The Multi-Homing Trap

For the better part of a decade, the business model of ride-hailing was the tech equivalent of a controlled bonfire. Giants like Uber, Lyft, Ola, and Grab burned tens of billions of venture capital dollars to subsidise rides, chase growth, and bleed each other into consolidation. Despite commanding some of the most sophisticated matching algorithms ever built, and despite operating networks with genuine two-sided effects, the structural profitability of Mobility as a Service remained, for a long time, genuinely terrible.

By 2024, Uber posted its first full year of operating profit. Lyft is generating positive free cash flow. The narrative shifted: ride-hailing had "finally figured it out." But that narrative, as commonly told, is incomplete, and the incomplete version leads to flawed strategic conclusions for anyone building in this space today.

To understand what actually happened, and what it means for the next generation of MaaS platforms, you have to go further than the subscription story.

The Structural Diagnosis: Multi-Homing and the Commodity of the Match

The core pathology of ride-hailing economics has a precise name: multi-homing. It is what happens when both sides of a two-sided marketplace face near-zero switching costs, making exclusive loyalty to any single platform irrational.

On the demand side, multi-homing is visible and well-documented. Riders open both Uber and Ola and book whichever is cheaper. The cost of downloading a second app is negligible. But here a common analysis makes a meaningful error: the switching cost is not actually zero. Accumulated rating history matters to drivers who can selectively accept higher-rated passengers. Stored payment methods and corporate account integrations create friction. Earned subscription benefits create sunk cost psychology. These are weak switching costs, but weak is not zero, and overstating the problem leads to overstating the solution required.

The supply side is where the analysis usually gets thin, and where the real damage is done. Driver multi-homing is structurally more dangerous than rider multi-homing because it attacks the platform's core quality signal: ETA reliability. When a driver mounts two phones and accepts whoever surges first, match rates decline, cancellations increase, and wait times grow. A rider doesn't abandon Uber because the app is ugly. They abandon it because the car took eleven minutes when Ola promised four. Supply-side multi-homing destroys the product experience in a way that demand-side multi-homing merely erodes the economics.

These are two separate problems. They require two different interventions. Most analyses, and most platform strategies, treat them as one.

Why "Network Effects" Didn't Save Anyone

The standard playbook for platform businesses is to grow fast, achieve density, and let network effects create a winner-take-all outcome. Ride-hailing had genuine network effects: more drivers mean shorter wait times, which attracts more riders, which attracts more drivers. The problem is that these effects are hyperlocal and the scale advantage saturates quickly.

Once a city has enough driver density to deliver a five-minute ETA, adding more drivers doesn't meaningfully improve the rider experience. The network effect plateaus. At that point, both Uber and a well-funded competitor can deliver the same core experience in the same city at roughly the same unit economics. You are back to competing on price.

This is the trap. The platform that wins on network effects reaches a ceiling. The platform that wins on price burns cash indefinitely. Neither outcome produces durable profitability.

What Actually Fixed Uber's Economics

By 2024, Uber had posted its first year of operating profit. The popular explanation is Uber One, the subscription product that bundled ride discounts with Uber Eats free delivery and scaled to over 36 million members by early 2026. This is partially correct. But attributing the profitability turn primarily to subscription psychology misses three equally important structural shifts.

1. Competitive consolidation. Lyft's retreat from aggressive market expansion in 2022–23 gave Uber effective pricing power in most US markets. You cannot separate Uber's margin improvement from the fact that its most credible domestic competitor stopped burning money to undercut it. Profitability came partly from winning the war of attrition, not just from product innovation.

2. Uber Eats reaching margin maturity. Food delivery at scale, where the logistics network is dense enough and restaurant partnerships are locked in, generates meaningfully better unit economics than its land-grab phase. By 2024, Uber Eats was contributing positive adjusted EBITDA in most of its core markets. This is a separate business improving its own economics, not a halo effect from ride-hailing.

