The LLM Model Just Became a Commodity

Commercial Real Estate AI Strategy

AI Models Are Converging. The Advantage Now Sits Above the Model.

Stanford’s 2026 AI Index put a number on something CRE owners need to feel: AI capability is converging fast. The competitive moat has moved up the stack, and most portfolios are not built for it yet.

Per Stanford’s 2026 AI Index, the performance gap between open-source AI and the most expensive frontier models shrank from roughly 8% to roughly 1.7% in a single year. The cost to hit benchmark performance is falling 5 to 10 times annually. Most enterprises now run three or more model families at once, which means multi-model operation is becoming the default rather than a specialized architecture choice.

In practical terms, the model itself is becoming a commodity. For commercial real estate owners and operators, that changes the AI conversation that has dominated boardrooms for the last two years. “Which AI tool should we buy?” is already the wrong first question, because tool selection only creates durable value when the owner already controls the data, connectivity, permissions, and workflows the tool depends on.

Why tool-first buying fails in CRE

Here is the pattern still common across portfolios. A vendor pitches a new building application with its own dashboard, its own AI claim, and its own model under the hood. The evaluation focuses on features. One building gets selected as the pilot. The contract is signed, the integration is custom, and the first reports look promising.

Then the conditions change. A better model ships. Pricing resets. The vendor is acquired. The next asset has a different BMS, access-control stack, or property-management system. The pilot does not transfer cleanly, so the team starts over. Each restart consumes capital, staff time, and political attention inside the ownership group.

The deeper cost is ownership of the operating record. Utility intervals, ticket histories, access events, maintenance cycle times, occupancy signals, and vendor performance histories become inputs to the vendor’s product improvement loop. When the contract ends, or when the owner wants a different decision engine, the history that would have made the next model useful is incomplete, poorly structured, or contractually hard to extract. The building may still have sensors and software. What it lacks is a portable operating memory the owner can reuse.

In a market where model performance is converging, that trap gets more expensive every quarter it remains in place. Cheaper models increase the number of tools that will be pitched into the same fragmented stack. Without owner-controlled foundations, each new purchase widens the gap between activity and durable capability.

What compounds when models converge

When AI capability converges across providers, competitive value moves above the model into assets only the ownership organization can assemble and keep. CRE makes that shift sharper than most industries because buildings generate private operational data every minute of every day, across property types, in formats that do not appear on the public internet. That data becomes valuable for AI only when it is captured in a form the owner can govern, retain, and reuse across assets and over time.

Proprietary operating data you can actually use

Frontier models already know the public internet. They do not know how a specific multifamily asset behaves during shoulder-season utility spikes, how long a particular vendor takes to close after-hours tickets, which access-control exceptions correlate with tenant complaints, or how occupancy patterns shift after a capital project. Those signals sit in meters, BMS logs, work-order systems, access platforms, and leasing tools. Legal ownership on paper is not enough. The operational test is whether the owner can export the history, retain lineage, join it across systems, and apply a new analytics or AI layer without renegotiating the original vendor relationship.

When that test fails, asset managers cannot benchmark true utility performance across peers in the portfolio. Insurance submissions rely on narratives instead of clean operating evidence. Diligence teams reconstructing recoverable NOI at sale or refinance do the reconstruction from vendor extracts and tribal knowledge, which slows the process and weakens the owner’s negotiating position.

Operating workflows that put decisions in the right seat

A strong model dropped into a broken process produces weak decisions faster. The useful design work is operational: where data enters the process, who reviews exceptions, what threshold triggers a work order or a capital request, and which actions an automated agent is allowed to take without human approval. In property operations, that might mean a utility anomaly route that reaches the chief engineer with context before the bill arrives. In leasing, it might mean renewal risk signals that reach the asset manager with the building history attached, not a disconnected dashboard alert.

Those workflows are an ownership design problem. They encode how the company wants buildings run. They do not ship inside a vendor demo, and they cannot be rebuilt from scratch at every address if the underlying data model keeps changing.

