DIY AI dominates DAC 2026 engineering track as users move beyond vendor tools
At DAC 2026 (Design Automation Conference, July 26–29, Long Beach), the Engineering Track—which showcases deployed user results—reads as a census of do-it-yourself AI. Hyperscalers and semiconductor firms are no longer waiting for vendors to ship finished AI products. Samsung, NVIDIA, IBM and others are building their own orchestration, agents, and models on top of vendor simulation, formal verification and signoff engines. The EDA tools become engines; the intelligence layer belongs to the user.
Samsung presented reinforcement-learning agents tuning SoC quality-of-service parameters in emulation, outperforming manual methods via Deep Q-Networks with dueling and double-DQN enhancements. IBM detailed agentic workflows automating hardware verification debug and root-cause analysis via Model Context Protocol (MCP) servers, achieving 15–40% reduction in manual effort. A second IBM paper demonstrated agents generating EDA utilities from specifications, cutting development time from four person-weeks to under 30 minutes using reusable MCP blocks.
For architects: The visible DIY AI wave signals a market bifurcation—vendors ship tooling; users own strategy. MCP (the connective tissue for data glue) is arriving in user methodology papers barely a year after entering discussion. But the lesson stays behind corporate firewalls: each hyperscaler learns from proprietary data, the rest of the industry loses propagation of lessons.
Sources
- Primary source
- eetimes.com
“Hyperscalers and semiconductor houses are not waiting for vendors to ship finished AI products. They are building their own orchestration, their own agents, and their own models”