OpenAI Partners with Broadcom for Custom AI Chips, Boosting Shares

OpenAI's Compute Capacity Expansion

• 🚀 OpenAI is aggressively expanding its compute capacity, securing 10 gigawatts with Broadcom, following previous deals with NVIDIA (10 GW) and AMD (6 GW).
• 💡 The key differentiator with Broadcom is the use of custom silicon, allowing OpenAI to embed learnings directly into the hardware for enhanced capabilities.

Custom Silicon vs. Merchant Silicon

• 🧠 Merchant silicon (NVIDIA, AMD) offers generalized chips usable by various vendors, while custom silicon is tailored for a specific company's needs, like Google's TPUs.
• 💰 Custom silicon, such as Broadcom's for Google, can significantly reduce costs, with an estimated chip cost of $6,000 compared to $30,000 for generalized chips.
• 📈 NVIDIA and AMD operate with high gross margins (75%), marking up their silicon prices, whereas custom silicon bypasses this markup.

Financial Implications and Deal Structure

• 💸 The OpenAI-Broadcom deal is distinct as it lacks investment or stock components, unlike agreements with NVIDIA and AMD.
• 💰 Scaling AI infrastructure is immensely expensive; for example, 10 gigawatts could cost around $500 billion, with NVIDIA investing up to $100 billion and AMD providing financing, leaving significant funding gaps.
• 🏦 OpenAI plans to finance these massive infrastructure costs through increasing revenue, private deals, and leveraging buy-in from major partners like NVIDIA and Microsoft.

Apple's Absence and Future Strategy

• 🍎 Apple is notably absent from these AI infrastructure partnerships, a