Custom silicon

OpenAI is aggressively pursuing strategic partnerships and hardware development, signaling a shift from solely focusing on algorithms to prioritizing infrastructure and custom silicon. Collaborations with Broadcom and acquisitions like Jony Ive’s startup aim to optimize AI accelerators and create innovative AI-native devices. The Stargate initiative solidifies OpenAI’s control over AI infrastructure through deals with Nvidia and AMD. Furthermore, OpenAI is actively cultivating its developer ecosystem, transforming ChatGPT into an AI operating system and fostering tight integration to enhance its platform’s stickiness. This vertical integration strategy mirrors those of Apple and Microsoft, aiming to establish OpenAI as a dominant force in the AI landscape.

Broadcom and OpenAI are collaborating to develop and deploy 10 gigawatts of custom AI accelerators, signaling intensified competition in AI infrastructure. This partnership, following OpenAI’s alliances with Nvidia, Oracle, and AMD, aims to diversify its supply chain and optimize compute resources. The initiative involves customized networking, memory, and compute components built on Broadcom’s Ethernet stack, potentially reducing OpenAI’s costs and enhancing efficiency. Broadcom’s stock surged, reflecting the growing demand for custom AI chips. OpenAI aims to expand compute capacity to meet the demands of advanced AI models and future superintelligence.

Apple’s iPhone Air features the A19 Pro chip with integrated neural accelerators in the GPU cores for enhanced AI. The device also debuts Apple’s first in-house wireless chip, N1, and the second-generation C1X modem, signaling a move towards greater control over core components. These chips offer improved power efficiency and performance. While Qualcomm and Broadcom aren’t immediately threatened, analysts foresee Apple gradually phasing out reliance on external suppliers, potentially including U.S. manufacturing. Apple emphasizes on-device AI for privacy and efficiency.

Broadcom (AVGO) is expected to release its fiscal third-quarter earnings, with investors focused on its AI sector performance. Revenue is projected to reach $15.83 billion, a 21% increase year-over-year, driven by custom silicon for cloud giants and networking solutions. Analysts anticipate continued growth due to escalating demand for AI infrastructure. Broadcom’s accelerator chips (XPUs) are gaining traction as cost-effective alternatives to GPUs. Cantor Fitzgerald projects significant revenue from custom silicon, potentially exceeding $40 billion by 2027. Stock has climbed 30% YTD. AI revenue reached $4.4 billion in Q2, and the VMware acquisition further strengthens Broadcom’s position.