Summary of Key Points
In May 2026, AI chip company Cerebras made its largest semiconductor IPO in history, raising over $5.5 billion on NASDAQ. Its founder and CEO, Andrew Feldman, shared profound insights into the AI industry during an interview, covering various topics such as: the reality of AI infrastructure (which is not a bubble with demand far exceeding supply), supply chain bottlenecks (especially the shortage of HBM memory), Cerebras' differentiated approach (using SRAM to avoid HBM-related issues), OpenAI's early strategic advantages, the complex dynamics of the Sino-US chip competition, the darkest moments of startup life (where months of funding can still not result in a working chip), and the real impact of AI on employment, among others. These insights span multiple dimensions including technology, business, geopolitics, and personal development.
Detailed Analysis
1. AI Infrastructure: Bubble or Real Demand?
While some claim there's an "AI bubble," and figures like Sam Kim predict that AI infrastructure will cost $3-4 trillion by 2030, Feldman is clear: this is not a bubble.
- Historical comparison: The fiber optic bubble in the 1990s involved building infrastructure first before waiting for customers; with AI, however, "customers need it now, and we can't produce fast enough." Cerebras has orders worth $25 billion in backlog, as do NVIDIA and AMD—not because they don't want to sell, but because data centers are being built too slowly.
- Why it's not a bubble: A bubble relies on betting on future demand, while currently, there's an urgent need to meet immediate demands, which are still growing exponentially (even 85-year-olds and 11-year-olds use AI).
- The role of delayed approval: Similar to traffic lights on highways, these delays help control the pace and prevent overconsumption, making the market healthier.
2. Supply Chain Bottlenecks: HBM Memory Prices Soar
HBM (high-speed memory used in GPUs) is the second major bottleneck in the AI supply chain (after TSMC's production capacity). Only Samsung, Micron, and SK Hynix can produce it, with prices rising by four to five times. Micron's gross margin has reached 80-85%, on par with software companies.
- Cerebras' differentiated strategy: Instead of using HBM, Cerebras uses SRAM, which is integrated directly into the chip during manufacturing, eliminating additional costs and avoiding shortages.
3. The Dark Moments of Startup Life
Cerebras spent a decade preparing for its IPO, including a challenging period where they burned through $8 million per month for 18 months without producing a working chip:
- The board's support: Feldman emphasizes that without internal pressure, the team would have been dismissed as incompetent. Hardware development is already difficult, and they were tackling unprecedented challenges with chip manufacturing at the wafer level.
- Family support: His wife's understanding helped him through this tough time; she realized it wasn't her fault but a production issue.
4. The Sino-US Chip Competition
This is a sensitive topic, but Feldman discusses it straightforwardly:
- Security concerns: Selling advanced chips to China would give them military advantages, making wars more difficult and costly for the U.S.
- Realistic challenges: Sanctions may encourage China to develop its own technology ecosystems (as they have done in solar and lithium batteries), which is detrimental to the U.S. However, as industrial rivals, the U.S. cannot allow its technology to be used for competitive advantage.
- Dependence on TSMC: The U.S.'s reliance on Taiwanese companies like TSMC poses geopolitical risks. Feldman suggests giving TSMC special incentives (e.g., 20 years of exemption from local regulations) to bring manufacturing back to the U.S.
5. AI and Employment
There's concern that AI will lead to job losses, but Feldman points out:
- Most layoffs are due to overhiring during the pandemic, with only 5-10% actually related to automation.
- Technology creates new jobs: New roles such as "Chief AI Officers" and "AI Governance Experts" will emerge. HR functions will evolve from mere question-answering to strategic roles.
- The bigger picture: Technology replaces old jobs but also creates new ones; the key is to adapt to change.
Conclusion
This interview not only highlights Cerebras' success but also provides a deep understanding of the AI industry's realities: explosive demand, supply chain challenges, the importance of differentiated technology strategies, and the impact of geopolitics on innovation. For everyone, understanding these aspects helps to see how AI will transform life, work, and even the global landscape.