3. Advertising as a high-margin revenue layer. Uber's advertising business, serving ads on the app during the booking experience and the ride, generates revenue on infrastructure the platform has already paid for. This is structurally the same model that Amazon used to make AWS's profitability look like it was subsidising retail: a high-margin vertical layered on existing operations. Uber's ad revenue is growing at triple-digit rates year-over-year and is likely the highest-margin business line in the company's portfolio.

4. Then Uber One. The subscription product genuinely works, members do spend 4x more and retain at significantly higher rates. The psychological mechanism is real: a paid sunk cost breaks the habit of opening a competitor app. But subscription lock-in is the reinforcement of competitive position, not its foundation. It works because Uber had already won on the other three vectors. A struggling platform launching a subscription product into a competitive market does not automatically replicate this outcome.

The strategic lesson is not "build a subscription." It is: subscriptions are powerful when they protect margins you have already built; they are poor tools for manufacturing margins you have not.

The Supply-Side Problem Subscriptions Don't Solve

Uber One is a demand-side lock-in mechanism. It converts a multi-homing rider into an exclusive one. It does nothing about the driver with two phones on the dashboard.

Supply-side multi-homing has historically been addressed through three mechanisms, none of them clean.

Surge pricing creates short-term driver commitment during high-demand windows but accelerates rider multi-homing (a rider facing surge will immediately check a competitor). The solution to one problem becomes the cause of the other.

Guaranteed earnings programmes (Uber Pro Shield, Ola's daily earnings guarantee) reduce supply-side churn but are expensive subsidies that erode the margin improvement the subscription product was supposed to create.

Exclusive driver incentives such as bonus schemes, preferred support, vehicle financing partnerships create partial lock-in but are easily matched by a well-funded competitor.

The honest answer is that supply-side multi-homing in human-driver platforms has not been structurally solved. It has been managed, expensively, through incentives. The only genuine solution to driver multi-homing is one that most MaaS platforms are not ready to discuss openly: the removal of the human driver from the equation entirely.

Autonomous Vehicles: The Structural End to the Multi-Homing Problem

Any analysis of MaaS business models in 2026 that does not address autonomous vehicles is analysing the wrong version of the industry.

Waymo is operating fully driverless commercial services in San Francisco, Los Angeles, and Phoenix. Uber has an AV partnership that routes Waymo rides through its demand-side network. Tesla's robotaxi service is in commercial deployment. These are not ten-year horizon events. They are operating today in real markets with real revenue.

Autonomous vehicles eliminate driver multi-homing by definition. A fleet-owned AV is exclusive to one platform. It does not mount a second phone. It does not chase a competitor's surge. The supply side becomes, for the first time, a controllable variable.

This has profound implications for platform economics. A MaaS platform operating an AV fleet controls both sides of the marketplace simultaneously, a structural position that is closer to a utility or a logistics company than to a two-sided platform. The unit economics are fundamentally different: higher fixed capital costs (the vehicle) but dramatically lower variable costs (no driver cut, which currently represents 70–80% of the fare in human-driver markets), and zero driver acquisition or retention spend.

The multi-homing trap is a human-driver problem. It may be a transitional problem, not a permanent one.

The Alternative Architecture: Open Protocols and the Cooperative Model

While the Uber model moves toward bundled subscription lock-in and eventual AV fleet ownership, a genuinely different strategic architecture is emerging, and it represents not a refinement of the incumbent model but a philosophical rejection of it.

Namma Yatri, built on top of India's Open Network for Digital Commerce (ONDC), is the most discussed example. Rather than operating as a centralised toll-booth that extracts a commission from every transaction, Namma Yatri charges drivers a flat daily or monthly subscription fee and passes the full fare to the driver. The matching protocol is open; the platform earns from the subscription, not the transaction.

This is a genuinely interesting model. But before accepting it as the template for the next generation, the critical weaknesses deserve equal attention.

