OPINION | Public options may turn global AI reasoning into a mixed oligopoly market

India may launch its public open AI reasoning model soon.

Two launches in recent weeks could prove to be the most consequential developments in artificial intelligence this year, and they did not come from Silicon Valley. These developments also raise the question of how other economies like India and Brazil will consider the AI reasoning race.

In Abu Dhabi, MBZUAI and G42 unveiled K2 Think, a 32-billion-parameter reasoning model claiming parity with much larger rivals on logic and mathematical benchmarks, released as open weights. And in Switzerland, a consortium led by ETH Zurich, EPFL, and CSCS launched Apertus, a fully open reasoning model family whose weights, code, documentation, and evaluation tooling are already proliferating across major platforms.

These are not just new artefacts on GitHub. They are public entrants into what has so far been an almost exclusively private market.

Their arrival will change the price–quantity equilibrium of the reasoning model market — in ways that economists studying mixed oligopolies would find entirely familiar.

Price elasticity segments a market

And there’s something here from West Bengal, India, that can be a lesson for AI and Silicon Valley. A recent paper of ours examines this in the Journal of Economic Behavior and Organization and explored what happened when government-backed generic dispensaries entered the pharmaceutical market in West Bengal, India. It offered a natural experiment: a public provider suddenly appeared alongside entrenched private sellers. The result was not uniform price erosion. Instead, the private market segmented.

In categories where demand was highly elastic (price-sensitive), private firms slashed prices to retain customers who might switch to the public alternative. In categories where demand was more inelastic (quality-sensitive, high perceived risk), private firms raised prices. They knew the customers who remained valued assurance and would not defect on price alone.

The entry of a public provider created a separating equilibrium. The public sector absorbed the price-sensitive mass market; private firms repositioned as premium suppliers to inelastic buyers. Average prices concealed a bifurcation: some prices collapsed; others rose.

Cost of commoditised reasoning will drop

This pattern is well established in the literature on mixed oligopolies. Such markets are where public and private producers coexist, serving different demand segments and reshaping each other’s strategies. And it is about to replay in AI with the entry of K2THINK and Apertus from UAE and Switzerland as we see it.

Exactly how will the new demand curve for reasoning models pan out then build on our findings? For the first time, credible open-weight reasoning models exist as outside options. This shifts the residual demand faced by private LLM providers.

Elastic users, start-ups, schools, municipal agencies, can now substitute towards zero-marginal-cost open models. Private firms will see the demand curve for commoditised reasoning (summarisation, classification, batch analytics) rotate downwards.

In textbook terms, the marginal revenue they can extract from these users has collapsed. To remain competitive, they must lower per-token prices or launch efficiency-optimised SKUs, driving them closer to marginal cost.

Premium segment may see enhancement in product features

Inelastic users, enterprises, defense contractors, health systems are different. They value assurance: safety guardrails, service-level agreements, indemnities, tool integration, continuous fine-tuning. They are less price-sensitive and more risk-sensitive. Public entrants will not fully displace their demand. Instead, private firms will re-optimise prices upward for this residual, inelastic segment. They will bundle reasoning with governance and charge a premium for it.

This bifurcation means private prices will not simply fall. They will diverge, falling in elastic segments, rising in inelastic ones. The same logic that governed oncology drugs in West Bengal will govern high-stakes reasoning APIs.

Expansion of demand follows public entry

Public entry will not just split prices. It will also increase equilibrium quantity (Q). Reasoning models have very low marginal costs but high fixed costs. Many potential users were excluded by high access prices and licensing constraints. Public entry shifts their demand into the market, increasing adoption and total welfare.

This is the classic mixed-oligopoly result: public entry draws new users in while forcing private firms to compete harder for the mass market. The net effect is more widespread use, even if average private margins shrink.

Crucially, this does not imply private exit. In the West Bengal case, private drug firms stayed — they just moved upmarket. In AI, private labs will do the same: abandoning the low-margin commodity tier to open models and focusing on high-assurance niches.

Related benefits ripple out across the ecosystem

One can also expect supply-side spillovers in capability, supply chain and ecosystem with subsequent movers in the AI reasoning space. India may launch its public open AI reasoning model soon.  Public entry overall will also create positive externalities that push the market frontier outwards.

Open checkpoints, training recipes and evaluation harnesses from Apertus will become public goods, lowering the fixed cost of new entrants and accelerating learning-by-doing across the ecosystem.

K2 Think’s ability to run on non-NVIDIA hardware like Cerebras chips signals a credible alternative to GPU bottlenecks, potentially shifting bargaining power in the hardware market.

Open models on mainstream platforms compress the time from lab to production, teaching systems integrators how to operationalize open weights securely and reducing lock-in to incumbent cloud providers.

These spillovers are non-trivial: they alter not only prices and quantities but also the production function of the entire sector. However, we should also be a little cautious about “open washing” risks. The price discipline of public entry depends on market structure.

Risks arising out of aggressive bundling

If a handful of private firms control distribution, cloud credits, enterprise channels, app stores, they may neutralize public competition by bundling aggressively. In such concentrated markets, even after public entry, private prices can rise across the board. Regulators should watch for “open washing”: advertising open backbones while locking essential capabilities (fine-tuning, deployment tooling) behind proprietary gates.

The degree and nature of openness will also matter. Apertus is fully open (weights, code, data documentation). K2 Think is open-weight but with narrower licensing. These design choices will influence how much price pressure they exert at the low end versus how much premium migrates to private bundles.

Takeaways for policymakers

So finally, what should be the policy and business playbook in the market for AI reasoning? For policymakers, the lesson is likely clear. Fund sovereign models that are parameter-efficient, multilingual and tool-use capable. Invest in open infrastructure for deployment (RAG, observability, safety layers). Mandate transparent evaluation and independent red-teaming so that quality-conscious buyers can trust public models. Guard against GPU allocation or cloud bundling practices that blunt the competitive bite of public entry.

For private firms, the path is also clear. Expect thin margins on elastic segments. Move upmarket to sell assurance, compliance, and integration, not just tokens. Build trust, not throughput.

A mixed oligopoly in the offing

As public models commoditise reasoning, the private premium will rest on reliability, governance, and risk transfer. This is what economists call a mixed oligopoly. It is not moral play; it is a predictable equilibrium. Public entry will lower prices where demand is elastic, raise them where demand is inelastic, and expand access overall. It will also accelerate innovation through spillovers while forcing private firms to reposition.

K2 Think and Apertus are not anti-private provocations. They are market-correcting forces — the first serious public competitors in the reasoning race.  If designed well, their entry will not end the private LLM era. It will sharpen it.

‘Freemium’ model is giving way to a two-tier system

Public entry into the AI reasoning race marks a shift from a private oligopoly to a mixed public-private market. Until now, firms like OpenAI, DeepSeek, Mistral, Claude and Anthropic have sold proprietary models behind paywalls, offering free teaser tiers to attract users while monetising enterprise-grade versions with safety guarantees, compliance and service-level agreements. They operate much like private hospitals or private schools, competing on brand, quality, and risk management rather than on price alone.

By contrast, new public entrants such as from UAE and Switzerland are offering open (weight) models that anyone can deploy at near-zero marginal cost. This is closer to how public hospitals, government schools, low-cost housing boards or state-run broadband networks work: they anchor the mass market, make basic access universal, and force private players to sharpen their offerings. Thus the marker structure separates into two sub-markets.

Overall, we are heading towards from a simple premium, or freemium model, to an elite premium and a public free sovereign model. All good for choices, quantity expansion, price competition and likely in the net for global social welfare from use of AI reasoning models.