In March 2000, the NASDAQ peaked at 5,048. By October 2002, it had lost 78 percent of its value. The companies that disappeared weren’t frauds. Most of them had real products, real customers, and real revenue. What they didn’t have was a path to profitability, a business model that made sense at normal capital costs, or a balance sheet that could survive a funding drought.
Their customers found out the hard way.
Twenty-five years later, a hotel owner in Palm Beach or Montecito is sitting in a demo. The product looks extraordinary. AI-powered personalization. Dynamic pricing intelligence. Predictive demand modeling. The founders are sharp. The deck is polished. The pilot terms are generous.
There is no reason to say no.
There is also no reason to ask the question that actually matters.
The Numbers Behind the Demo
The scale of AI infrastructure investment in 2026 is genuinely unprecedented. Alphabet, Microsoft, Meta, and Amazon are projected to spend nearly $700 billion combined this year on AI-related capital expenditure, a figure representing more than triple their combined spending from just two years ago. Goldman Sachs projects total hyperscaler capex from 2025 through 2027 will reach $1.15 trillion. These are the pipes. The foundational layer. The infrastructure that makes AI products possible.
Very little of that capital is flowing to the company that just booked a demo at your property.
AI funding hit record levels in 2024 and 2025, but the composition matters enormously. A disproportionate share went to foundation model developers – OpenAI, Anthropic, Google DeepMind – and infrastructure plays. The application layer, companies building products on top of someone else’s model, has faced a different reality.
Here is the economic reality most hotel owners are never shown. For every dollar a hospitality AI vendor collects from you, a significant portion flows directly back out to their model provider as inference cost: the compute tax. This is not a marginal line item. It is a structural constraint on gross margin that traditional software never carried. Some vendors manage it through caching, fine-tuning, or tiered pricing architectures. Many do not, or cannot at their current scale. The result is a vendor landscape where gross margin pressure is widespread and unevenly managed, and where the difference between a vendor that survives and one that doesn’t often comes down to financial architecture the demo was never designed to reveal.
The market data reflects the broader pressure. By early 2026, software price-to-sales ratios had compressed from 9x to 6x, levels not seen since the mid-2010s, as institutional investors rotated away from application software toward semiconductor and energy infrastructure, according to market analysts tracking the selloff. MIT’s NANDA initiative, drawing on 150 leadership interviews and analysis of 300 public AI deployments, found that 95 percent of generative AI pilots deliver no measurable impact on profit and loss. McKinsey’s State of AI 2025 identified only 6 percent of companies as genuine AI high performers, those achieving meaningful EBIT impact from AI investment, against a field where the vast majority reported little to no financial return.
The market is not collapsing uniformly. Some application-layer vendors have built profitable businesses with usage-capped pricing, proprietary workflow value, or hybrid architectures that create real switching costs. The infrastructure layer is strengthening. A cohort of AI vendors with genuine proprietary data and defensible economics will survive this cycle and consolidate. What is happening is a bifurcation between companies with durable unit economics and companies that were built on venture subsidy, running a product layer over someone else’s model, waiting for a funding round that is becoming harder to close.
The hotel industry is being actively pitched by both categories. Most owners have no reliable way to tell the difference from a demo alone.
What Actually Happens When an AI Vendor Fails
This is the question nobody in a hotel sales cycle is asking. It deserves a direct answer, with the appropriate caveat: outcomes vary based on contract language, jurisdiction, and deal structure. That variance is precisely why the question matters before the contract is signed, not after.
When a VC-backed software company runs out of runway, the sequence typically begins long before the lights go out. Support tickets go unanswered first. Integration updates slow and then stop. The engineering team that built your custom configuration has already moved to a better-capitalized company. The platform degrades quietly while the demo environment stays pristine. By the time a revenue manager notices something is wrong, the vendor is already in conversations with acquirers.
