Here's a fact that should make every AI travel startup founder uncomfortable: there are dozens of AI travel apps on the market, and almost none of them are profitable. Meanwhile, the companies building MCP infrastructure — the boring, unsexy, behind-the-scenes tools that power AI travel booking — are already generating revenue.
This isn't a coincidence. It's a structural pattern. And if you're building or investing in AI travel, you need to understand it, because it determines who wins and who loses in this market.
Let me break it down.
The AI Travel App Graveyard: Where the Money Goes
Take a quick tour of the AI travel app landscape. There are apps that plan itineraries. Apps that find flight deals. Apps that book hotels through chat. Apps that act as "travel concierges." Apps that generate packing lists. Apps that translate foreign languages. Apps that do all of the above.
Most of them have raised seed or Series A funding. Most have impressive user growth numbers. Almost none are profitable. Here's why:
Customer acquisition cost (CAC) is brutal. The travel market is dominated by OTAs that spend billions on marketing. Booking.com alone spent over $5 billion on marketing in 2024. An AI travel startup trying to acquire users is competing with companies that can outspend them 100-to-1. The average CAC for a travel app user is $30–$80. For a user who might book one $200 hotel room per year, generating $20 in commission, the math doesn't work.
Retention is terrible. People don't travel every day. They book a trip, use the app for a week, and then don't open it again for 6–12 months. By the time they travel again, they've forgotten the app exists, or they've seen an ad for a competitor. Monthly active users (MAU) spike around booking periods and crash in between.
Monetization is indirect. Most AI travel apps make money through affiliate commissions — they send users to OTAs and get a cut of the booking. This means they're not really a travel business; they're a marketing channel for OTAs. And OTAs know it — they can change commission rates, reduce cookie windows, or launch competing products at any time. The app's revenue is entirely dependent on the goodwill of companies that see them as either customers or threats.
The AI isn't defensible. Every AI travel app uses the same underlying models (GPT, Claude, Gemini) and the same inventory sources (OTA APIs). The "AI" in "AI travel app" is a thin layer on top of commoditized infrastructure. There's no moat. A competitor can build the same app in a weekend.
Put it all together: high CAC, low retention, indirect monetization, no moat. This is a recipe for burning investor money until the funding runs out. And that's exactly what's happening.
The MCP Money Machine: Where the Profit Is
Now look at the other side of the market: MCP infrastructure. These are the companies building the servers, protocols, and tools that let AI agents connect to real-world services like hotel booking. They're not building consumer apps. They're building the pipes.
And they're making money. Here's why:
Customers are developers and businesses, not consumers. MCP infrastructure companies sell to developers who build AI agents, and to businesses that want their services accessible through AI. These customers have budgets, make purchasing decisions rationally, and don't require Super Bowl ads to acquire. CAC is a fraction of consumer app CAC.
Revenue is recurring and transaction-based. MCP servers charge either a monthly subscription, a per-transaction fee, or a freemium model with paid tiers. Every time an AI agent books a hotel through the MCP server, the server generates revenue. This is direct monetization — no affiliate links, no cookie windows, no dependence on OTA goodwill.
Retention is structural. Once a developer integrates an MCP server into their AI agent, switching costs are high. The agent is built around the MCP server's tools, data formats, and APIs. Moving to a competitor requires rewriting integration code, retesting, and risking breakage. Developers stay. Businesses stay. Revenue compounds.
The infrastructure is the moat. A hotel MCP server that has real-time inventory from 2 million+ hotels, reliable pricing APIs, and customer support is not something a competitor can build in a weekend. It requires supplier relationships, technical infrastructure, compliance certifications, and operational scale. This is a real moat.
RollingGo is the case study. It aggregates 2 million+ hotels from 500+ suppliers across 200+ countries, including 110,000+ directly contracted properties. Its v1.0 exposes three tools (searchHotels, getHotelDetail, getHotelSearchTags) through a single MCP endpoint. The business model is deliberately counterintuitive: API keys are free, calls are unlimited, approval is instant. No credit card required. No rate limits. No per-query fees.
How does that make money? Four identity tiers capture value at different levels:
- Personal — free, for individual AI tool users (builds ecosystem adoption)
- Developer — free, with Partner Center access for traffic analytics (locks in integration)
-
KOL/Affiliate — free, with tracked
bookingUrllinks that earn commission on every completed booking (revenue share without infrastructure cost) - Enterprise — paid, with team management, budget controls, audit trails, and volume pricing (the real revenue engine)
The free tier isn't a charity — it's a distribution strategy. Every developer who connects RollingGo to their agent is another distribution channel. Every KOL who uses the affiliate link is another salesperson. The enterprise tier captures the value when those channels scale. This is the AWS playbook: give away the infrastructure to developers, then monetize the enterprise demand that follows.
