Y Combinator's Requests for Startups · Summer 2026

Inference Chips for Agent Workflows

Requested by Diana Hu · View in YC · Start with a Landing Page

Build a marketing landing page for a custom AI inference chip startup targeting cloud providers and AI infrastructure teams. The page must include: Hero section: - Bold headline: "GPUs were built for training. Agents need something different." - Subheading: agentic AI workloads — multi-step reasoning, tool calls, memory retrieval — have fundamentally different compute patterns than batch inference. This chip is designed from the ground up for the agent era. - Primary CTA: "Talk to Our Silicon Team" - Hero visual: a side-by-side architecture comparison showing GPU compute patterns vs. agent-optimized silicon, highlighting KV cache access, branching, and variable context Problem section: - Title: "Running agents on GPUs is like running a city on a highway" - 3 data points: cost per agent step on current GPUs, latency overhead from KV cache misses, energy wasted on workloads GPUs weren't designed for - Short copy explaining why batch inference chips are a poor fit for interactive, multi-step agent workflows The opportunity: - 3-column grid: Lower cost per agent step, Faster tool-call latency, Better energy efficiency at scale - Supporting stat or quote from a public AI infrastructure benchmark Architecture highlights: - 3 technical differentiators (written accessibly): native KV cache architecture, branching-aware execution units, near-memory compute for retrieval - "Designed for the workloads that matter in 2026 and beyond" Who it's for: - 3 customer profiles: Cloud providers, Enterprise AI teams running agents at scale, Edge AI deployment — each with a one-line benefit Team and credibility: - "Built by chip architects from [top semiconductor companies]" (placeholder) - 2–3 placeholder quotes from AI infrastructure leads Early partner form: - Name, company, current inference spend, workload type - "Become a Design Partner" CTA Footer with company name, tagline, and contact email. Create a modern startup design inspired by Y Combinator (YC) companies. Choose one bright primary color and build a clean, minimal color scheme around it. The design should feel bold, simple, and product-focused with strong typography, generous whitespace, and clear hierarchy.

Builds a marketing landing page for an agent inference chip design platform. Explains why GPU architectures are a poor fit for agentic workloads, the custom silicon opportunity, and a sign-up for chip designers and AI infrastructure teams. Cycle-accurate simulation, RTL design tooling, and tapeout pipelines require specialized engineering infrastructure.