Voice AI had a moment in 2025, and SoundHound AI (SOUN) was riding the wave. The company spent the year proving that its conversational AI technology could handle everything from enterprise customer service to ordering coffee from your car. Now, as the calendar flips to 2026, CEO Keyvan Mohajer is laying out an ambitious vision for what comes next.

MarketDash caught up with Mohajer to talk through the company's key wins from 2025 and where he sees the biggest opportunities ahead.

When Amelia Customers Put Polaris To The Test

SoundHound started 2025 with momentum behind Polaris, its proprietary foundation model for speech and language. The headline numbers looked impressive: up to four times lower latency and 35% lower word error rates compared to big tech models, with support for nearly 30 languages. Those are the kinds of benchmarks that look great in press releases.

But here's where it gets interesting. When actual customers started testing Polaris with their own data in real-world scenarios, the performance gap widened dramatically.

"While we showed a 35% better error rate than big tech models using industry standard benchmarks, when our customers tested Polaris with their own data, the improvement was as high as 70%," Mohajer explains. That's not a rounding error. That's the difference between a system that works okay and one that actually solves problems.

This performance advantage became increasingly visible as customers from the Amelia platform started migrating away from legacy third-party vendors. Amelia, for context, is the agentic AI platform SoundHound acquired last year, bringing along hundreds of enterprise customers across financial services, insurance, and healthcare. "We've cut error rates, lowered costs, and enabled faster iteration cycles," Mohajer notes.

The numbers back up the story. By the second quarter, revenue surged to $42.7 million, a 217% year-over-year jump. The platform was processing more than 1 billion queries per month, a meaningful milestone for any AI infrastructure play.

Enterprise Customers Who Care About Boring Things Like Reliability

The Amelia acquisition turned out to be more than just a customer grab. It brought SoundHound into some of the most heavily regulated industries out there, sectors where flashy demos don't cut it and downtime has real consequences.

"While we have diversified the industries we support, the core ingredient to succeed is the same: having the best conversational AI for customer service and user interface experiences," Mohajer says.

What surprised him most wasn't the technology integration, which went smoothly thanks to complementary stacks and zero vertical overlap. It was the cultural fit.

"What stood out most was how closely these enterprise customers aligned with our core ideal that dependable, real-world performance matters more than flashy demonstrations," he explains.

The result is a platform that Mohajer describes as capable of powering AI agents that "not only perform reliably but also make decisions, drive outcomes, and adapt to enterprise workflows." In other words, the kind of boring, dependable AI infrastructure that enterprises actually need rather than the exciting, unreliable kind they get pitched.

India's Linguistic Complexity Problem

By the third quarter, the growth continued. Revenue hit $42.0 million, up 68% year-over-year, bringing nine-month sales to $114 million, a 127% increase. Cash on the balance sheet rose to $269 million, giving the company runway to expand into new markets and form factors.

One of those new markets is India, where SoundHound landed a major deal to embed its Chat AI technology into millions of smart devices, including prominent two-wheeler brands. India represents a massive growth opportunity, but it's also a linguistic nightmare for voice technology.

"One of the biggest challenges in India is the diversity of languages, dialects, and accents. Users are often even switching between these within a single sentence," Mohajer says.

This isn't a problem you solve by training on more data. It requires fundamental architectural decisions about how your system handles language. SoundHound's technology is designed specifically to handle code-switching, regional accents, and colloquial speech in noisy environments. The goal is to let users speak naturally without forcing them to adopt some standardized pronunciation.

"That ability to understand people as they are is what will drive adoption and trust in the Indian market," Mohajer notes.

Voice Commerce Gets Serious In 2026

SoundHound also spent 2025 expanding its in-vehicle monetization strategy. The company extended voice commerce capabilities beyond the typical food and coffee ordering to include restaurant reservations and parking payments. Partnerships with OpenTable and Parkopedia, combined with wins across European and U.S. automakers, set the stage for what Mohajer sees as a defining moment.

"One of the milestones I'm most excited about is seeing agentic voice commerce become a real, everyday behavior inside the vehicle and ultimately any IoT device with a microphone," he says.

The company is running pilots with major automotive brands for hands-free ordering and paid parking. At CES 2026, it plans to showcase an expanded vision that goes well beyond commerce. Think intelligent vehicle diagnostics, automated service appointment booking, calendar management, messaging, and travel planning, all handled through voice.

The bigger play here is about platform openness. "What makes our agentic voice commerce platform especially powerful is that it is limitless," Mohajer argues. Enterprise customers can integrate their own AI agents into vehicles, and SoundHound can host any agent compatible with MCP and A2A protocols. The vision is "creating an ecosystem where innovation can scale quickly."

"Seeing these pilots mature into production deployments and new logo wins will be a defining moment for us in 2026," he concludes.

The transition from pilot programs to production deployments will tell us whether voice commerce in vehicles is a genuine behavior shift or just a novelty feature. Based on SoundHound's trajectory through 2025, Mohajer seems confident the technology is ready. Now it's about execution.