An orchestration layer the owner controls

As portfolios adopt multiple AI tools, often without a single architecture decision, the controlling layer is the one that decides which model may act, on which data, under whose identity, with what retention rules, and with what review trail. That layer covers identity, permissions, privacy, lineage, retention, and rules of use. If it lives inside each vendor product, the owner cannot answer basic questions across the portfolio: who accessed tenant data last quarter, which model wrote to a building system, or whether a retired tool still holds historical credentials.

For CRE, that is not an abstract IT concern. Access control, HVAC setpoints, elevator systems, and metering are operational technology with real safety, liability, and tenant-experience consequences. Orchestration without owner governance turns automation into unowned action.

Institutional knowledge encoded as reusable decision logic

Every ownership group has patterns that currently live in the heads of strong asset managers and senior property managers: how renewals are judged in a soft submarket, which capital deferrals create downstream OpEx, what “normal” looks like for a class of assets after weather events. Owners who convert that judgment into structured decision logic (rules, prompts, agent policies, and review checkpoints) create an asset that compounds as people turn over and as model vendors change. The model can be swapped. The encoded operating standard remains.

Those four layers, proprietary data, workflows, orchestration, and encoded institutional knowledge, are where durable differentiation sits once model performance compresses toward a narrow spread.

What the market looks like when the foundation is missing

Without those layers, every new tool becomes another silo. Data remains inconsistent across properties, difficult to trust in an investment committee packet, and locked inside contracts that make portability slow or expensive. Each building needs custom integration. The portfolio ends up with a patchwork of locally “smart” sites that cannot transfer learning from one asset to the next.

The economic effects show up in familiar CRE workflows. Utility programs cannot be managed as a portfolio discipline because consumption history is fragmented. Insurance underwriters receive incomplete operating evidence and price uncertainty into premiums. At disposition or refinance, diligence teams discover recoverable operating improvements the owner never captured in a governed form, and that discovery can show up in valuation. AI deployments become automation without clear permissions, which creates risk without creating a reusable portfolio capability.

Cheaper models do not solve that condition. They increase the volume of tools that can be layered onto the same fragmentation, which raises the cost of delay.

What changes when the owner owns the stack above the model

When the foundation is owner-controlled, properties become portable intelligence assets rather than vendor-bound software installs. A shared data model lets an asset manager compare utilities, work-order performance, and occupancy drivers across the portfolio with consistent definitions. Decision platforms can be swapped without rewiring every building network. Vendor transitions no longer erase multi-year operating history. Privacy, security, and compliance reviews become answerable because identity, retention, and lineage are governed in one place the owner controls.

At refinance, recapitalization, or sale, that capability matters because the portfolio’s operating intelligence can be shown as an owned asset: clean history, governed access, and transferable workflows. The alternative is a stack of subscriptions that must be explained, replaced, or written down in diligence.

How OpticWise builds that owner-controlled stack

The requirements above define the solution. OpticWise delivers it as a two-layer model designed for owner control in commercial real estate.

The path from a fragmented stack to that posture is the PPP 5C™ plan:

1. Clarify: define success metrics, map who owns what, identify leakage, and document what is trustworthy and portable.
2. Connect: establish secure, owner-controlled connectivity that is repeatable from property to property.
3. Collect: capture and normalize the data buildings already generate into a consistent model the owner can reuse.
4. Coordinate: govern identity, access, privacy, lineage, retention, and rules of use.
5. Control: enable any decision engine or workflow (vendor platform, internal analytics, or AI model) to act under owner permissions.

Where to start

Start with one building. Map who owns what, where the data lives, where operational burden is stacking up against KPIs, and what would need to become portable before another AI tool is added. That is the Peak Property Performance® Review: a one-building diagnostic of digital control posture and data ownership reality, with specific monthly plays the team could run in the next 90 days once the foundation is right. Your team can even run the diagnostic themselves.

The LLM model is becoming a commodity. The strategic risk is remaining architected for a market in which the model was scarce and the owner’s operating foundation did not have to carry the advantage.

Own your data & digital infrastructure. Operate with strategic foresight. Build for the long game.

Bill