In an era dominated by cloud-based AI, a growing movement toward edge computing is emerging. Canadian edge AI startup Picovoice offers a full portfolio of voice AI and LLM technologies, distinguishing itself by delivering cloud-level performance directly on edge devices, combining high performance with privacy and low latency.
Founded in January 2018 by Alireza Kenarsari, Picovoice aims to accelerate the transition of AI processing to where data is generated – at the user's fingertips.
Kenarsari observed the inherent inefficiencies of cloud-dependent AI assistants, questioning the necessity of sending even simple voice commands to distant data centers for processing.
"If you tell Alexa to turn off the lights, does it really need to go through half of the planet, go to a data center, and come back? Probably not," Kenarsari said. "I knew what was going to happen, so my goal was to accelerate that transition."
He envisioned a future where AI could operate more locally, mirroring human intelligence. This vision, combined with his entrepreneurial spirit, led him to establish Picovoice.
Kenarsari's journey to Picovoice is marked by a rich blend of experiences. His career began with stints at three different startups in Vancouver, where he witnessed the full spectrum of outcomes from acquisition and IPO to complete failure. This was followed by a role at Amazon, focusing on machine learning-based financial fraud detection. His close interactions with the Alexa unit at Amazon provided him with key insights that would shape his vision for his startup.
End-to-end optimization
Picovoice's core technology is distinguished by its end-to-end optimization. The company owns its data pipelines, training mechanisms, and inference engines, enabling it to fine-tune its AI models for optimal performance on edge devices. This approach allows Picovoice to match the accuracy of cloud-based APIs while providing the low latency and reliability crucial for real-time applications.
What differentiates Picovoice's solution from other edge-computing competitors? "Many edge deployment solutions use post-training optimization, where a pre-trained model (often open source) is modified to be smaller and faster. This approach has limitations: the original model isn't designed for edge deployment, restricting potential optimizations," explained Kenarsari. "Additionally, reliance on open-source runtimes like PyTorch and TensorFlow limits the optimization techniques available." These limitations make it difficult to achieve cloud-level accuracy on edge devices. That's why Picovoice addresses this by creating its own data pipeline and training mechanism.
To empower non-developers to design voice-enabled product interactions, Picovoice has developed a web-based platform that simplifies the creation and customization of voice commands and wake words. It adopts a customer-centric approach, tailoring its business model to suit the unique needs of each client. As Kenarsari notes, "If you're making an AI PC, your constraints are very different compared to someone who is making a $5 million surgical robot."
Its solutions are designed to be cross-platform compatible, supporting a wide array of operating systems and hardware configurations from Linux, macOS, Windows, Android, iOS, Chrome, Safari, Edge, and Firefox to NPU, GPU, CPU, MPU, and MCUs. "This versatility is particularly valuable for enterprises with diverse product portfolios," said Kenarsari. This is especially valuable for large enterprises that want to provide consistent user experiences across different devices.
Value proposition for the edge
Picovoice targets industries where privacy, reliability, and real-time processing are paramount, including consumer electronics, automotive, healthcare, public safety, and government tech. The company's business model is B2B-focused, offering cost-effective solutions for high-volume applications. Its technology is being utilized in various innovative applications, including LLM-based voice assistants, agentic AI, and even NASA's next-generation spacesuits. In public safety, Picovoice's on-device solutions ensure that sensitive data remains secure.
Picovoice initially earned recognition by delivering technically advanced solutions that outpaced competitors, gaining credibility among sophisticated technical buyers. As the company evolves, expanding its product portfolio and refining its go-to-market strategy, its value is increasingly recognized by non-technical stakeholders as well. With a commitment to transparency, Picovoice offers open access to its technology and resources. Its flexible pricing model is designed to adapt to the unique needs and scale of each customer.
Looking ahead
Looking ahead, Picovoice's vision is to power a billion devices with its AI technology. It is actively expanding its team and exploring strategic acquisitions to further its reach and impact.
The company remains committed to continuous innovation, investing heavily in research and development to stay at the forefront of the AI revolution.
Picovoice has achieved profitability and secured significant multi-year deals with Fortune 100 companies. While the company is not actively seeking fundraising, it is open to strategic acquisitions to expand its team and capabilities.
Article edited by Jack Wu
