Voice is regaining ground as an interface for enterprise artificial intelligence. Starlink is already using Grok to handle more than 15,000 calls a day, while carriers such as Deutsche Telekom, Orange Business and Ooredoo build their own AI Voice offerings. The competition is starting to shift away from telephony as a channel and toward a far more valuable layer: the one that understands what the customer wants and acts on it across CRM, support, orders, logistics or billing.
Grok Voice and the telecoms’ new race, in 30 seconds
- Starlink says Grok handles more than 15,000 calls a day and takes part in over 3,000 weekly orders across voice and chat.
- The agent is said to autonomously resolve roughly 70% of support queries.
- xAI has built telephony, enterprise tools, an API and Model Context Protocol (MCP) support into its voice platform.
- Deutsche Telekom, Orange Business and Ooredoo are already developing similar services.
- Value is starting to move from the voice minute toward automating the business process behind the call.
The change looks small if you only look at the phone. The customer still dials a number, explains the problem and waits for an answer. What’s changed is everything that happens afterward.
Traditional phone automation was built around menus, decision trees and call routing. The goal was to roughly identify what the customer wanted and transfer them to the right department.
Generative agents open up a different possibility. They can interpret an open-ended conversation, hold context, look up information and use tools to complete part of the work.
That’s where Grok Voice starts to get interesting.
Starlink is using Voice as more than a contact center
According to figures released by SpaceXAI, the system used by Starlink handles more than 15,000 inbound calls a day related to support and sales.
The company also credits the agent with autonomously resolving close to 70% of support queries, while sales conversations are said to convert at around 20%.
These figures come from the company itself, so they should be read as corporate numbers rather than independently audited results.
What’s more interesting than the exact percentage is what’s happening behind those conversations.
Starlink’s agent works with 28 tools spread across hundreds of workflows. It can take part in processes that include hardware diagnostics, equipment replacement, service credits or sales management.
That considerably changes the nature of an automated call.
When a customer explains that their terminal isn’t working, the goal is no longer simply to classify the issue. The agent can interpret the problem, query internal tools, run certain checks and trigger a follow-up action.
Voice works as the input.
The value sits in the automation behind it.
From voice chatbot to enterprise software orchestrator
This evolution also explains why xAI wants to turn Grok Voice into a platform available beyond Starlink.
The company has voice models and has built out telephony capabilities, API integration and support for Model Context Protocol (MCP), a mechanism that lets models and agents connect to external tools and information sources.
The approach increasingly resembles that of an orchestrator.
The customer speaks, the model identifies the intent, and the agent picks the tools it needs to complete the process.
For a business, that can mean checking a CRM, verifying an invoice, updating an order, booking an appointment, starting a shipment or pulling information from a support system.
Telephony then stops being the main product.
It becomes one of several possible interfaces for triggering business processes.
The strategy is especially interesting because Starlink gives xAI a production environment where it can test this architecture against a considerable volume of real conversations.
Not every AI company has its own telecom carrier to deploy voice agents at that scale.
Pricing is starting to put pressure on the traditional contact center
Another factor that could speed up adoption is price.
Grok Voice Think Fast 2.0 is listed at a public rate of $0.08 per minute of audio, while the first Voice Agent Builder offerings launched with lower prices for certain components.
These figures shouldn’t be confused with the total cost of running a contact center.
A company still needs telephony, integration, business systems, storage, security, regulatory compliance, oversight, analytics and staff for the cases the agent can’t resolve.
Even so, having a public per-minute price sets a benchmark for a market used to far more complex cost structures.
The pressure could be especially significant for repetitive first-line tasks, simple sales, issue triage or queries about structured information.
The business debate will stop being just about how many human agents can be replaced.
The question will be how much it costs to fully resolve a customer’s intent, regardless of whether the interaction lasts two minutes or ten.
Telecoms also want to control the AI layer
Traditional operators don’t seem willing to settle for being just connectivity providers.
