Let’s be frank: artificial intelligence (AI) monetization and AI billing today is not conceptually unclear. It is operationally immature.

The current state of AI feels similar to the early days of Voice-over-Internet-Protocol (VoIP) technology-based telecommunication in the 2000s. Not because the markets are identical, but because the same pattern is emerging. Technology is moving fast. Business models exist. However, the operational discipline required to sustain them is still catching up.

The VoIP analogy is useful, but only if applied carefully. It’s not a template for how AI will evolve. It’s a reminder of what happens when monetization is treated as something to solve later. And that’s exactly what many AI companies are doing today, often without realizing it.

From VoIP Lessons to Reality: An AI Billing Analogy

When PortaOne launched in 2001, telecom operators were not confused about how to charge. They already well understood voice minutes, messages, and data volumes as billable units. The industry knew what it was selling.

What it lacked was the ability to execute consistently.

When the Model Exists but the System Does Not

In the 1990s, the market was dominated by established operators running expensive, complex infrastructure: carrier-grade switches, Cisco equipment, and heavyweight billing systems from vendors like Oracle. These systems were powerful. But they were also rigid, costly, and designed for a relatively stable environment.

Then VoIP changed the landscape. It lowered the barrier to entry and enabled a new wave of small, agile operators, effectively startups. These companies operated in a fast-changing environment, building businesses from scratch with limited resources. They could not rely on legacy systems, yet they faced the same fundamental challenges: measuring usage accurately, charging in real time, preventing fraud, and maintaining margins.

And, like with AI billing today, the environment was not static. New models like prepaid calling cards emerged, introducing entirely new risks. What seemed a simple operation, such as deducting the balance per call, quickly became complex when fraud appeared. For example, multiple simultaneous calls from a single prepaid account could instantly multiply losses by tens. Operators had to react fast, redesign logic, and implement controls under pressure.

This is where the real problem surfaced: billing was never just billing. It expanded into discounting, fair usage policies, taxation, partner settlements, and fraud prevention. Each new requirement consumed engineering time that companies could otherwise have spent building revenue-generating features. Many operators learned these lessons the hard way: through revenue leakage, fraud losses, and operational failures.

For example, when the market introduced new sets of termination rates (these are the wholesale costs operators pay to deliver a call to another network), many operators failed to update profit guarantee policies for their end-user tariffs in time. That led to immediate and sometimes severe margin erosion.

Anyone Can Build an AI Product, but Few Can Price It Correctly

Today’s AI market is in a similar position, but with an additional layer of complexity.

The monetization models already exist. Across the companies we spoke to, we saw the full spectrum: flat subscriptions, usage-based pricing per token or Application Programming Interface (API) call, hybrid models with overages, seat-based pricing with caps, prepaid credits, enterprise contracts, and even bring-your-own-key (BYOK) approaches where customers bring their own API keys for pass-through pricing models.

But what has changed is the structure of the market.

Previously, building a SaaS company required significant upfront investment: teams of developers and testers, long development cycles, and predictable cost structures dominated by payroll. Today, a single developer can build and launch a functional AI-powered product in weeks. The barrier to entry has collapsed. As a result, a new wave of small, fast-moving AI startups is emerging, much like VoIP operators over two decades ago.

However, their cost structure is fundamentally different. The primary cost driver is no longer internal development, but external AI usage, such as API calls to third-party models. This introduces several AI billing challenges: variability, unpredictability, and tight coupling between product usage and cost.

As in early VoIP, these companies operate in a rapidly evolving environment where products, pricing, and usage patterns change constantly. And just like before, many will underestimate the complexity of AI billing and monetization infrastructure.

This is not a market that lacks ideas. It is a market where execution has not yet caught up with ambition. And, as history shows, most players will only fully understand the problem after they have already paid for it.

The Real Problem: Visibility, Not Tools

One observation stood out across almost every conversation we had during the event in Barcelona. Revenue lives in payment processors. Costs live in vendor dashboards. Margins live in spreadsheets. And no system connects all three. This is the real AI billing gap.

Most early-stage AI startups do not feel strong billing pain yet. They rely on simple setups: flat subscriptions, BYOK models, or basic usage thresholds. Many build lightweight internal systems using PostgreSQL and scripts. At this stage, it works well enough.

However, the absence of pain is misleading. It’s not because the AI billing or monetization problem is solved. It is because the scale has not yet exposed it. 

The Hidden AI Billing Gap: Fragmented Economics

At first glance, billing (including AI billing) looks deceptively simple, almost like the classic “stone soup” story, where something trivial appears sufficient. However, as the product evolves, requirements accumulate: discounts, fair usage policies, taxes, credits, overages, and partner settlements. Each addition seems incremental, yet together they form a complex system that demands constant attention.

