OpenAI Unveils First Custom AI Chip With Broadcom

2026-06-24 · Source: Bloomberg Tech · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cloud Computing & IT Infrastructure, Emerging Technologies & Innovation · Depth: Intermediate, extended

Summary

OpenAI, in partnership with Broadcom, unveiled "Jalapeno," its first custom AI chip, aiming for 50% lower inference costs and greater infrastructure control, with plans to spend tens of billions on these chips. Simultaneously, SK Hynix plans a landmark \$29.4 billion US listing on July 10th to expand HBM memory chip capacity, capitalizing on surging AI demand and structural shifts in the memory market. Cerebras reported its first quarterly earnings post-IPO, showing record revenues of \$191 million (up 92% YoY) but a stock drop due to margin impacts from renting back gear to meet demand, highlighting its unique architecture that avoids HBM bottlenecks. Separately, SpaceX secured a record \$25 billion investment-grade bond sale, significantly lowering borrowing costs for its integrated entities, including XAI, and Andreessen Horowitz discussed its \$10 billion-plus investment in SpaceX, emphasizing the long-term vision for Starship and potential orbital data centers.

Key takeaway

For Directors of AI/ML evaluating compute strategies, this brief highlights the imperative to diversify hardware beyond traditional GPUs. You should explore custom ASICs like OpenAI's Jalapeno or alternative architectures such as Cerebras's wafer-scale approach to mitigate supply chain bottlenecks and reduce inference costs. Additionally, consider the structural shifts in memory demand, which necessitate securing HBM supply or adopting HBM-free solutions to ensure long-term capacity.

Key insights

Custom AI hardware and strategic capital are critical for scaling AI infrastructure and managing costs.

Principles

• AI infrastructure demands diverse hardware architectures.
• Memory's strategic value is structurally rising, not just cyclical.
• Founder-led companies attract significant long-term capital.

Method

Cerebras's wafer-scale architecture avoids HBM, CoWoS, and 3nm node constraints, enabling faster compute deployment for customers.

In practice

• Explore custom silicon for AI inference cost reduction.
• Investigate HBM-free architectures for supply chain resilience.
• Consider long-term founder-led ventures for strategic growth.

Topics

• AI Hardware
• Custom ASICs
• Memory Chips
• HBM Technology
• Data Center Infrastructure
• Venture Capital

Best for: CTO, VP of Engineering/Data, AI Product Manager, Investor, Director of AI/ML, Consultant

Related on AIssential

• Why everyone from OpenAI to SpaceX is building their own chips (and turning up the heat on Nvidia) — Major tech firms are developing custom AI chips to reduce supplier dependence and optimize performance.
• Rebellions Bets on Memory-Centric Architecture as It Weighs IPO Options — Memory-centric AI architectures and supply chain security are critical for large-scale LLM inference and market traction.
• OpenAI's spicy new custom AI chip — OpenAI's custom "Jalapeño" chip, co-built with Broadcom in nine months, uses AI for design, achieving superior inference efficiency.
• OpenAI, Broadcom unveil custom AI chip; Qualcomm to buy Modular for $3.9 bn - Business Standard — AI leaders are diversifying hardware and software strategies to optimize performance and reduce costs.
• Broadcom’s AI Chip Guidance Raises APAC Supply-Chain Questions — Broadcom's AI chip guidance highlights supply chain volatility and margin scrutiny for HBM, not a cooling of AI demand.
• OpenAI’s Jalapeño chip is Big Tech’s spiciest move away from Nvidia — Major tech firms are developing custom AI chips to gain control and optimize performance, reducing reliance on single suppliers.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Bloomberg Tech.