ngrok bets the LLM gateway war goes to network companies: flat fee, PH #4

ngrok is the San Francisco developer-infrastructure company behind the tunneling and ingress service used to expose local servers and production apps; its AI Gateway, launched in 2026, is a hosted OpenAI-compatible endpoint that routes calls to OpenAI, Anthropic, Google and self-hosted models with server-side keys, automatic failover and access controls.

Network companies, not routers, will win the gateway fight: one endpoint for local and hosted models, priced as flat per-token infrastructure, not a cut of spend.

US$50M Series A in December 2022 led by Lightspeed Venture Partners with Coatue — ngrok's first outside funding after seven bootstrapped years.

Alan Shreve started ngrok on a small credit-card-funded server and built it into the developer default for exposing local services; the v1 open-source repository was actively developed from 2013 to 2016. The company stayed bootstrapped for seven years, signing up millions of developers and 30,000 paying customers such as Databricks and Zendesk, before its first raise: a US$50M Series A led by Lightspeed with Coatue in December 2022, which Shreve framed as funding the shift from developer tool to production infrastructure.

With the AI wave, ngrok first shipped an AI gateway inside its platform in December 2025. Early users liked centralized routing across hosted and self-hosted models but found endpoints and traffic policy too much friction for a base-URL swap, so in July 2026 ngrok rebuilt the front door as app.ngrok.ai: a dedicated dashboard, API and gateway URL. The rebuilt product publicly launched on Product Hunt on 2026-08-05 and finished #4 of the day with 356 upvotes and 67 comments; its wedge was the local model, with Ollama or vLLM on a Mac mini or idle GPU reaching the gateway over ngrok's private connectivity and hosted providers acting as fallbacks.

ngrok already owned the unsolved half: every other gateway assumed the network was solved, while ngrok's private connectivity made a self-hosted model reachable with no open ports or VPN.

Distribution was free: millions of developers and 30,000 paying customers trusted ngrok for ingress, so the AI Gateway rode existing billing, team management and brand instead of starting from zero.

Flat-fee pricing attacked the category's tax: a per-token fee stays constant against expensive frontier models, where percentage-of-spend fees grow with the price of the token, not its count.

Ship, listen, rebuild: the December 2025 alpha proved demand, and the July 2026 app.ngrok.ai rebuild removed the platform friction users complained about before the public launch.

Compete from an asset rivals lack, then price against the market's tax: ngrok's tunnel made local-model routing possible, and a flat token fee undercut gateways skimming a percentage of spend.

As of 2026-09-06 ngrok is scaling the AI Gateway as a dedicated product at app.ngrok.ai after its 2026-08-05 Product Hunt debut. Independent analysis credited the launch with eight months of usage behind a few-weeks-old interface and flagged three gaps: no semantic caching, no per-key rate limiting, and observability thinner than Portkey's or Helicone's. The open questions are whether the flat US$0.05-per-million-token model wins against percentage-of-spend and self-hosted rivals, and whether the local-model wedge becomes a meaningful revenue line on top of ngrok's ingress business.