Cognition hits $48 billion valuation as ElevenLabs eyes 2028 IPO

The rapid scaling of these AI entities indicates a shift from experimental pilots to production-grade enterprise deployments. Cognition and ElevenLabs are proving that specialized applications in coding and voice can command massive valuations and revenue even without owning the underlying frontier models. This trend challenges the dominance of incumbents like Google and Microsoft, forcing cloud providers to optimize their infrastructure as seen with the Amazon AWS Bedrock overhaul. The focus is moving from raw model capability to distribution and engineering efficiency. Watch for ElevenLabs' profitability margins to fluctuate as it continues to pay Anthropic and OpenAI for the model access that powers its enterprise voice agents.

Cognition's Devin coding agent has become a focal point in the enterprise AI tooling race, where multiple vendors are vying for developer workflow budgets. In June 2026, Ericsson launched its AI in RAN commercial software subscription claiming up to 20% higher downlink throughput across more than 15 live deployments, illustrating how agentic AI is moving from isolated pilots into production-grade operations across industries. That same shift from experimentation to production deployment is what underpins Cognition's $900 million annualized revenue figure and validates the broader thesis that agentic AI market growth can command enterprise-grade pricing without owning frontier models.

ElevenLabs is navigating a complex dependency structure as it scales toward its 2028 IPO target. The company's voice agents rely on third-party model access from providers like Anthropic and OpenAI, creating margin pressure that investors will scrutinize closely. Meanwhile, Amazon AWS has been aggressively optimizing its own AI infrastructure to serve as the platform layer beneath companies like ElevenLabs. Nokia combined with AWS and Databricks to build a telco AI control layer, with Nokia's Autonomous Networks Fabric running on AWS from later in 2026, demonstrating how AWS is positioning Bedrock and its broader cloud stack as the default runtime for agentic AI workloads across verticals. Amazon's strategy of rebuilding Bedrock's inference layer with minimal engineering headcount signals a broader push to reduce per-token costs, which directly benefits application-layer companies like ElevenLabs that consume massive inference volumes.

On the technical side, the competitive dynamics between AI application companies and cloud providers are intensifying. Nokia and Google Cloud announced six specialized Gemini-powered agents for telecom network operations at DTW Ignite 2026, claiming 50% to 80% reductions in problem-solving times, showing how hyperscalers are bundling their own agent frameworks on top of foundation models. This vertical integration from Google and AWS creates both opportunity and threat for independent application builders like Cognition and ElevenLabs: cloud providers offer distribution and infrastructure but also compete directly at the agent layer. The pattern mirrors what , where platform choices (GPU versus CPU for compute-heavy functions) determine long-term competitive positioning. For Cognition and ElevenLabs, the analogous strategic question is whether to remain model-agnostic application layers or deepen integration with a single cloud provider's stack.