AI’s value is in the application layer

As telecom CEOs and CTOs weigh their artificial intelligence (AI) investments, a crucial question emerges: Where does AI’s real value lie—in foundation models or applications? With telecom giants like SK Telecom investing $100 million in Anthropic and the Global Telco AI Alliance forming to create industry-specific large language models (LLMs), it’s time to challenge conventional wisdom.

The foundation model Black Hills

Does it feel like we’re in the middle of an AI gold rush? It should. The industry is witnessing massive investments in AI foundation models:

• Microsoft’s $13 billion stake in OpenAI
• Amazon’s $8 billion investment in Anthropic
• SK Telecom, Singtel, Deutsche Telekom, e&, and SoftBank forming the Global Telco AI Alliance
• T-Mobile committing $100 million to OpenAI

But here’s what many in the telecom industry haven’t fully grasped: foundation models are rapidly becoming commoditized, while the real value is accumulating elsewhere. This isn’t speculation—it’s happening right before our eyes. Foundation models are becoming utilities, not differentiators.

Think of foundation models like any other technology component that initially provided competitive advantage but eventually became a commodity—just like cloud computing infrastructure, which once differentiated hyperscalers but is now an expected utility. The real differentiator now lies in how you apply these models to solve specific business problems.

AI is advancing so rapidly that foundation models hailed as revolutionary just a couple months ago are now becoming increasingly standardized offerings available from multiple providers. This commoditization, like most things with AI, is happening faster than anyone predicted, driven by several factors:

• Plummeting prices: Training and inference costs have been dropping by 10x year over year. This is a classic sign of commoditization, where competition drives down prices.
• Ready availability: Every week brings a new model from Anthropic, Google, Meta, OpenAI, and dozens more. AWS Bedrock alone offers access to over 100 foundational models. For an enterprise, accessing a powerful LLM is no longer an issue. Plus, there are low switching costs as applications can swap underlying models with minimal friction.
• Similar capabilities: The performance gap between models is disappearing. Sure, there are nuances depending on the modality you need, but the core capabilities of different LLMs are becoming increasingly similar. Most offer strong performance in tasks like code generation, mathematics, video and image comprehension and generation, audio and visual processing, and general knowledge access.

As AI foundation models become increasingly standardized and accessible, their strategic value will continue to decline. Forward-thinking telecom companies are already shifting focus to where the lasting competitive advantage will be found—in the application layer.

The real gold is in the application layer

The true differentiator won’t be owning a slightly better foundation model. It’ll be offering domain-specific applications that make a huge impact on the bottom line—like increasing revenue by double-digits, providing new insights, or dramatically cutting costs.

Models can’t do that. But applications can, and do:

Turn AI into ROI: Applications translate AI potential into bottom-line results. While models generate text, applications generate revenue.

Fuse domain knowledge with AI: Your industry expertise is worthless inside a general model, but priceless inside a tailored application. This knowledge asymmetry is your competitive edge.

Break system silos: Applications coordinate between your business systems, creating intelligence no LLM could achieve alone. Foundation models don’t know your systems exist; applications make them work as one.

Leverage proprietary data: Your unique data can’t feed public models, but it can power private applications. PlanAI proves this by turning subscriber data into personalized offers that drive measurable revenue growth.

Adapt as models commoditize: Applications let you swap foundation models like utilities, protecting your investment as the AI landscape evolves.

This layer is where the smart money is placing its bets.

AI agents are just the beginning

The value living in the application layer becomes even more evident when we look at the emergence of agentic AI. These autonomous agents won’t excel because they run on marginally better foundation models; they’ll excel through domain-specific knowledge and sophisticated application architecture.

Aaron Levie, CEO and cofounder of Box.com, describes this emerging paradigm brilliantly:

This is exactly where the application layer creates its value—not in having a marginally better language model, but in how these applications coordinate and extract intelligence from your business data and processes.

Even Satya Nadella, Chairman and CEO of Microsoft (which has spent the MOST money on foundation models), is now talking about how the value will be at the app level. He recently noted:

The message couldn’t be clearer: The value is not in owning the model. The value is knowing your domain and building applications that solve real problems within it.

The alchemist: Totogi’s BSS Magic

At Totogi, we’ve put this principle into practice. Rather than chasing after better foundation models, we’ve focused on creating applications that leverage existing models to solve real telecom problems. With BSS Magic, we’ve built a domain-specific ontology that allows you to build a digital twin of your telco’s processes.

Here’s how BSS Magic demonstrates the value creation at the application layer:

1. Immediate business impact: A major telecom provider in Singapore was struggling with an inefficient order creation process. Each order required 50+ clicks and 5-6 minutes to complete. With BSS Magic, we reduced the order creation process by 80%: from 5+ minutes to about 50 seconds. Implementation took just four weeks.
2. Interoperability breakthroughs: One BSS migration project (from one vendor to another) that would have taken six months with traditional methods was completed in just four days with BSS Magic. A North American Tier-1 telco wanted to switch billing systems for a small services offering. With BSS Magic, we mapped the data in mere minutes and migrated the integrations with generated AI code. The whole project took less than two weeks, saving $210,000 per year.
3. Revenue generation: A Tier-1 operator in EMEA with 15+ million subscribers was struggling with traditional customer value management software that delivered only 1-2% revenue uplift. We implemented PlanAI, which uses predictive and generative AI to create hyper-personalized, real-time offers for individual subscribers based on network usage patterns. The results? A 60% increase in offer acceptance, 105% increase in data usage from personalized offers, and a 10% increase in overall revenue—with 20% growth in sight. This wasn’t just tweaking a model; it was building an application that turned network data into a genuine money-making machine.

These examples aren’t about having a better LLM model—they’re about applying AI intelligently at the application layer to solve specific business problems. The underlying foundation models are interchangeable commodities; the real value comes from how we’ve applied them to telecom-specific challenges.

The bottom line

The gold rush for foundation models is a distraction. The real value isn’t in owning a slightly better language model—it’s in building applications that deliver measurable business results.

While competitors sink hundreds of millions into foundation models that rapidly commoditize, telecom companies focused on application development stand to capture an additional $15-20 billion in global revenue by 2027, according to Analysys Mason. That’s not just market share—that’s market leadership.

Remember: Anyone can access a model. Few can build applications that matter.

Recent Posts

Get my FREE insider newsletter, delivered every two weeks, with curated content to help telco execs across the globe move to the public cloud.