance on artificial intelligence in real estate valuation for public consultation in the second quarter of 2026, with publication expected later in the year.

Both point the same direction. Human accountability for the judgment, machine assistance for the assembly, and a documented trail from output back to source.

What actually separates the 5% from everyone else

Read the nine trends together and a single requirement runs underneath all of them.

An underwriting model that reads a rent roll is only useful if the extracted figures can be checked against your own portfolio history and the market. A leasing agent is only useful if it can see the actual lease terms, not a summary of them. A portfolio NOI view is only useful if every property is measured with the same yardstick. A regulator-ready pricing tool has to produce lineage. A data center underwrite needs power and interconnection data that no property system holds.

Every one of those is a question about whether your systems can be reasoned over as a whole. That is what the 5% built and the 88% skipped.

The failure pattern is consistent, and it is not really about AI. A firm buys a capable tool, connects it to one system because that is what the integration supports, and gets answers that are correct within that system and wrong at the portfolio level. Someone notices, trust drops, and the pilot quietly ends. The postmortem calls it an AI problem. It was a context problem.

CRE is not unusual here. The same gap between adoption and results appears in private equity, where 86% of dealmakers use generative AI and nearly 40% of GPs expect no material financial impact this year, and in healthcare operations, where 80% of health systems are acting on revenue cycle AI and 3% report significant returns. The common factor is federated data, not immature models.

Four things have to be true before AI produces numbers a firm will act on:

One definition per metric, owned centrally. Same-store NOI, economic occupancy, effective rent, and recovery ratio each need one firm-level definition, precise enough that two analysts compute the same number, with every property system's local source mapped to it in code rather than in a policy document. This is finance work, not technical work, and it is the highest leverage two weeks in the entire program.

Entity resolution across the portfolio. One tenant, one owner, one property, regardless of how five systems spell them. Without it, tenant exposure, portfolio concentration, and counterparty risk are all approximate.

Documents made computable. Leases, loan agreements, JV agreements, and management agreements hold the terms that drive the cash flows. Until those terms are structured facts, any model that touches them is guessing.

Lineage on every figure. Every number traceable to source records, in seconds. This is what makes a number defensible in an investment committee, an audit, an LP conversation, and now a regulatory one.

How OutcomeCatalyst fits

OutcomeCatalyst builds the layer that sits between the systems you already run and the AI you want to use. We connect Yardi, MRI, RealPage, AppFolio, ARGUS, your general ledger, your CRM, and your lease and loan documents into one governed model your team and your agents can reason over. Nothing gets migrated and no platform gets replaced.

For a CRE owner, operator, or investment manager that means:

• A firm-owned metric registry, so same-store NOI means one thing across every property and every system that reports it.

• Entity resolution across the portfolio, so tenant exposure and counterparty concentration are visible rather than estimated.

• Leases and loan agreements made queryable, so escalations, renewal options, recovery structures, and covenant terms are facts a model can compute against instead of PDFs someone has to reread.

• Lineage on every number, which is what makes a figure defensible in an IC memo and in front of a regulator.

• Live connections rather than exports, so the portfolio view reflects this week rather than last quarter.

Our NOI and occupancy intelligence demo and deal origination scoring demo show what this looks like in practice, using fictional data. The full picture is on our commercial real estate page, and brokerage-side workflows are on our real estate page.

For the deployment side of this, which agents to assign first and what to keep under human sign-off, see AI agents for commercial real estate workflows. If you are deciding between building the layer internally and buying it, we compared both paths in buy versus build an AI context layer, and covered sequencing in data unification in 2026.

If your firm has AI budget approved and keeps stalling at the same place, where the tool works on one system and breaks across the portfolio, that is the problem we solve. Start a conversation about your portfolio.

Common questions about AI and data in commercial real estate

What percentage of commercial real estate firms are using AI in 2026?

JLL's 2025 Global Real Estate Technology Survey found 88% of investors, owners, and landlords piloting AI and 92% of corporate occupiers, but only 5% report achieving all their AI program goals. Altus Group found roughly two thirds experimenting with fewer than one in ten operationalized. Piloting is close to universal, production is rare.

What is the biggest obstacle to AI adoption in CRE?

Data fragmentation. Altus Group found roughly 50% of CRE executives name it as their top data problem, ahead of data quality at 24% and governance at 23%. A portfolio split across Yardi, MRI, RealPage, ARGUS, and CoStar cannot answer a portfolio-level question without manual reconciliation.

Can AI underwrite a commercial real estate deal?

AI can now extract an offering memorandum, trailing twelve month statement, and rent roll into a structured model in minutes rather than hours, and vendors report five to ten times more deals reviewed with the same staff. The judgment, the assumptions, and the recommendation stay human. The gain is in assembly and screening capacity, not in the decision.

How much are data centers driving CRE demand?

CBRE recorded 2,497.6 MW of North American net absorption in 2025, up 38%, with vacancy at a record low near 1.6% and about 74% of construction preleased. Hyperscaler capex guidance for 2026 runs roughly $630B to $725B. Concentration risk and power availability are the two underwriting factors that traditional CRE data stacks do not cover.

