Large Model Companies and the Race to the Capital Markets: Who Can Complete the MaaS Business Loop First?

    As large model companies seek IPOs and public financing, the more important question is not who will become the first publicly listed AI model company, but who can turn model capabilities into repeatable, measurable business value. This article examines the role of the capital markets, the technical and commercial foundations of Model-as-a-Service (MaaS), the competitive position of independent model vendors, and the indicators that reveal whether a sustainable business loop has been established.

    In recent years, several large model companies in China have reportedly explored Hong Kong listings or prepared for public financing. Market attention has often focused on which company will become “the first large model stock.” From the perspective of technology and industrial development, however, listing order is not the most important issue.

    What determines whether a model company can survive over the long term is its ability to consistently provide model services that enterprises are willing to purchase, developers can call through APIs, and financial statements can verify.

    For an industry still operating under heavy investment, the capital market is more than a source of funding. It is also an ongoing mechanism for evaluating commercial performance. Companies must regularly disclose their revenue structure, research and development spending, cash flow, customer concentration, and future risks. In other words, model companies are moving from an early stage dominated by technology narratives and financing expectations into a stage where they must demonstrate their ability to convert technology into sustainable business results.

    1. Becoming the “First Publicly Listed Company” Is Only the Starting Point

    Going public typically offers large model companies three major benefits.

    First, public markets can provide a relatively long-term source of capital. Foundation model development requires continuous spending on training, inference, data processing, talent, and safety governance. These are not one-time expenses. If a company relies mainly on a small number of strategic investors, its financing schedule can be affected by individual funding sources and industry cycles. Public markets broaden the investor base and can provide more stable funding for long-term research and development.

    Second, a listing helps establish a more transparent system for evaluating operations. During private financing, a company’s valuation often depends heavily on technical capabilities, the background of its team, market expectations, and the financing environment. Once the company enters the public market, metrics such as revenue growth, gross margin, R&D spending as a percentage of revenue, and customer renewal rates become subject to continuous scrutiny.

    Third, going public encourages more standardized corporate governance. Model services involve complex issues including data security, content safety, intellectual property, model accountability, and customer compliance. As a company grows, algorithmic capability alone is not enough. It also needs sound financial, auditing, risk-control, and service-delivery systems.

    However, an IPO cannot solve an unresolved business model. If model usage revenue is insufficient, customer repurchase rates are low, and inference costs remain higher than revenue, public markets may expose operational pressure even faster. Therefore, “who lists first” mainly answers a question about financing speed. “Who can continue selling model capabilities?” is the question that determines enterprise value.

    2. What Does MaaS Actually Sell?

    Model-as-a-Service, or MaaS, can be understood as the delivery of model capabilities as a service. Instead of providing only a model file that customers must deploy and maintain themselves, a model provider combines the model with computing resources, inference systems, APIs, access management, monitoring, and security capabilities. Customers then pay according to usage, resource consumption, or a service period.

    For enterprise customers, the main value of MaaS is a lower barrier to adoption:

    • They do not need to purchase large amounts of computing resources themselves.
    • They do not need to build an inference cluster from scratch.
    • They do not need to assemble a complete algorithm and model operations team.
    • They do not need to independently manage model upgrades, elastic scaling, and failure recovery.
    • They can validate business value through an API before making a larger investment.

    From a technical architecture perspective, a complete MaaS offering generally includes the following layers.

    1. Model Layer

    The model layer provides capabilities such as text understanding, generation, classification, extraction, vector representation, and multimodal processing. In practice, enterprises often need more than a single general-purpose model. They may require a combination of models with different sizes, latency profiles, and cost levels.

    For example, a lightweight model may be sufficient for simple intent classification, while a more complex reasoning task may require a larger model. Model providers therefore need to route requests according to task complexity rather than sending every request to the most expensive model.

    2. Inference and Compute Layer

    The inference system determines the actual cost of each request. Relevant factors include input length, output length, concurrency, batching efficiency, cache hit rate, hardware utilization, and service stability.

    Even a technically impressive model may struggle to achieve healthy margins if each request consumes too many resources. Inference optimization commonly includes:

    • Dynamic batching
    • Request queuing and priority scheduling
    • Long-context caching
    • Lower-precision computation
    • Model compression
    • Tiered routing across multiple models
    • Off-peak resource reclamation and elastic scaling during demand peaks

    These capabilities may not be visible in the customer interface, but they can have a significant impact on a model company’s financial performance.

    3. Platform and Engineering Layer

    Enterprise customers care about more than whether a model can generate an answer. They also need a service that can be integrated, monitored, and audited. The platform layer typically provides identity authentication, access control, usage limits, logging, version management, anomaly alerts, and usage reporting.

    A production-oriented request flow can be abstracted as follows:

    1. Business request
    2. ↓
    3. Identity verification and authorization
    4. ↓
    5. Request preprocessing and safety checks
    6. ↓
    7. Model routing and resource scheduling
    8. ↓
    9. Inference and generation
    10. ↓
    11. Output filtering and format validation
    12. ↓
    13. Logging, metering, and alerting
    14. ↓
    15. Return result to the business system

    MaaS is therefore not simply the act of opening an API. It packages model capabilities into an enterprise service that can be operated, governed, measured, and billed.