This complexity does not come for free. It drains engineering resources. Instead of building features that drive revenue, AI billing developers find themselves debugging edge cases in taxation logic or fixing inconsistencies in usage calculations. For early teams building everything in-house, this is almost inevitable. As the PortaOne founders’ team remembers very well from our time in Telenor, companies tend to step on the same problems sequentially: losing time, resources, money, and only then fixing what broke.

Very few AI companies systematically connect cost, revenue, and customer behavior in one place. As usage grows, this disconnect becomes increasingly dangerous. Companies make pricing decisions blindly. Margin erosion remains invisible until it becomes material. And by the time they notice it, the fix is no longer trivial. This is exactly the kind of risk we witnessed with AssemblyAI’s experience, where scaling exposed the limitations of fragmented, in-house AI billing and forced a high-stakes migration under time pressure.

When Visibility Comes Too Late: From Fragmentation to Failure

There is also a structural advantage in systems that aggregate experience across markets. In telecom, different regions evolved at different speeds. Features that seemed novel in one market had often already been solved elsewhere. 

For example, capabilities around prepaid services or fraud controls would emerge in more advanced VoIP markets, such as Australia, and only later be recognized as critical in others. Operating across multiple markets enabled the PortaOne team to anticipate these needs, integrate them into our software platforms, and introduce them to all of our customers around the globe before they became urgent in a particular market.

The same dynamic is beginning to appear in AI. Companies building in isolation tend to rediscover the same problems, often only after they have already incurred losses.

A Case Study in AI Billing Fragmentation

Consider a simple AI billing example: launching an AI product with a free trial. On the surface, this looks like a standard growth tactic. But if the company has not measured the actual cost per trial user, it can quickly become a liability. In practice, subscribers will optimize for free usage. 

Some will create multiple accounts with different email addresses, repeatedly exploiting the trial period. If each trial user generates, for example, $50 in API costs while paying nothing, the company is effectively subsidizing abuse at scale.

Without clear visibility into cost per user and usage patterns, this remains invisible until losses accumulate.

The only way to manage this is to connect cost, usage, and policy enforcement in real time in your AI billing processes. For instance, limiting free trial usage to a defined threshold, such as a fixed number of tokens, based on actual cost economics. But this requires infrastructure that most early-stage teams do not have in place.

Software vendors that consolidate learnings across multiple customers and use cases are far more likely to anticipate and prevent these issues. Vendors building everything from scratch tend to encounter them the hard way.

This is not just a tooling gap. It’s a visibility gap and, increasingly, a strategic risk.

AI Billing Complexity Is Already Here (but Not Yet Recognized)

Another pattern become clear in Barcelona. The complexity is already present. Companies are just not labeling it as a billing problem.

Take voice AI as an example. A single request may involve speech-to-text, a large language model (LLM) call, and text-to-speech synthesis, often across different vendors. Each component has its own pricing unit, its own rate structure, and its own variability. This is no longer a single cost stream. It is a multi-vendor, multi-unit system.

And this is becoming the norm in AI billing. In production, APIs are increasingly driving usage, making it programmatic, continuous, and high-volume. User interfaces are not disappearing. They are just shifting to a control and operations role.

As a result, consumption is scaling faster than most companies can effectively monitor and control.

Pricing, Control, and the Path Forward

The Hardest Problem Is Not AI Billing

It’s easy to assume that the main challenge in AI billing and monetization is technical. How to meter usage, how to bill customers, and how to integrate payments. In reality, those are solvable problems. The harder problem is aligning price with value.

In telecom, cost and value are closely linked. A minute of voice has both a cost and a clear meaning for the customer. This alignment simplifies pricing.

In AI, the primary cost driver is token consumption, but tokens are an internal abstraction. Customers do not think in tokens. They think in outcomes. They care about whether something was done, solved, or improved.

This creates a structural disconnect between how AI companies produce their services, and how the market perceives them.

That is why current pricing models feel incomplete. Token-based pricing aligns with cost but not with customer understanding. Subscriptions simplify packaging but often hide true usage. Outcome-based pricing aligns with value but is difficult to define and measure consistently.

This is not a temporary issue. It’s a fundamental characteristic of AI services.

Hybrid Pricing Is the New Normal

There is a belief that the industry will eventually converge on a single model, most likely outcome-based. That is unlikely. Hybrid pricing is not a transitional stage but a stable equilibrium.

Most AI providers will continue combining subscriptions, usage-based components, and outcome-driven elements. Each serves a purpose: where subscriptions provide predictability, usage ensures cost alignment, while outcome pricing captures value where it can be measured. Removing any one of these layers tends to create new problems rather than solve an existing one.

This is already evident in current AI billing practice: many companies are gradually shifting from flat subscriptions to hybrid models as usage grows and cost structures become clearer.

Where Telecom Experience Helps and Where It Does Not in AI Billing

Telecom experience remains highly relevant to AI billing and monetization, but only within clear boundaries.

