Rented or owned? The market starts separating the wheat from the chaff
If your competitive edge depends entirely on a model someone else owns, do you actually have a competitive edge at all? A look at what a century of technology history says about who really wins.
Text by Martti Asikainen, 13.7.2026 | Photo by Adobe Stock Photos
In the first phase of the AI boom, money flowed into almost anything with AI attached. It was enough to bolt a chat window onto an old product for investors to get excited. Sometimes even the mere intention was enough. That era is now over. Investors have begun asking uncomfortable questions, such as what happens if an AI model vendor rolls out the same feature into its own product tomorrow, or sets up a rival company altogether (e.g. OpenAI 2026). Or what happens if the price of tokens for generative AI models rises tenfold next year?
If your entire competitive advantage rests on a readily available external model, then it is hard to defend (Barney 1991). The company does not own the model — it is using a resource whose pricing, availability and direction of development it cannot control. That is why competitive advantage is likely to be found less in ready-made models and more in things an external AI vendor cannot copy or conjure up overnight: a company’s own data, distribution channels, processes and organisational capabilities (Teece 2018).
A cynic might even argue that if your competitive advantage rests entirely on a model someone else owns, then you do not actually own a competitive advantage at all. You own a licence to use a tool whose owner can reprice it, copy it, or even discontinue it with a single update. This is part of why the conversation is increasingly turning towards data, distribution and workflows — the things an AI vendor cannot simply switch off with the click of a button.
Technology platforms and the logic of dependency
Every major technological shift creates a new dependency. This is not the fault of individual companies but a recurring structural pattern (David 1985; Arthur 1989). In the mid-19th-century United States, factory location, raw-material sourcing and distribution all became dependent on the routes and timetables of railway companies. Firms that managed to locate themselves at the nodes of the rail network gained a significant competitive edge — one that arose not from their own efficiency, but from their relationship to their location and to the railway company (Atack et al. 2008; Donaldson & Hornbeck 2016).
The same pattern repeated during the so-called Wintel era of the 1980s and 1990s, when software houses were no longer simply competing with one another, but found their success increasingly dependent on how well they could adapt to the terms set by Microsoft’s Windows operating system and Intel’s processor architecture. Two companies formed the backbone of an entire industry (Bresnahan & Greenstein 1999; Gawer & Cusumano 2002).
Especially within digital ecosystems, a company’s position increasingly depends on its relationships with other players (Jacobides, Cennamo & Gawer 2018). Cloud services, for instance, reduced the need for companies to run their own data centres, while simultaneously shifting control to the cloud provider. Social media lowered marketing costs, but made growth hostage to an algorithm.
Search engines brought free visibility, but made online business so precarious that everything was one algorithm update away from collapse. Apple’s App Store and Google Play repeated the same pattern: developers’ success no longer depended solely on the quality of their product, but also on the platform’s rules, visibility algorithms and revenue cuts.
AI does not break this pattern — it reinforces it.
In particular, the development of foundation models has deepened dependency on a small handful of large model vendors (Bommasani et al. 2021). The more a business leans on continuous model calls, the greater the share of its cost structure that sits in someone else’s hands (Bommasani et al. 2021). That does not make the model a bad choice — things are rarely so black and white — but it does create the same kind of dependency as any other outsourced resource. Many only notice this once usage volumes and the bill start climbing in lockstep.
Free today, expensive tomorrow
AI labs are not turning a profit at today’s prices. They operate on a model of continuous funding rounds, in which each round of investment buys the next year of below-cost access, which in turn grows the user base and justifies the next round at an even higher valuation. The structure is a familiar one. Amazon did it with e-commerce. Uber did it with taxis. Spotify did it with music. AI labs are now doing it with intelligence itself (Asikainen & Masala 2026).
The same four-stage pattern recurs in every major technological shift. The first is the land-grab phase, in which we currently find ourselves: labs price tokens below their real cost and flood the market with free tiers and cheap APIs. The goal is simple — developers build on top of the platform, businesses integrate the interface, and ecosystems form around the model. Lock-in accumulates quietly, almost unnoticed.
The second stage is consolidation: smaller players are bought out or simply disappear from the market. Those who survive soon face investor pressure to demonstrate profitability, which leads to the third stage: repricing. Prices become anchored to real costs plus margin, and free tiers shrink or vanish altogether. Companies that have built their business on AI-driven workflows or assistants then face cost items they never budgeted for — and are forced either to raise their prices sharply or to abandon the business as unprofitable (ibid. 2026).
The fourth stage is the fundamentals stage. At this point, teams’ effectiveness is no longer measured by how fast they can build with AI, but by how cost-efficiently they do so. Token efficiency becomes the competitive advantage that people currently imagine springs from AI use alone. The first signs of this shift are already visible. In April 2026, GitHub announced it would move its Copilot tool to usage-based billing from June onwards. Since the change, developers have reported that a single agentic working session can already consume more than an entire month’s subscription fee (GitHub Blog, 2026).
At this point, many want to dispute the direction of travel: hardware costs are falling, model efficiency is improving, and competition may well keep prices low for longer than we expect. That, too, is true. But current pricing is unlikely to reflect a lasting market equilibrium — venture-subsidised, below-cost access is a temporary strategy, not a permanent state of affairs.
