Two stories landed this week that look unrelated and aren't. On Wednesday, RingCentral reported Q3 earnings that beat its own guidance, with revenue of $676 million, up 5.9% year over year. The number that mattered most wasn't revenue. It was the $400 million in annual recurring revenue from AI products, a figure that has doubled in twelve months, and a net retention rate above 100% among customers who use those paid AI tools. People who adopt AI communication features keep paying for them. That's the strongest signal yet that AI voice is a durable business line, not a demo. The second story came from RingCentral's home country's northern neighbor, and from this company's own backyard. In Nova Scotia, a $10 billion Anthropic-backed data centre at Point Tupper has drawn an 18,000-signature petition, a skeptical editorial in the Winnipeg Free Press, and a federal AI minister telling the Senate that Ottawa can't regulate the project's electricity use. Canadians are asking whether AI infrastructure benefits them or just consumes their power and water. Put the two stories together and you get the real question for any business thinking about AI calling in 2026. The technology works. The economics work. What's unproven is public tolerance for AI that contacts people without their knowledge, consent, or a clear way out. Businesses that get consent right will ride the boom. Businesses that don't will supply the cautionary tales.
Key points
- RingCentral's AI product annual recurring revenue hit $400M in Q3 2026, roughly double a year earlier, and now makes up about 15% of total ARR.
- Retention tells the real story: customers using RingCentral's paid AI products stayed at over 100% net retention, above the ~99% platform-wide figure.
- Public skepticism about AI is showing up outside the industry too, with an 18,000-signature petition against the Point Tupper, Nova Scotia data centre and provinces drafting their own rules.
- Regulators treat AI-generated voices as artificial voices under the TCPA, so prior express consent and clear disclosure are now baseline requirements, not extras.
- Managed, consent-checked calling campaigns give small and mid-sized organizations the benefits of AI voice without buying seats, building integrations, or becoming compliance experts.
RingCentral's Q3 report is worth reading closely if you run a clinic, a franchise, a staffing firm, or anything else where the phone still does real work. AI Receptionist, the company's automated front-desk product, passed 22,000 customers, roughly four times last year's count. Its Ace AI Conversation Expert sits near 7,000 customers, up 60%. The Customer Engagement business grew about 50% quarter over quarter. More telling than adoption is stickiness. RingCentral's net retention runs above 99% platform-wide. Among customers using paid native AI products, it's over 100%, meaning those customers expand their spend over time. When a product cohort retains better than the base platform, that's not hype. That's people finding the tool useful enough to buy more of it. The company now processes about 40 billion minutes and 14 billion calls a year, and it raised full-year guidance to $2.65 billion in revenue. Meanwhile, market research on conversational AI for contact centers points the same direction: buyers want AI that completes multi-step tasks and hands off cleanly to humans, especially in healthcare scheduling and patient support. So the category My AI Call Center operates in just got the strongest commercial validation it has ever had. Here's the part that should make you pause before celebrating.
While RingCentral's investors were reading a transcript, Nova Scotians were signing a petition. The Point Tupper project, which EverWind will lease to Anthropic in an 18,600 square metre facility growing to 420 megawatts by mid-2028, is the largest private investment in the province's history. Premier Tim Houston frames it as a vote of confidence in rural Nova Scotia. Critics, and there are now tens of thousands of them, see a region being targeted because it's economically desperate, with few permanent jobs to show for it. The regulatory picture adds to the unease. Federal AI Minister Evan Solomon told a Senate committee that Ottawa cannot impose electricity obligations on data centres; that's provincial jurisdiction. Nova Scotia and Saskatchewan are now drafting their own laws. Over 160 data centre proposals have landed federally, and the rules are being written while the bulldozers warm up. Why does this matter to a business running outbound calls? Because it shows the mood. People are increasingly willing to say no to AI when they don't see the benefit or the guardrails. A data centre that drinks a town's water and powers someone else's model is an easy no. An robocall that mimics a human voice, calls without consent, and offers no way to reach a person is an easier one, and it comes with a federal lawsuit attached.
Under the TCPA, AI-generated voices are treated as artificial voices, which means prior express consent is required before you call. State-level quiet hours, day restrictions, and registration rules stack on top. Every call needs AI disclosure, and recipients need a real path to opt out, a human, or both. Keyword opt-outs like STOP and REVOKE have to be honored immediately and carried across every campaign. That's the floor. The ceiling is list discipline, and it's where most AI calling operations quietly fail. A campaign is only as compliant as the list behind it. Bought lists with no clear permission records are the classic trap: they look cheap until the complaint letters arrive. A serious operator reviews the list source and consent records before launch, flags anything shaky, and in most cases declines it, before the client spends a dollar. That refusal is a feature. If a vendor will take any list you hand them, ask yourself who carries the legal risk. It isn't the vendor. It's you, the business that initiated the contact.
The companies dominating the conversational AI market, RingCentral, Five9, Genesys, NICE, AWS, Salesforce, all sell the same thing: software and seats. You buy the platform, build the integrations, write the scripts, train the staff, and own the compliance burden forever. For a 200-person multi-location organization, that's a second job. What most of these businesses actually want is an outcome. Appointment reminders that cut no-shows. New leads called within minutes, inside approved calling windows. Renewal calls placed 30 to 60 days before the date, not after the customer has left. A dispositioned list back in the CRM with real outcome codes, not a dashboard of invented metrics. That's the gap a managed, campaign-based model fills. One clear goal per campaign, quoted before launch. Calls run only against approved, permissioned, or reviewed lists. Outcomes route back into the CRM and scheduling tools you already use, with hot leads transferred live to your team. Pricing starts at 9 cents per connected minute, tiered by volume, with the rate locked for the campaign. No per-seat charges, no platform bill, no minimums you didn't choose. The agentic AI trend the market analysts describe, AI that accesses authorized systems and completes tasks, is exactly what a managed campaign does. The difference is that someone else operates it, and someone else checks the consent records first.
The AI calling boom is real, and RingCentral's retention numbers prove it's not a fad. But the same week's headlines from Nova Scotia show what happens when AI arrives without visible benefit or guardrails: 18,000 signatures and a regulatory scramble. For businesses that call customers, the lesson is simple. Use AI where it's genuinely useful, on lists where permission is documented, with disclosure on every call and an opt-out that actually works. Do that, and the boom works for you. Skip it, and you become the reason the next petition gets signed. If you're weighing your first structured calling campaign, the campaign review is free, and the full cost is known before you approve anything.