The pattern is not theoretical. A personalization layer that appeared in multiple luxury property RFPs in 2024 was acquired eighteen months after launch. The roadmap was discontinued. Guest behavioral data, including preference profiles, booking intent signals, and spend patterns accumulated across hundreds of stays, was locked in a deprecated system the acquiring company had no interest in maintaining. The hotels involved had no repatriation clause. They had no exit right. They had a product that no longer existed and data they could not reach.
But outright failure is not the most common or the most dangerous outcome. Acquisition is. And acquisition introduces a risk that is almost never discussed in hotel technology procurement: strategic misalignment.
When a luxury-positioned AI vendor is acquired by a mid-scale volume player, or absorbed into a private equity portfolio optimizing for margin rather than product quality, the roadmap frequently shifts away from high-touch guest relationship features toward functionality that serves the acquiring company’s larger, less demanding customer base. The account managers who understood your property, your guest profile, and your service standards are often among the first to leave. What remains is a platform that technically still functions, pointed in a direction that no longer serves your guests or your brand.
Distressed acquisitions are not rare. Enterprise SaaS M&A hit $83.7 billion across 245 deals in Q4 2025 alone, according to PitchBook. The acquirers are primarily large platform companies and private equity roll-ups buying assets at compressed valuations. What transfers in those transactions, including your guest behavioral data, preference profiles, and demand signals, depends entirely on language that was almost certainly not in the contract you signed.
In a standard asset sale, the acquiring entity does not automatically inherit your contractual protections with the seller. Your data rights, your SLAs, your pricing commitments, your integration guarantees must be renegotiated, or they are not honored. Whether the behavioral intelligence your guests generated inside that platform can be cleanly repatriated to your property, or whether it transfers to a new owner whose interests may directly conflict with yours, is a function of contract language most hotel buyers never negotiate.
The hotel industry has already lived one version of this. OTAs absorbed guest relationship data from properties that believed they were simply listing inventory. The mechanism was different. The outcome, demand intelligence owned by an intermediary used to serve your guests to your competitors, was not.
The Hype Cycle Has Already Moved On
The AI market narrative has shifted from chatbot AI to agentic AI in under two years. Investors are now funding autonomous agent frameworks and outcome-based enterprise systems. The personalization and revenue management tools that dominated hospitality technology conversations in 2024 occupy a different position in the investment thesis today, not obsolete, but no longer the category attracting the capital and attention they commanded eighteen months ago.
Think of it this way. First-generation AI tools were a better calculator, faster and more accurate, but fundamentally still a tool operated by a human. Agentic AI is positioned as a digital employee, autonomous, self-directing, capable of executing multi-step workflows without human intervention. Investors have moved their attention and their capital toward the employee-makers. The calculator-makers are operating on what they raised.
A hotel that signed a two-year contract with an AI personalization vendor in 2024 may find itself in 2026 holding a contract with a company operating on reduced staff, servicing a product that no longer sits at the center of its investors’ thesis. The product still runs. The integrations still function. The support team is smaller. The roadmap has slowed. That trajectory is already visible across the broader application-layer market, in a hospitality vendor landscape that lacks the public market transparency that signals these problems earlier in enterprise software companies.
Not All AI Vendors Carry the Same Risk
This is where the argument requires precision, because indiscriminate vendor skepticism is neither useful nor accurate.
Some AI vendors will survive this cycle and become genuine infrastructure. The distinguishing characteristics are identifiable. Proprietary training data that cannot be replicated by a competitor spinning up on the same foundation model. Embedded workflow integration with real switching costs. Enterprise contract structures with installed bases that predate the current hype cycle. Unit economics that do not depend on inference-cost subsidies from their model provider.
A revenue management platform with fifteen years of hospitality-specific pricing data and deep PMS integration is a fundamentally different risk profile than a guest-facing personalization layer built in eighteen months on top of a commercial API. A back-office automation tool from a well-capitalized incumbent with hundreds of millions in annual recurring revenue is a different conversation than a Series A chatbot vendor with twelve months of runway.
The problem is that in the demo environment, both categories are indistinguishable. Both have polished interfaces, sharp founders, and compelling ROI projections. The difference is entirely in the financial architecture underneath: the compute dependency, the investor conviction, the contractual protections, the data portability. None of which the demo was designed to reveal.
