ReadingResearchRadarInvestment framework
Sign in / Sign up
← Back to reading
NewsTopic · 新思AI芯片设计2 min read·2026-10-19

Synopsys CEO Sassine Ghazi: AI Models Turn Over Every 3-4 Months, Chip Design Cycles Under Pressure From 18 Months Toward 9, R&D Spending Keeps Rising

In a keynote interview at SEMICON West, Sassine Ghazi, president and CEO of Synopsys, explained the core tension: AI models change every three to four months, while chip design cycles currently take about 18 months and are being compressed by customers. Autonomous agents can now handle spec-to-verified-RTL, but physical implementation and sign-off still depend on human engineers. As customers push for faster delivery, EDA companies' R&D spending continues to increase rather than decrease.

ImportanceMaterialEvidenceE3 inspectableWrite-upStandard

2026-10-15 15:01 GMT+8

When asked what keeps him up at night, Sassine Ghazi, president and CEO of Synopsys, answered with three words repeated: pace, pace, pace. The interview took place at SEMICON West in October 2026, conducted by Semiconductor Engineering editor-in-chief Ed Sperling before a live audience.

Ghazi's account of where automation actually stands is honest. He confirmed that Synopsys already has customers using autonomous agents to move from a spec to verified RTL — a step that used to require many human engineers working for months. But he was equally clear that physical implementation, physical verification, and chip sign-off are not yet autonomous.

For EDA companies, AI has not cut R&D spending. Customers are pushing design cycles from 18 months down to 14, to a year, and system companies are asking for shorter than nine months. Synopsys must use AI tools to meet that pace, but developing, deploying, and continuously optimizing those AI tools itself requires more investment. Interviewer Ed Sperling noted that EDA has always exceeded almost every other hardware company in R&D as a percentage of revenue; Ghazi confirmed that trend has not changed and that investment is actually continuing to increase, not decrease.

Sign in / Sign up
ReadingResearchRadarInvestment framework

Here is a structural point worth noting for investors: EDA companies' revenue growth depends on customer design activity volume, which in turn depends on AI model iteration speed. The faster models change, the greater the demand for custom silicon, and the higher EDA revenue — but EDA companies' own margins may face pressure from sustained R&D intensity. Ghazi did not provide specific figures; this inference requires cross-checking against Synopsys financial reports.

As Synopsys CEO, Ghazi's narrative naturally serves company interests. He positions autonomous agents as complementing rather than replacing human engineers, and frames pace pressure as a source of greater demand for EDA tools — both framings favor Synopsys' business story.

However, Ghazi made two admissions that work against his own narrative: physical implementation and sign-off are not yet autonomous, and AI has not reduced R&D spending. If both hold, the EDA industry's AI story is not short-term cost reduction but a long-term capability race requiring sustained high investment.

Sources

  • 原始出处semiengineering.com↗
Topic

新思AI芯片设计

美国EDA巨头新思科技(Synopsys)正通过其AgentEngineer平台推动芯片设计从“辅助”向“自主”跨越。尽管CEO Sassine Ghazi指出物理实现和签核仍依赖人工判断,但最新数据显示其AI代理已将验证周期压缩50倍,并…

4 steps1 piecesUpdated 2026-10-20 06:49

How it got here

  1. 2023-02-15新思科技总裁Sassine Ghazi在AI硬件峰会提出AI是实现芯片自主设计的唯一出路,强调强化学习可优化PPA并降低能耗。
  2. 2026-07-27新思科技宣布与AMD和微软合作,在Microsoft Discovery上推出业界首个自主EDA工作流,调试周期缩短25-40%。
  3. 2026-09-28新思科技发布AgentEngineer,基于NVIDIA基础设施,声称将验证闭环速度提升50倍,并移除了传统流程中的最终人工审批门控。

What is still unsettled

  • 移除最终人工审批门控后,大规模量产中的良率风险如何量化?
  • Cadence等竞争对手宣称的L5级自主里程碑与新思科技的AgentEngineer在实际部署效果上有何差异?
  • 客户对‘无人在环’(no human in the loop)设计的接受度是否足以支撑2026年底的全面商用?
Read the full topic →
Continue researching →