Do you own your future
Many readers might conclude from this piece that AI is best used quietly in the background rather than sold as a feature. It’s a tempting thought, but it is not a neutral strategy — it is simply a different kind of risk.
A company that builds its entire productivity on AI-assisted development but invests nothing at all in its own expertise, data pipelines or understanding of its processes is just as dependent as a company that plasters its externally-sourced AI across its homepage in giant letters. The only difference is that no outsider sees it — until it’s too late.
The real lesson of technological history is not that technology should be hidden away, but that alongside the technology itself, one must also understand the risks and challenges bound up with how it is used (David 1990; Brynjolfsson et al. 2021). Electricity transformed factories, but the competitive advantage did not come from the power cable — it came from the redesigned production process. The internet transformed commerce, but the competitive advantage did not come from the internet connection — it came from the online stores and business models built on top of it (Teece 2018).
Much the same logic applies to AI. It hardly matters where in the production line AI becomes visible; what matters far more is how deeply it is woven into how the organisation actually operates (Vial 2019). In practice, this means investing in three things that no model vendor can take away from you: the quality and availability of your own data, an understanding of your processes deep enough to rebuild them using a different model, and your organisation’s own ability to judge where AI delivers benefit and where it does not (Teece 2018).
This understanding is especially important because AI’s benefits are not spread evenly across all tasks. In one field experiment involving 758 consultants using AI in their work, AI significantly improved performance on tasks that fell within its zone of competence, but clearly worsened outcomes on a more complex task that exceeded its capabilities (Dell’Acqua et al. 2026).
An organisation that fails to recognise this boundary risks applying AI precisely to the tasks where it does the most damage. It is a great pity, then, that AI discourse today is dominated by talk of AI’s supposed omnipotence and capabilities, while far too little is said about what using it actually costs over the long run, in money, in expertise, and in strategic room for manoeuvre.
Next time your organisation discusses its AI strategy, ask these two questions in succession:
- Are we building our business on top of AI, or are we building it with the help of AI?
- If an AI provider raises its prices tomorrow, or changes its terms of service, what will we have left?
The difference between these two questions might seem like a small turn of phrase, but in practice it can be a question worth billions of euros — whether your company owns its future, or merely rents it, one month at a time.
Author
References
Asikainen, M. & Masala, S. (2026, 20 May). AI is currently almost free. That’s exactly why we’re making our most expensive mistakes right now. Finnish AI Region.
Barney, J. (1991). Firm Resources and Sustained Competitive Advantage. Journal of Management.
Teece, D. J. (2018). Business Models and Dynamic Capabilities. Long Range Planning.
Bommasani, R., et al. (2021). On the Opportunities and Risks of Foundation Models. Stanford CRFM.
Vial, G. (2019). Understanding Digital Transformation. Journal of Strategic Information Systems.
Jacobides, M. G., Cennamo, C., & Gawer, A. (2018). Towards a Theory of Ecosystems. Strategic Management Journal, 39(8), 2255–2276.
Atack, J., Haines, M. R., & Margo, R. A. (2008). Railroads and the Rise of the Factory: Evidence for the United States, 1850–70. NBER Working Paper No. 14410. https://doi.org/10.11126/stanford/9780804771856.003.0007
Donaldson, D., & Hornbeck, R. (2016). Railroads and American Economic Growth: A “Market Access” Approach. The Quarterly Journal of Economics, 131(2), 799–858. https://doi.org/10.1093/qje/qjw002
Bresnahan, T. F., & Greenstein, S. (1999). Technological Competition and the Structure of the Computer Industry. Journal of Industrial Economics, 47(1), 1–40.
Gawer, A., & Cusumano, M. A. (2002). Platform Leadership: How Intel, Microsoft, and Cisco Drive Industry Innovation. Harvard Business School Press.
David, P. A. (1985). Clio and the Economics of QWERTY. American Economic Review, 75(2).
Arthur, W. B. (1989). Competing Technologies, Increasing Returns, and Lock-In by Historical Events. Economic Journal, 99(394).
David, P. A. (1990). The Dynamo and the Computer: An Historical Perspective on the Modern Productivity Paradox. American Economic Review, 80(2), 355–361.
Brynjolfsson, E., Rock, D., & Syverson, C. (2021). The Productivity J-Curve: How Intangibles Complement General Purpose Technologies. American Economic Journal: Macroeconomics.
Dell’Acqua, Fabrizio, Edward McFowland III, Ethan Mollick, Hila Lifshitz, Katherine C. Kellogg, Saran Rajendran, Lisa Krayer, François Candelon & Karim R. Lakhani (2026). Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of Artificial Intelligence on Knowledge Worker Productivity and Quality. Organization Science, 37(2), 403–423. DOI: 10.1287/orsc.2025.21838.
OpenAI. (2026, 11 May). OpenAI launches the OpenAI Deployment Company to help businesses build around intelligence. Published on OpenAI’s website.
GitHub. (2026, 27 April). GitHub Copilot is moving to usage-based billing. The GitHub Blog. Accessed 1 July 2026.
