Market — AI Gateways & LLM Routing

Verified claims and product-axis read for AI Gateways & LLM Routing. Every fact below is sourced; every product judgment traces back to underlying signals.

Verified facts

• Portkey, LiteLLM, and Helicone all expose Prometheus-compatible metrics endpoints. ↗ (other)
• OpenRouter's terms allow models to be deprecated with 14-day notice to API consumers. ↗ (other)
• OpenRouter natively exposes provider-side privacy policies (training-data opt-out) per model. ↗ (other)
• Helicone introduced cost alerts via Slack/webhook in 2024. ↗ _(historical_event)_
• OpenRouter offers BYOK (bring your own key) at 0% markup, monetizing only on prepaid credits. ↗ (other)
• LiteLLM project maintainer count grew to 12+ regular committers in 2025. ↗ _(historical_event)_
• LiteLLM hosts router benchmarks showing <5ms p99 added latency on small-payload requests. ↗ _(technical_spec)_
• Portkey integrates with AWS Bedrock, Azure OpenAI, and GCP Vertex out of the box. ↗ (other)
• Portkey claims 99.99% uptime SLA on its Enterprise tier with regional failover. ↗ (other)
• LiteLLM Proxy publishes weekly Docker pulls exceeding 500K on Docker Hub. ↗ (other)

Cross-cutting opportunities (industry read)

• PCIe Gen5/Gen6 & CXL Retimer / Switch IC — Hyperscale AI racks need PCIe Gen5→Gen6 + CXL retimers/switches just to keep host-to-accelerator links alive at scale; this is Astera Labs' core ALAB-listed business.
• 800G / 1.6T Co-Packaged Optics & Silicon Photonics module — Scale-out beyond ~100K GPU clusters (verified Colossus class) is bandwidth-bound; pluggable optics are hitting reach/power limits, so the industry is racing to ship CPO at 800G→1.6T while still selling AEC for short reach.
• UALink AI Scale-Up Fabric (GPU-to-GPU) — NVLink lock-in is the single biggest non-NVIDIA accelerator pain point; the UALink consortium is the answer and Astera is the first merchant silicon making it ship-worthy.

See the Products and Strategy modules for the full product list and forward-looking judgment.