It is directly applicable to execution. Metering, rating, charging, fraud control, and margin management are disciplines that telecom has refined over decades. These are precisely the areas where AI companies are now encountering challenges.

In telecom, even small inefficiencies could lead to significant losses. Non-perfectly configured least-cost routing (LCR) settings, for example, often resulted in traffic being sent through the wrong termination partner, with immediate financial consequences.

A VoIP Example That Is Relevant to AI Billing Today

A VoIP operator might have two vendors for terminating calls to Canada: one at $0.01 per minute and another at $0.015. Due to outdated LCR configuration or delayed rate updates, traffic continues to be routed to the more expensive vendor.

If that operator is handling 500,000 minutes per day to that destination, the $0.005 difference results in an unnecessary cost of $2,500 per day. That is over $75,000 per month without any visible change in service quality. And because the routing appears to be working correctly from a technical perspective, companies might not even notice the issue until margins are reviewed in detail.

This kind of leakage is not caused by a broken system. It’s caused by a system that is not properly configured and not continuously aligned with real-time cost conditions.

The same logic now applies to AI workloads. However, telecom experience does not fully translate to value definition. The telecom industry has standardized value units. In AI, value is contextual and varies across use cases. Telecom solves the “how to charge.” AI still needs to define the “what to charge for.”

Monetization and Cost Control Are the Same AI Billing System

One of the most underestimated aspects of AI monetization is the link between pricing and cost control. AI systems have a capability that telecom systems did not. They can make real-time decisions about how to execute tasks.

The same request can be handled by different LLMs, each with different costs, latencies, and performance characteristics. Without control, systems tend to default to the most capable, and often the most expensive, option.

At scale, this becomes a margin problem. A more disciplined approach involves automated routing of AI agents based on predefined policies. The system can direct tasks to different models depending on their nature, with decisions based on token cost, model availability, throughput, or specialization. This is effectively least-cost routing for AI.

It’s where monetization becomes operational. Pricing defines revenue. Routing defines cost. Without aligning the two, AI companies cannot control their profitability.

Two Layers of Monetization

To make sense of the current state of AI billing, it is useful to separate monetization into two layers.

• The first is strategic. This is where pricing logic is defined. It includes how value is packaged, how products are structured, and how pricing supports positioning.
• The second is infrastructural. This is where execution happens. Metering, rating, charging, invoicing, and access control all belong here. The strategic layer is where companies differentiate. It most likely must remain in-house.

The infrastructural layer is where reliability matters most. It benefits from standardization and can be externalized. Companies make the most mistakes when they mix these layers. They either overbuild infrastructure as if it were strategy, or oversimplify strategy as if it were infrastructure. 

And the question is not whether to build or buy. The real question is what should remain strategic, and what should be treated as infrastructure. AI companies should retain control over how they define value and structure pricing. That is their competitive edge.

However, building internal AI billing systems for metering, charging, and access control often leads to the same outcome we saw in early telecom: growing complexity, hidden inefficiencies, and, eventually, scaling problems. The goal is to externalize execution without losing control over strategy.

The PortaOne Perspective: Closing the AI Billing Gap

What we see today is not a lack of tools, but a lack of integration.

AI usage is scaling rapidly. At the same time, implementation challenges in AI billing remain significant. Skills gaps, budget constraints, and regulatory uncertainty continue to slow adoption. Pricing is volatile. New models appear with new rates. Older models become cheaper. Routing strategies change constantly. Static pricing and ad-hoc billing cannot keep up with this environment. This is the gap that we designed PortaAIM to address.

By combining cost and revenue tracking per transaction, real-time access control via API, and a flexible product catalog that supports any pricing model, PortaAIM connects the pieces that are currently fragmented. It enables companies to see per-customer profitability, enforce usage limits in real time, and adapt pricing without rewriting code.

In practical terms, this means moving from reactive AI billing to proactive monetization control.

What Comes Next

The AI market will not suddenly stabilize. It will mature gradually. Usage will continue to grow faster than cost control capabilities. Multi-vendor stacks will remain standard. Pricing will remain dynamic. Hybrid models will dominate. Outcome-based elements will expand where measurement improves. Regulation will introduce additional constraints.

The companies that succeed will not be those with the most innovative pricing ideas. They will be the ones who can see, control, and adapt their monetization systems in real time.

AI monetization is not a mystery. It is a system that is not yet fully operationalized. The concepts already exist. The tools are emerging. The challenge is connecting them into a coherent, scalable system.

The lesson from telecom is simple. Discipline is not optional. It’s just delayed. The only question is when you decide to implement it.

If this resonates, it’s worth treating monetization not as a feature, but as a core system. PortaAIM is one approach to building that system, based on experience from industries where getting billing wrong was never an option.

FAQ