Skip to content
Industry inflectionsAPIAUDCBANDCLCMFTWLO

AI Voice Communications Shifts Toward Agent-Led Network Purchases

Editorial illustration for AI Voice Communications Shifts Toward Agent-Led Network Purchases
Published 6 min read

Summary

Sinch and Agora say AI voice agents now buy network capacity; Agora prices AI minutes 10-100x above standard and lifted net retention to 110%, but AI revenue is still small.

AI voice agents are becoming a new class of buyer for communications networks, according to Sinch AB (CLCMF) and Agora, Inc. (API). On their quarterly earnings calls on November 5 and November 16, 2026, the Swedish communications platform and the real-time audio and video provider both said that AI programs that answer and place calls and hold live conversations with people are now customers of their networks.


Why AI voice agents buy network capacity instead of building it

An AI voice agent is an AI program that can answer or make phone calls and talk with a user in real time, typically for customer service, scheduling or personal assistance. The model only understands and responds. Moving the audio back and forth fast enough, letting the user interrupt at any moment, and supplying a phone number and a trusted caller identity all depend on the underlying communications network. Historically, the buyers of these network services were mainly social apps, online education companies and enterprise call centers, and they paid by the minute.

AI companies have now joined that buyer list. Agora said that OpenAI published a technical report this year on using WebRTC to support high-concurrency, low-latency conversations between people and AI, and that several other leading AI companies adopted WebRTC in their products around the same time [1]. WebRTC is the standard technology for real-time audio and video inside web pages and apps, and Agora sells transport services built on it. Sinch takes the telephone-network route: it launched the second version of its voice API, aimed at AI agent development, built on what it describes as the largest independent carrier-grade voice network in the U.S., and it is integrated into agent development tools including Anthropic Claude, OpenAI Codex and Google Agentspace through MCP servers [2]. MCP is a connection standard that lets AI tools call outside services directly, so developers can plug Sinch's voice capabilities into an agent while building it. Because AI companies are not building these networks themselves and instead buy from existing network operators, the demand is unlikely to land on just one or two suppliers.


Agora disclosed pricing and retention data; Sinch gave a timeline

Agora gave the most specific disclosure. Its dollar-based net retention rate rose to 110% from 97% in the same quarter a year earlier, and management attributed the improvement to strength in its core human-to-human engagement business and to growing revenue from customers building human-to-AI voice applications [1]. The metric measures how much more existing customers pay than a year earlier; a reading above 100% means existing customers are spending more. The CFO also said that, based on official website pricing, its conversational AI products cost one to two orders of magnitude more per minute than its standard real-time engagement products, or 10 to 100 times as much [1]. The CEO said its conversational AI products are deployed in more than a dozen use cases and handle more than 1 million AI agent calls per day [1].

Sinch provided timing and scale. Management expects AI to contribute meaningfully to growth starting next year, likely adding low-single-digit percentage points. It bases that expectation on ongoing discussions with fast-growing AI-native companies that need enterprise-grade communications infrastructure, even if initial volumes may be limited [2].

AI revenue is still small at both companies. On Agora's figures, AI-related revenue is about 5% of fourth-quarter revenue, its AI products pulled gross margin down to 63.4% from 66.0%, and its target for reaching operating profit has moved to the second half of 2027 [1]. Taken together, the two companies' statements show that AI voice is already generating paid usage on communications networks, but it is not yet large enough to carry overall revenue.


Phone numbers and voice transport gain a customer willing to pay more

As AI voice moves into production, communications networks gain a new party to bill. Phone numbers, caller identity, low-latency transport and interruption handling were mature, basic services; they are now components AI companies must buy from outside. Agora's pricing also shows that an AI minute can sell for more than a standard minute. The two transport routes should be read separately: Sinch runs over the traditional phone network and phone numbers, while Agora runs over WebRTC inside apps, so the two may not be competing for the same minutes.

The limits of this conclusion matter. Sinch itself cautions that initial volumes may be limited, and Agora's AI revenue is only about 5% of the total and is diluting gross margin. For now, the change looks like a new source of usage rather than a shift in revenue mix. Three measures are worth tracking: Agora's net retention rate and AI revenue share, whether the gross-margin drag from AI products narrows, and whether Sinch delivers its low-single-digit growth contribution in 2027.


Which companies could be affected

  • Twilio (TWLO): A major U.S. communications API platform. Its voice revenue grew more than 20% in the third quarter, and management said AI-native companies are moving voice agents from proofs of concept into production, while acknowledging that voice AI accounts for only about 1% of all voice interactions [3]. It faces the same new class of buyer as Sinch, so whether AI calling scales should show up directly in its voice volumes.
  • Bandwidth (BAND): Like Sinch, it owns its own U.S. voice network. Management said every AI voice agent needs a number, identity and media streaming, and enterprise voice revenue grew 15% in the third quarter [4]. It sells exactly the number-plus-transport layer, so more AI agents could raise demand for that layer.
  • AudioCodes (AUDC): It sells connectivity software that links operator voice networks to AI models, for example connecting Deutsche Telekom's network to Mistral AI. Its Voice AI business is growing 40% to 50% year over year, with targets of $25 million in 2026 and $50 million in 2028 [5]. It sits between the network and the model, so how much it is affected depends on how quickly operators and enterprises deploy AI agents.

Sources

[1] Drillr · Agora, Inc. (API) · 2026-11-16 · Earnings call

Our dollar-based net retention rate was 110%, compared to 97% in the same quarter last year. Both our revenue growth and net retention rate improved further from Q2. This reflects the strength of our core human-to-human real-time engagement business, as well as the increasing revenue contribution from customers building human-to-AI voice applications.

[2] Drillr · Sinch AB (CLCMF) · 2026-11-05 · Earnings call

[3] Drillr · Twilio (TWLO) · 2026-10-29 · Earnings call

[4] Drillr · Bandwidth (BAND) · 2026-10-29 · Earnings call

[5] Drillr · AudioCodes (AUDC) · 2026-11-04 · Earnings call

This article is meant to help you spot industry changes and companies that may be overlooked. It is not a stock recommendation.

Want deeper analysis?

Ask drillr anything about API, AUDC, BAND, CLCMF, TWLO — powered by SEC filings, earnings calls, and real-time data.

Try drillr.ai for free

drillr can make mistakes. Information only — not investment advice. Learn more