Ooredoo Qatar has built outbound sales calls powered by artificial intelligence. Deutsche Telekom introduced a call assistant built into the network, and Orange Business is combining enterprise telephony with AI, caller identity and synthetic-content detection.
These are different approaches, but they share an idea.
Voice is once again a platform to build software on.
For Deutsche Telekom, being able to integrate the assistant directly into the network itself, rather than forcing users to install a separate app, is especially relevant.
Orange Business is putting more weight on another of carriers’ traditional strengths: the trust tied to identity and telephone infrastructure.
With synthetic voices getting harder to tell apart from a real person, knowing who’s calling can become as valuable as understanding what they’re saying.
Deepfake detection, branded calling, number reputation and authentication could become differentiators for telecom operators.
Carriers’ biggest edge is still the network
AI companies can build very capable voice models, but carriers control assets that can’t simply be replicated by deploying more GPUs.
They hold phone numbers, interconnections, subscriber identity, network signals, anti-fraud systems, regulatory relationships and a privileged position to verify certain elements of a communication.
That infrastructure could let them build a trust layer around agents.
A company could use a generative agent to handle a call while, at the same time, relying on the operator’s network to verify the caller’s identity, spot suspicious behavior or secure certain processes.
The opportunity lies in combining both worlds.
The risk for carriers would be settling for carrying calls while another company controls the entire intelligent layer sitting on top.
That has already happened, partly, in other corners of the internet.
Networks provided connectivity while much of the economic value moved to digital platforms, cloud services, apps and software vendors.
Voice AI could repeat part of that story.
Musk doesn’t need to replace the telecoms
The threat also isn’t necessarily that Starlink ends up replacing the major mobile and fixed-line carriers.
It could be enough to control the layer that interprets the customer.
If a call arrives over a Deutsche Telekom, Orange, Telefónica or any other carrier’s network, but the system that understands the conversation, queries the data and executes the action belongs to an outside platform, a meaningful share of the business value sits outside the network.
That’s the scenario that makes Grok Voice’s trajectory especially interesting.
The potential business isn’t just in charging cents per minute processed.
It’s in becoming the platform that connects human intent with enterprise software.
One customer says they want to change a plan.
Another asks where their order is.
Another needs a piece of equipment replaced.
For the user, these are conversations. For the agent, they’re events that can turn into API calls, database queries and status changes across different applications.
That orchestration layer could end up being worth much more than telephony itself.
Voice is back, but now it looks more like an API
For years it looked like voice would keep losing ground to WhatsApp, mobile apps, email and chat.
Generative AI could partly reverse that trend.
Talking is still one of the most natural interfaces for explaining a complex need. The problem was that machines were poor at understanding free-form conversation, and traditional systems had to quickly force it into a set of predetermined options.
Today’s models reduce that limitation.
And if they can also use tools, the call stops being an isolated conversation.
It becomes a way of controlling software using natural language.
For telecom companies, the opportunity is considerable — but so is the risk.
Grok, voice agents from other AI companies, and specialized platforms can build the intelligent layer without necessarily owning the network the call travels over.
Carriers still hold advantages that are hard to reproduce: identity, infrastructure, security, regulation and millions of relationships with businesses and users.
The question is whether they use those advantages to build the next generation of voice services, or let the value shift once again to a platform sitting above their networks.
Frequently asked questions
How many calls does Grok handle for Starlink?
Figures released by SpaceXAI put the volume at more than 15,000 inbound calls a day related to support and sales. The company also cites more than 3,000 weekly orders handled across voice and chat.
What share of support does Grok resolve without people?
xAI credits the agent used by Starlink with autonomously resolving roughly 70% of support queries. It’s a figure provided by the company itself.
What’s the difference between a voice chatbot and a voice agent?
A chatbot mainly holds the conversation and provides information. An agent can also use tools, query systems and execute actions such as modifying an order, starting a replacement or updating business records.
Why could this shift affect carriers?
Because part of the value can move from carrying the call to controlling the intelligence that interprets the customer and executes what comes next. Carriers can differentiate through network integration, identity, security and fraud prevention.