    3. Where Can Independent Model Vendors Compete?

    Large cloud platforms typically have advantages in computing, storage, networking, security, and enterprise customer relationships. Their model services can also be bundled with other cloud products. Independent model vendors may not have the same infrastructure scale, but they still have several areas in which they can compete.

    The first is model differentiation. Enterprises do not always need the model with the largest parameter count. They are often more concerned with accuracy for a particular task, stability, response speed, and total cost. Independent vendors can optimize for specific scenarios such as code generation, knowledge-base question answering, complex reasoning, content moderation, and industry document processing.

    The second is product iteration speed. Smaller model companies may be able to adjust their models, APIs, and service policies more quickly in response to user feedback. In a market where technology is changing rapidly and requirements have not yet been standardized, fast iteration is itself a competitive advantage.

    The third is neutrality and compatibility. Some enterprises do not want their business systems to become dependent on a single infrastructure provider. They may therefore prioritize API compatibility, model replaceability, private deployment, and multi-cloud architectures. Vendors that help customers reduce migration costs may find it easier to enter complex enterprise environments.

    However, differentiation cannot remain limited to benchmark rankings or demonstration results. Enterprises ultimately purchase a complete outcome, including accuracy, response time, availability, data isolation, incident handling, and service accountability. Model quality is only one part of the procurement decision.

    4. The Core Challenges of the MaaS Business Model

    1. Revenue Growth Does Not Equal Profit Growth

    An increase in model usage does not necessarily produce a corresponding increase in profit. If customer demand is concentrated in long-context processing, complex reasoning, or high-concurrency scenarios, revenue may grow more slowly than computing expenses.

    Companies therefore need to track several metrics at the same time:

    • Revenue per thousand or per million input and output units
    • Inference cost per request
    • Resource utilization
    • Renewal rates among high-value customers
    • The gap between revenue growth and computing consumption growth
    • Losses caused by service outages and refunds

    MaaS can create economies of scale only when unit costs decline as the business expands.

    2. Customers Can Switch Providers Relatively Easily

    Model APIs have a degree of standardization. If a customer does not depend deeply on a model’s proprietary capabilities, it may be able to switch between multiple providers. This can intensify price competition and reduce vendors’ bargaining power.

    Improving customer retention is not simply a matter of increasing switching costs. Providers need to deliver broader business value through capabilities such as knowledge-base processing, evaluation systems, workflow orchestration, access governance, data security, and ongoing operations. The more customers depend on an integrated solution rather than a single API, the more stable the revenue is likely to be.

    3. Revenue Concentration Among Large Customers

    Early-stage model companies often obtain substantial revenue from a small number of large customers. These customers usually have strong bargaining power, long procurement cycles, and complex project-acceptance requirements. If revenue is overly concentrated, a change in any one contract can affect the company’s overall operating results.

    A healthier structure combines:

    • Reference customers with a sufficient number of small and midsize customers and developer users
    • One-time project revenue with recurring usage and subscription revenue
    • General-purpose model services with value-added services for specific scenarios

    4. Technology Updates Create Asset and Cost Pressure

    Model capabilities evolve rapidly, and older versions may lose their competitiveness quickly. Companies must continue investing in training and inference infrastructure, but a new model may not immediately generate a proportional increase in revenue.

    This pattern—research and development spending occurring before commercial returns—means that model companies must maintain sufficient cash reserves and plan model releases carefully.

    5. How to Determine Whether a Model Company Has Completed the Business Loop

    Instead of focusing primarily on listing dates, investors and industry observers should ask five questions.

    First, have customers actually integrated the model into production systems? Trial usage, demonstration traffic, and production calls are very different. Stable demand is created only when models become part of core business processes.

    Second, is the revenue recurring? One-time customization projects can generate cash flow, but continuous usage, subscriptions, and long-term contracts provide stronger evidence of business quality.

    Third, does gross margin improve as the business scales? If higher usage produces increasingly severe losses, the inference system, pricing model, or customer mix may still have fundamental problems.

    Fourth, does R&D spending create productized capabilities? High R&D expenditure is not necessarily a problem. The key question is whether research results translate into lower costs, higher accuracy, stronger stability, and more paid use cases.

    Fifth, does the company have adequate safety and governance capabilities? For industries such as financial services, manufacturing, education, healthcare, and government, data handling, access control, audit logs, and content risk management are often prerequisites for procurement rather than optional features.

    6. What Will the Capital Markets Ultimately Reward?

    The large model industry remains in a long-term development phase. In the short term, the final market structure is unlikely to be determined by a single product or financing event. Capital markets are more likely to reward companies with the following characteristics:

    • Model capabilities that closely match real business requirements
    • Effective control over inference costs
    • A stable and diversified customer base
    • Recurring and predictable revenue
    • R&D investment that translates into product advantages
    • Platform capabilities for security, monitoring, and governance
    • The ability to maintain technical and commercial independence within an open ecosystem

    The label of “the first large model company to go public” is therefore a temporary milestone rather than a final conclusion. What matters is whether a company can transform expensive research capabilities into reliable engineering services, and then transform those services into sustainable cash flow.

    From this perspective, long-term competition among model companies is not simply a contest over parameter scale or capital resources. It is a combined competition involving model quality, inference costs, product engineering, customer management, and organizational governance. The companies that can first build a complete loop—from research and development to delivery, from usage to revenue, and from revenue to profit—will be better positioned to lead the next stage of the artificial intelligence industry.