La empresa que creó un chip gigante para competir con NVIDIA

La empresa que creó un chip gigante para competir con NVIDIA

by Inteligencia Artificial

Applied Sciences & Engineering Engineering & Technology TJFElectronics engineeringTJFCElectronics: circuits and components
🎙 Inteligencia Artificial 👥 725K 📅 July 30, 2026 ⏱ 82 min 👁 26K 📄 expert opinion 🧭 2026-08-03
Available in: English (current) Français

Keywords

CerebraschipNVIDIAAIdata center

Summary

In this interview, Andrew Feldman, CEO of Cerebras, explains how his company created the world’s largest chip, the size of a dinner plate, to address a 75-year-old problem in chip manufacturing. He details the advantages of large chips for AI inference, particularly their use of SRAM memory which is thousands of times faster than the HBM used in GPUs, enabling speeds up to 15-20 times faster than competitors. Feldman discusses the immense demand for AI compute, citing a backlog of $24 billion in orders and the industry-wide bottleneck in data center construction. He argues that the AI boom is not a bubble because demand currently exceeds supply, contrasting with past infrastructure bubbles. The conversation covers the challenges of building data centers, including energy, permits, and community opposition, and touches on geopolitical aspects like China’s chip ambitions and ASML’s monopoly. Feldman also reveals OpenAI’s strategic investments in compute, including their partnership with Cerebras, and mentions NVIDIA’s acquisition of Groq. He emphasizes that the real bottleneck is not chips but electricians and infrastructure, and discusses the potential of distributed computing. The interview concludes with reflections on leadership, the future of AI, and its societal impacts, including job displacement and education.

198 words

Critical Evaluation

The interview provides valuable insights into the AI hardware industry from a key insider, Andrew Feldman, CEO of Cerebras. The discussion is technically informative, explaining complex concepts like SRAM vs. HBM memory and the importance of data movement in AI inference in an accessible manner. Feldman’s arguments about the AI infrastructure boom are compelling, backed by concrete figures such as the $24 billion backlog and the comparison of data center spending to office construction. The claim that the AI industry is building behind demand, not ahead of it, is a strong counter to bubble narratives, though it is inherently self-serving. The conversation also touches on critical issues like energy consumption, permitting, and public opposition, providing a realistic view of the challenges ahead. However, the interview is essentially a promotional platform for Cerebras, with Feldman naturally highlighting his company’s advantages. The presence of sponsored segments (InfoJobs and Plaud) further indicates commercial interests. While the information is likely accurate, it lacks independent verification, and some statements, such as the speed comparison with GPT-5.6, are not independently confirmed. The title accurately reflects the content, focusing on Cerebras’ giant chip and its competition with NVIDIA. Overall, the interview offers a valuable perspective on the AI hardware landscape, but viewers should be aware of the potential bias and the need to cross-reference with other sources.

220 words

Title / Content Match

The title accurately reflects the main topic: Cerebras' giant chip and its competition with NVIDIA.

Quality & Reliability

8/10

The interview features Andrew Feldman, CEO of Cerebras, providing firsthand insights into chip architecture and industry trends. Claims are plausible and align with public knowledge, but lack independent verification and contain promotional elements.

Chapters

Cited Sources

Concurring Sources

  • Cerebras Systems — Wikipedia article on Cerebras, consistent with the company's claims.
  • NVIDIA — Wikipedia article on NVIDIA, providing context on the competitive landscape.

Contribution & Novelties

The interview provides unique insights into Cerebras’ chip architecture and its implications for AI inference speed, as well as the strategic decisions of major AI companies. It offers a firsthand perspective on the infrastructure challenges and the competitive landscape.

Pour aller plus loin :

  • Cerebras Systems — Background on the company and its technology.
  • SRAM — Explanation of SRAM memory used in Cerebras chips.
  • HBM — Explanation of HBM memory used in GPUs.
  • NVIDIA — Overview of NVIDIA’s role in AI hardware.
  • Data center — General information on data centers and their challenges.

93 words

Radar Profile

The radar profile shows high scores in quantity and quality of information, with a moderate technical level and high reliability. This indicates a well-informed expert discussion with practical insights, though the promotional nature slightly tempers the reliability score.

Reliability 8/10