Understanding what the real AI risk looks like for independent hotels requires looking past the product and into the business model underneath it.
What to Ask Before You Sign
The hospitality industry’s instinct in technology procurement has always been to evaluate the product. Does it work? Does it integrate? What does the ROI model look like? These are not wrong questions. In 2026, they are insufficient ones.
The questions that belong in every AI vendor evaluation are different in kind. They are the questions a distressed-debt investor would ask, not the ones a hotel operations team typically runs on a software vendor. They require legal counsel in the room before the contract is signed, not after a problem surfaces.
On financial health
What is the company’s current runway and when was it last independently verified? Who are the lead investors and have they participated in subsequent funding rounds, or have they marked the position down? What is the burn multiple: for every dollar of new revenue added, how many dollars is the company spending? What percentage of gross revenue flows to the model provider as inference cost, and what happens to the product economics if that provider changes its pricing?
On data rights
Who owns the guest behavioral data generated inside the platform, and is that answer in the contract in plain language, not buried in terms of service? What specific rights does the property retain to export, port, and delete that data? What happens to those rights in the event of an acquisition, an asset sale, or a wind-down? Is there a data escrow provision that survives a change of ownership?
On operational continuity
What are the contractual SLAs for platform continuity and support response, and what are the remedies if those SLAs are not met? In the event of acquisition, does the hotel retain the right to exit the contract without penalty? Is there a migration assistance provision requiring the vendor to support transition to an alternative system?
On model defensibility
Is the AI capability genuinely proprietary, built on hospitality-specific training data that competitors cannot replicate, or is it a product layer over a commercial foundation model that any well-funded team could rebuild in a single development cycle?
The one clause that matters most
Before signing any AI vendor agreement, insist on a change-of-control termination right: the explicit contractual right to exit the agreement and receive full data extraction in the event of acquisition, insolvency, or material change in service ownership. Full data extraction should be defined in the contract to include raw behavioral signals, enriched guest profiles, and any model outputs generated from your guests. A vendor that refuses to negotiate this clause has told you something important about how it views your interests relative to its own.
The Posture That Protects You
The argument here is not that hotels should reject AI. The infrastructure is real, the capability is advancing, and independent luxury properties that ignore AI’s growing role in shaping guest preference over the next three years will face structural disadvantages that are difficult to reverse.
The argument is that the procurement standard needs to match the risk profile. Signing a multi-year AI vendor contract without financial due diligence is the hospitality equivalent of listing on an OTA without reading the commission structure. The product looked useful. The dependency was in the fine print. The cost compounded quietly.
The hotels that navigate this period intact will be the ones that treated AI vendor selection the way they treat any capital commitment with a multi-year horizon, with scrutiny on financial durability, with contractual protection on data ownership and continuity, and with a clear answer to the question: if this vendor is acquired or degraded in eighteen months, what have we lost, and can we get it back?
Not all vendors will fail. Not all acquisitions are damaging. Procurement discipline is not the same as vendor rejection. It is the difference between a relationship that serves the property and one that quietly transfers its most valuable long-term asset to someone else’s balance sheet.
The industry has been through this before. The lesson was expensive. The terms were in the fine print. The dependency was called a distribution strategy.
Sources
MIT NANDA Initiative, The GenAI Divide: State of AI in Business 2025 (July 2025), based on 150 leadership interviews, 350 employee surveys, and analysis of 300 public AI deployments.
McKinsey & Company, The State of AI 2025 (November 2025), annual global survey on AI adoption and financial impact.
Goldman Sachs, hyperscaler capex projections (2025-2027), as reported in market analysis and CNBC, February 2026.
PitchBook, Q4 2025 Enterprise SaaS M&A Review, $83.7 billion across 245 deals.
Software price-to-sales multiple compression (9x to 6x), as reported by institutional equity analysts and financial market reporting, February 2026.
CNBC, “Tech AI spending approaches $700 billion in 2026,” February 6, 2026.
