How AMD's Software Beat Nvidia For The OpenAI Deal

Key Concepts

• AI Software: Crucial for unlocking chip performance, optimizing functionality, and enabling hardware operation and programming.
• Chip Stickiness: Engineers' familiarity with a chipmaker's software platform creates a barrier to switching.
• MI450 Chips: AMD's specific AI chips chosen by OpenAI for compute power.
• Compute Power (GW): A measure of the processing capacity required for AI workloads.
• Triton: OpenAI's open-source language for programming GPUs, now compatible with AMD chips.
• CUDA: Nvidia's proprietary programming software, known for creating a "moat" in AI model training.
• AI Model Training: The process of teaching an AI model using large datasets, often requiring specialized software like CUDA.
• AI Model Inference: The process of running a trained AI model to make predictions or decisions, generally considered less complex than training.
• Hyperscalers: Large cloud service providers (e.g., Google, Amazon) that require massive data center infrastructure.
• Project Stargate: A proposed $500 billion investment in US data centers and AI infrastructure by OpenAI, Oracle, SoftBank, and potentially the US government.

---

AMD's Strategic Emphasis on Software

AMD, a leading chip giant, places significant emphasis on software development, a strategy that may seem unusual for a semiconductor company but is critical for unlocking the full potential of its chips in the AI industry. According to Vomsy Bopana, Senior Vice President of AI at AMD, CEO Lisa Su consistently pushes for faster progress in software efforts, stating, "Great job. You need to go faster."

Software is vital because it:

• Optimizes Performance: Coaxes better performance from silicon and optimizes functionality.
• Enables Operation: Allows engineers to operate and program hardware effectively.
• Creates Stickiness: Once engineers learn a chipmaker's software platform, they are less likely to switch to another, creating a competitive advantage.

This software-centric approach was instrumental in AMD securing one of its most significant deals in over 50 years.

The Landmark OpenAI Partnership

AMD has secured a multi-billion dollar deal with OpenAI, the creator of ChatGPT, marking a major victory for the company. Under this agreement, OpenAI will utilize AMD's MI450 chips to power 6 gigawatts (GW) of compute for its popular AI products. As part of the partnership, OpenAI will also have the option to purchase up to 160 million shares, or 10%, of AMD. This deal is a substantial boost for AMD as it strives to compete with market leader Nvidia.

OpenAI's Influence on AMD's Development

The foundation of the partnership began in 2023 when OpenAI started running some of its models on AMD hardware. This initial engagement deepened, leading to OpenAI influencing both AMD's hardware design and software development.

• Hardware Design: OpenAI provided counsel on the design of AMD's forthcoming MI450 chips.
• Software Compatibility: AMD collaborated with OpenAI to make Triton, OpenAI's open-source language for programming GPUs, compatible with AMD chips. Previously, Triton only supported Nvidia GPUs. Vomsy Bopana noted, "As our relationship with OpenAI deepened, we've expanded the engagements across all portions of stack, but certainly on the software side."

The Broader AI Compute Market Landscape

The AMD-OpenAI deal comes amidst an "almost insatiable" demand for compute power in the AI sector.

• Massive Investments: Earlier this year, President Donald Trump, OpenAI, Oracle, and SoftBank announced Project Stargate, a proposed $500 billion investment in US data centers and AI infrastructure. Apple followed weeks later with a similar $500 billion commitment.
• Nvidia's Dominance: The AI frenzy has transformed Nvidia, once primarily known for gaming chips, into a $4.5 trillion behemoth. OpenAI recently entered a $100 billion partnership with Nvidia to power 10 GW of compute.
• Emerging Competitors: A new generation of semiconductor startups, including Cerebras, SambaNova, Groq, and D-Matrix, are developing specialized chips optimized for specific AI workloads, challenging the generalized approach of legacy players.
• Market Opportunity: Sid Sheth, CEO of D-Matrix, argues that the explosion of AI services means the market won't be a "winner takes all" scenario. He told Forbes, "The opportunity is just so large, there's no way OpenAI works only with Nvidia. Customers are now willing to go through the process of learning what it takes to work with AMD's software."

Nvidia's Software Moat vs. AMD's Opportunity

Industry observers have long viewed Nvidia's proprietary programming software, CUDA, as a formidable "moat" that entrenches its position, particularly in the complex process of training AI models. However, the barrier to entry is lower for inference, which is the computing required to actually run trained AI models, as it is a less complicated process. This distinction is key to AMD's strategy, as OpenAI plans to use the AMD chips specifically for inference.

AMD's Turnaround and Future Ambitions

The OpenAI deal signifies a remarkable turnaround for AMD under CEO Lisa Su. In 2014, when Su took over, AMD was struggling, facing layoffs that cut a quarter of its workforce and a stock price of around $2. The company had missed the mobile device boom, and PC sales were weak. Under Su's leadership, AMD successfully pivoted to win data center business from hyperscalers like Google and Amazon. Today, AMD's stock is worth over $235, with a market cap of $382 billion. Su's current goal is to establish AMD as a true leader in the next technological wave: artificial intelligence.

---

Synthesis and Conclusion

AMD's multi-billion dollar deal with OpenAI, centered on its MI450 chips and 6 GW of compute, represents a pivotal moment in the company's history and its strategic push into the AI market. This success is largely attributed to AMD's intensified focus on software development, which not only enhances chip performance but also creates crucial "stickiness" with customers. The collaboration with OpenAI has directly influenced AMD's hardware design and expanded software compatibility, notably with Triton. While Nvidia currently dominates the AI chip market, the immense and growing demand for compute power, evidenced by massive investments like Project Stargate, suggests a market too large for a single winner. AMD's opportunity lies particularly in the inference segment, where the software barrier is lower compared to the training phase dominated by Nvidia's CUDA. Under Lisa Su's leadership, AMD has transformed from a struggling company into a formidable player, now poised to become a significant force in the artificial intelligence era.