AI, APIs, and IoT

 Telecommunications service providers have high aspirations for network APIs. Meanwhile, hyperscalers are equally ambitious about AI (language models, prompt engineering, and agents). In both cases, players are looking for ways to capitalize, respectively, on network and (GPU-equipped) data center investments. AI and APIs developments are now starting to converge because they are complementary technologies. Here, for example, is BT’s Managing Director of Research and Network Strategy, Gabriela Styf Sjöman, discussing emerging prospects in a TelecomTV interview. 

It starts with Consumers, then switches to Enterprise 

The initial slate of telco APIs focuses on reducing fraud via consumer applications. Examples include KYC (know your customer) Match, and Number Verification. The consumer theme is also common to AI, with success characterized by ChatGPT being the fastest-growing consumer application in history.

Facing growing monetization pressures, telcos and hyperscalers are turning their attention to enterprise and vertical industry opportunities, where there is more value at stake. Gabriela Styf Sjöman described how telcos are moving on from building and operating mass-market networks. The new focus is on solving B2B, B2G, and edge cases for enterprise and government customers. Relative to consumer propositions, these are smaller in scale and require a different commercial approach. Moreover, connectivity alone does not solve for the outcomes that these (non-consumer) customers seek. 

Understanding AI-API convergence for IoT 

The enterprise and government shift also brings IoT into discussions about AI and API. An illustration of this came during a recent on-line presentation on the topic of the evolution from APIs to AI ecosystems. For business operations and critical public infrastructure, IoT is a critical source for the data that feeds monitoring and decision-making systems. During the presentation, many audience questions were about integrating IoT with AI agents. Also, what mechanisms or standards exist to share connected device/sensor data?

One way that hyperscalers are shaping the ground rules around data sourcing from enterprise databases is via new protocols. The two leading ones, MCP (model context protocol) and A2A (agent to agent) are like USB hubs that users rely on to interconnect different peripherals to their PCs. This ‘plug-and-play’ representation sounds appealing. 

While it encourages adoption by offering a ring of familiarity, the USB-hub metaphor skims over other aspects required for enterprise adoption. MCP, for example, functions on top of client-server systems. It requires developers to abstract data from edge devices (e.g., IoT sensors and gateways) to a server. That is the aggregation point from which an MCP server can access the IoT data. It represents one-way traffic, from data source to application. The simple USB analogy masks a lot of underlying infrastructure. Where, for example, are the data models and semantic parameters necessary to integrate components from multiple vendors, and to access data from legacy databases and proprietary systems?

Competitive implications for AI Hyperscalers and Telcos 

Data sharing infrastructure rests on several foundations, as identified in a recent research paper on AI agents. The challenges of deploying AI agents as autonomous participants in digital markets requires adequate capabilities to manage identity and authorization, service discovery, interfaces, and payment systems. Put differently, AI Agents are the tip of the iceberg with many enabling capabilities remaining invisible but essential for success. Most of the capability terms will be familiar to IoT platform and service providers, being part of their everyday connectivity service offerings. That suggests a service bundling opportunity to extend that expertise beyond connectivity to API and generative AI services. 

For the communications industry, the challenge is how to mobilize support for API and service-enabling capabilities in communications systems without losing ground to the plug-and-play protocols emanating from the hyperscaler sector. Easy onboarding with plug-and-play components are a carrot to entice developer communities. Once data becomes abstracted and ingested in hyperscaler environments, the telco opportunity to capitalize on value higher up the application stack is lost. This is an issue for IoT technology suppliers, and enterprise users alike. 

 

Image Credits: Alexander Hafemann and Jaoinath Ponnala via unsplash.com