Synopsys is adding a royalty on its customers' chips to its custom IP business and selling AI engineers alongside its design software. Ravi Subramanian, its chief product management officer, said in a podcast interview that its custom "application optimized IP" for Amazon carries three charges: a license fee, a customization fee to prioritize scarce engineers, and, for the first time, a royalty on the final chip. He tied the model to Amazon's earlier Arm relationship and to 12-month design cycles, and said hyperscalers whose revenue rests on one or two applications want IP tuned to them over standards-based blocks. The Amazon agreement, valued at more than $1B, also covers Synopsys EDA, simulation and custom agentic AI for Amazon's engineers across its Nitro, Graviton and Trainium chips. The second product is labor. GPT-Synopsys, built with OpenAI, is a model that operates Synopsys tools to run power, performance and area optimization, timing closure and verification, sold as bundled compute, model and licenses on OpenAI-hosted infrastructure, with customer design data excluded from training. (Synopsys)
Private Companies
Firmusother
Firmus withdrew its ASX listing application on Friday in Australia, a day after closing its institutional book on a float that sought about A$7B at A$11 a share, a valuation near A$44B, and will pursue private capital. It reported $50.8M of revenue last year and expects about $30B of debt once its sites are built, against $5B of forecast 2028 operating earnings. Priced on October 6 and pulled three days later, the deal returns neocloud valuations to private markets. (ABC)
VEIRfundraise
VEIR raised a $110M Series C co-led by Matter Venture Partners and Tyche Partners, with LG Technology Ventures and Gates Frontier among new investors, for $225M raised in total. The Massachusetts company sells superconducting power cables as cooled systems, has demonstrated a 3 MW system, and is aimed first at behind-the-meter power in AI data centers. No customers are named. Proceeds go to manufacturing capacity as it moves from testing to first commercial projects. (The Next Web)
Public Markets
Micron fell with the Philadelphia Semiconductor Index, down 3.4%, after the Financial Times reported OpenAI's annualized revenue near $50B, against roughly $70B previously circulated. Part of the gap is how cloud-partner sales are counted. (CNN)
Emerging
Flash tier: A paper from Jaehoon Yang, Jeongmin Lee, Haneul Park, Seung Yul Lee, Nam Sung Kim and Jae W. Lee, Lachesis: Lifetime-Aware KV Cache Placement for Agent Serving across HBM and High-Bandwidth Flash, places agent KV cache by expected lifetime so short-lived data stays in HBM and fewer writes reach flash. In trace-driven simulation it extends flash lifetime 1.19x to 3.13x over HBM-first placement, to 3.3 to 12.2 device-years. Write endurance, not capacity, is the limit on a flash tier for agent serving, and placement software decides how much HBM is still needed. (arXiv)