Do I need to replace Yardi or MRI to use AI?

No, and replacement projects usually make things worse before they make them better. The workable path is to leave the systems of record in place and add a governed layer above them that resolves entities and standardizes definitions. Migration is a multi-year project that does not need to be won to get portfolio visibility.

What does the RealPage settlement mean for firms using algorithmic tools?

The November 2025 DOJ settlement bars RealPage's rent recommendation tools from using nonpublic competitively sensitive data from competing landlords and installs a court-appointed monitor with access to code and model logic. The practical takeaway for any operator is that algorithmic decisions need documented inputs and lineage, because you may have to explain them to a third party.

How is AI used in property management specifically?

Leasing communication, tour scheduling, maintenance triage, lease audits, and rent collection are the workflows with production deployments. AppFolio reported roughly sevenfold growth in AI actions year over year, and EliseAI reached a $2.2B valuation on a platform covering the same set. Lease abstraction is the most measurable single win, with reported time reductions near 85%.

Is proptech AI funding real or a bubble?

CRETI put 2025 proptech venture funding at $16.7B, up about 68%, with $4.5B into AI-native companies and four new AI-centered unicorns. At that level, tools improve quickly and consolidate quickly. The practical implication for buyers is to weight data portability and integration openness heavily in procurement, because the vendor set will look different in a few years.

What is the single highest-leverage first step?

Writing down one definition per portfolio metric, agreed by the people who own the numbers, and mapping each property system's local source to it. Twenty to thirty definitions. It requires no technology, it is unglamorous, and every AI initiative downstream depends on it.

Sources

• JLL, 2025 Global Real Estate Technology Survey (88% of investors/owners/landlords piloting AI, 92% of occupiers, 5% achieving all goals, 56 use cases, 87% raising technology budgets): jll.com

• JLL Research, "AI for business growth," October 2025 (more than 60% of investors unprepared technically and strategically for full-scale AI adoption): jll.com

• Deloitte, 2026 Commercial Real Estate Outlook, survey of 850+ global CRE executives across 13 countries (22% using industry-specific AI platforms, 20% using public LLMs, 19% early in the AI journey, 27% reporting implementation challenges, data centers top property type): deloitte.com

• Altus Group (roughly 50% cite data fragmentation as top data problem, 24% quality, 23% governance; two thirds experimenting, fewer than one in ten operationalized): altusgroup.com

• CBRE, North America Data Center Trends and 2025 records release (2,497.6 MW absorbed in 2025, up 38%; vacancy near 1.6%; 74.3% of construction preleased; Northern Virginia 1,102 MW): cbre.com

• Futurum Group, "AI Capex 2026: The $690B Infrastructure Sprint" (combined hyperscaler 2026 capex guidance of roughly $630B to $725B against about $388B in 2025): futurumgroup.com

• Cushman & Wakefield, May 2026 (AI to add roughly 330 million square feet of US CRE demand over the next decade, a 12.2% upside revision to 2035 absorption): cushmanwakefield.com

• CBRE, Q4 2025 US Capital Markets Figures ($499B full-year 2025 investment volume, up 22%; Q4 up 29% to $171.6B): cbre.com

• Mortgage Bankers Association (17%, or $875B, of $5.0 trillion in outstanding commercial mortgages maturing in 2026): mba.org

• Trepp office CMBS delinquency data via Colliers Knowledge Leader (record 11.66% in August 2025; 11.71% in March 2026; prior post-crisis peak near 10.7%): knowledge-leader.colliers.com

• CRETI via Multifamily Dive and Commercial Observer (2025 proptech venture funding $16.7B, up about 68%; $4.5B into AI-native companies; $11.2B in rounds of $100M+): multifamilydive.com

• Bisnow, February 2026 (four newest proptech unicorns all AI-centered: EliseAI $2.2B, Bedrock Robotics $1.75B, Juniper Square $1.1B, Vantaca $1.25B): bisnow.com

• EliseAI, August 2025 ($250M Series E led by Andreessen Horowitz, valuation to $2.2B, past $100M ARR): eliseai.com

• AppFolio Realm-X figures via earnings coverage (roughly 7x year-over-year growth in AI actions, about 500% quarter-over-quarter growth in Performers adoption, 98% of customers using at least one AI-native capability): finance.yahoo.com

• Realcomm on AI lease abstraction (per-lease review time down roughly 85%, from about 2 hours to 17 minutes; outsourcing costs down 50% to 90%): realcomm.com

• Wilson Sonsini and Multifamily Dive on the DOJ RealPage settlement, November 2025 (no nonpublic competitor data in runtime pricing, court-appointed monitor with access to code and model training documentation): wsgr.com

• RICS, valuation standards and AI guidance (2025 Red Book update requires governance and human oversight for AI in valuation; first global practice guidance on AI in real estate valuation out for consultation in Q2 2026): rics.org

OutcomeCatalyst connects the systems you already run into a governed intelligence layer your team and your agents can reason over. Demos on this site use fictional data. To see this on your own portfolio, start a conversation.