The flat subscription has a cold-start problem for drivers. A commission model charges drivers only when they earn. A flat fee charges them regardless. For a driver with uncertain daily income, which describes most gig economy workers, paying a fixed cost upfront is financially riskier than paying a variable commission on earnings. Namma Yatri has faced real driver acquisition friction precisely because of this dynamic. The model works better as income certainty increases, not as a universal replacement for commission structures.

ONDC adoption in mobility remains early-stage. ONDC transaction volumes in ride-hailing are a fraction of private platform volumes. The open protocol promise, that commoditising the network layer forces competition to move up to service quality, is theoretically sound but operationally unproven at scale.

The cooperative model has a capitulation problem. When a centralised competitor decides to subsidise fares in a market where Namma Yatri operates, the cooperative has no capital reserve to match it. The absence of a take-rate also means the absence of a war chest.

This is a Blue Ocean move, but it competes in a different ocean from Uber. It is not a superior version of Uber's strategy. It is a different bet entirely: that the marginal urban commuter in a price-sensitive market will route around incumbent platforms once a good-enough, lower-cost open alternative exists.

Whether that bet pays off depends entirely on whether ONDC achieves sufficient liquidity to make the matching quality competitive.

Two Strategies, Not One Playbook

The deepest inconsistency in most MaaS strategic analysis, including earlier versions of this thinking, is treating Uber's subscription path and Namma Yatri's open protocol path as a unified "next generation playbook." They are not. They are opposite strategic bets with opposite risk profiles.

The right strategic choice is a function of the market you are building in, the capital available, and your view on AV timeline. In a high-GDP urban market with AV deployment underway, the Uber architecture is increasingly defensible. In a price-sensitive emerging market with ONDC infrastructure developing, the cooperative model is worth serious consideration, but with clear eyes about its constraints.

What This Means If You Are Building Today

If you are a MaaS startup in 2025, the legacy aggregator game, match supply and demand, take a commission, grow fast, is almost certainly a losing strategy. The incumbents have density, brand, and increasingly, subscriptions and AV pipelines. You cannot outspend them.

The viable paths are narrower and more specific.

Vertical specialisation: owning a trip category that incumbents under-serve. Airport transfers, executive travel, medical patient transport, last-mile logistics. Categories where reliability and compliance matter more than price, where multi-homing is lower because quality variance is unacceptable, and where you can build direct relationships outside the commodity matching market.

Supply-side ownership: if AV economics are your bet, the platform advantage shifts dramatically toward whoever controls the fleet. This is asset-heavy and capital-intensive, but it is the one structural solution to driver multi-homing.

Open protocol infrastructure: building on ONDC or similar public infrastructure to deliver lower costs in price-sensitive markets, with subscription revenue replacing transaction revenue. Viable in the right geography with the right regulatory environment, and requiring patient capital that accepts slower growth.

Bundled adjacent services: not ride-hailing plus food delivery (that ship has sailed), but ride-hailing embedded into an existing high-frequency use case: a corporate travel platform, a healthcare system, a university campus network. Captive distribution channels where multi-homing behaviour is suppressed by the embedding context.

The common thread is that none of these paths compete with Uber on Uber's terms. The multi-homing trap was built by trying to be a better generic ride-hailing app. Escaping it requires being a genuinely different one.

Conclusion: The Trap Was Never Just About Price

The multi-homing trap is real. But it is not primarily a pricing problem, and the solution is not primarily a subscription product. It is a structural problem that arises when a platform's core value proposition is a commodity, its supply side has no reason to be exclusive, and its network effects plateau before they create genuine defensibility.

Uber escaped it through a combination of competitive consolidation, margin maturity in adjacent verticals, a high-margin advertising layer, and a subscription product that locked in its best customers. That is a five-year story compressed into a single neat narrative.

The next generation of MaaS platforms will escape it differently: through vertical focus, supply-side control, open protocol cost structures, or AV-driven elimination of the driver variable entirely. These are not refinements of the Uber model. They are responses to the fact that the Uber model has already been built, and built well.

The question is not how to do what Uber did. It is what comes after it.

Edited with Claude. Image generated with Gemini.