SemiEngineering's Q1 2026 startup funding tracker puts hard numbers on what the industry has been sensing anecdotally: the AI hardware investment cycle is not cooling. Eighty companies raised a combined $8.4 billion in the first three months of 2026, with 18 deals above $100 million and two -- Rapidus and Cerebras -- crossing $1 billion. The composition of the cohort tells you more than the headline number.
The two dominant themes are inference infrastructure and interconnect. Kandou AI's $225M round from a syndicate that includes Synopsys and Cadence is a signal worth noting: the major EDA vendors are buying stakes in chiplet and SerDes IP companies, which reflects where they see the value chain shifting as multi-die designs become the default. Eridu's $200M Series A to build a unified switch architecture for AI data centers -- covering silicon, optics, packaging, systems, and software -- is the other archetype: companies trying to vertically integrate the interconnect stack because the existing switch incumbent model doesn't scale to hundreds of thousands of GPUs.
The second signal is the EDA AI tooling cohort. Multiple agentic-flow startups drew substantial rounds this quarter, and the funding is coming in less than 12 months after prior raises -- which means early customers are telling investors that the tools are shipping, not vaporware. Physics-informed models tuned specifically for chip design are a recurring theme. This is not the same market as general-purpose coding agents applied to RTL. The companies building EDA-specific training data and domain-aware agents are in a different tier from the ones doing prompt engineering on top of commodity LLMs.