The connection takes five minutes:
{
"mcpServers": {
"RollingGo-Hotel": {
"url": "https://mcp.rollinggo.ai/mcp",
"type": "streamable-http",
"headers": {
"Authorization": "Bearer mcp_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"
}
}
}
}
The result: MCP infrastructure companies have low CAC, high retention, direct monetization, and real moats. They're profitable or on a clear path to profitability. Meanwhile, AI travel apps burn cash chasing consumers who don't come back.
The Numbers: A Side-by-Side Comparison
Let me make this concrete with a hypothetical but realistic comparison:
| Metric | AI Travel App | MCP Infrastructure |
|---|---|---|
| Customer type | Consumer traveler | Developer / business |
| CAC | $40 | $200 (but 1 customer = 1000s of end users) |
| Retention (12-month) | 15% | 85% |
| Revenue per customer | $20/year (affiliate) | $1,200/year (subscription + transactions) |
| LTV:CAC | 0.75:1 (losing money) | 5:1 (profitable) |
| Time to profitability | 36+ months (if ever) | 12–18 months |
| Moat | None | Inventory + relationships + scale |
The AI travel app spends $40 to acquire a user who generates $20 in lifetime revenue. It loses $20 per user. The MCP infrastructure company spends $200 to acquire a developer who generates $1,200 in annual recurring revenue. It makes $1,000 per customer per year, and that customer stays for years.
This isn't a close call. The infrastructure business is fundamentally better.
Why This Pattern Keeps Repeating
This isn't the first time we've seen this pattern. In every major technology shift, the companies that make money are the ones that build infrastructure, not the ones that build consumer applications on top of it:
- Gold rush: Levi Strauss sold jeans to prospectors. The prospectors mostly went broke. Levi Strauss became wealthy.
- Internet boom: Cisco sold routers. Akamai sold CDN. AWS sold cloud computing. The consumer apps built on top of them (Pets.com, Webvan) went bankrupt. The infrastructure companies thrived.
- Mobile boom: Apple sold iPhones. Google sold Android. App developers made pennies. The platform owners made billions.
- AI boom: NVIDIA sells GPUs. OpenAI sells models. MCP servers sell infrastructure. The AI apps built on top are fighting for scraps.
The pattern is clear and consistent: in a technology gold rush, sell picks and shovels. Don't dig for gold.
In AI travel, the "picks and shovels" are MCP infrastructure — the servers, protocols, and tools that let AI agents connect to real travel inventory and complete bookings. The "gold" is the consumer travel booking, which everyone is chasing and almost no one is capturing profitably.
What This Means for Founders and Investors
If you're a founder building an AI travel app, you have three options:
Option 1: Keep building the app and hope for the best. Maybe you'll be the one consumer app that breaks through. The odds are against you, but it's possible. You'll need a huge marketing budget, a viral growth loop, or a unique value proposition that no one can copy. Most apps don't have any of these.
Option 2: Pivot to infrastructure. Stop building a consumer app and start building the infrastructure that powers AI travel. This is harder technically but easier commercially. You'll sell to developers and businesses, not consumers. You'll have recurring revenue, high retention, and a real moat. This is the path that's already generating profit.
Option 3: Hybrid model. Build a consumer app that's powered by your own MCP infrastructure. The app is a demonstration of the infrastructure's capabilities (and a source of direct booking revenue), while the infrastructure is sold to other developers and businesses. This is the AWS model — Amazon built AWS to power its own retail business, then sold it to everyone else.
If you're an investor, the math is simple: infrastructure investments have better risk-adjusted returns than consumer app investments. You're not betting on a single app's viral growth. You're betting on a structural shift in how AI agents connect to real-world services. That shift is happening regardless of which consumer app wins.
The Counterargument: "But Someone Has to Build the Apps!"
I know what you're thinking. "If everyone builds infrastructure and no one builds apps, there's no demand for the infrastructure." And you're right — there needs to be a healthy ecosystem of AI travel apps to drive demand for MCP infrastructure.
But here's the thing: the apps will be built. They'll be built by thousands of developers, small teams, and niche players. They won't be giant, venture-backed consumer apps. They'll be specialized agents — a ski trip planner, a business travel optimizer, a family vacation coordinator — each serving a specific niche. These apps will be profitable at small scale because they have low CAC (niche targeting) and high retention (serving a specific use case).
The infrastructure providers will serve all of them. That's the beauty of the infrastructure model: you don't need to bet on which app wins. You win regardless, because every app needs your infrastructure.
The Bottom Line
The AI travel market is undergoing a massive shift. Consumer-facing AI travel apps are burning investor money with no clear path to profitability. MCP infrastructure companies are generating real revenue with strong unit economics. The pattern is as old as technology shifts: infrastructure wins, applications fight for scraps.
If you're building in this space, ask yourself: am I selling picks and shovels, or am I digging for gold? The answer determines whether you'll be profitable in 18 months or out of business in 36.
The gold rush is on. Everyone's chasing the consumer. The smart money is building the rails.
The hotel MCP infrastructure that's powering the next generation of AI travel apps is open-source and free to start using: RollingGo Hotel MCP on GitHub.
Top comments (0)