The rapid growth in computing demand for artificial intelligence continues to drive the accumulation of Nvidia‘s wealth. Its graphics processing units (GPUs) excel in the training and inference tasks of AI models, leading to continuously rising chip prices. Major tech companies are scrambling to purchase them, almost keeping pace with Nvidia’s production rhythm. Due to its significant technological advantages, it is estimated that Nvidia holds about 80% of the market share in the AI chip rental market for cloud computing platforms. The upper limit of this market share currently appears to be primarily constrained by its ability to stably supply chips to its core large customers.
However, recent moves by a major customer may signal potential challenges to Nvidia’s market dominance. If other major customers follow suit, Nvidia’s stock price growth momentum could be dampened. It is worth noting that Nvidia’s staggering revenue growth heavily relies on a few large tech companies, and customer concentration is showing an upward trend. Last quarter, just two customers contributed 39% of its total sales, while the top six customers accounted for as much as 85%. In comparison, the top two customers accounted for 25% of sales in the same period last year, and the top six customers collectively accounted for less than 66%. These core customers are generally considered to be industry giants such as Microsoft, Amazon, Alphabet, Meta Platforms, OpenAI, and Oracle.
The key turning point is that OpenAI has recently demonstrated a clear intention to reduce its reliance on Nvidia. Although it entered into a large agreement with Nvidia involving capital investment of up to $100 billion, OpenAI simultaneously signed a massive procurement contract with its competitor AMD. This agreement includes an incentive clause: upon meeting certain conditions, such as reaching specific AMD chip purchase volumes, OpenAI has the right to subscribe to AMD equity at a very low price. According to the plan, OpenAI will deploy a total of 6 gigawatts of AMD GPUs over the next few years, and for every 1 gigawatt of purchase volume achieved, it will receive a batch of warrants. This move essentially equates to AMD offering OpenAI a discount on chip purchases while consolidating its own position within the AI ecosystem. For reference, the agreement with Nvidia only fully takes effect after the purchase volume reaches 10 gigawatts. This means that although Nvidia will remain OpenAI’s primary supplier in the short term, its absolute dominant share has begun to loosen.
This collaboration might just be the beginning of a new industry trend. The effectiveness of the OpenAI-AMD deal largely depends on AMD’s upcoming MI450 GPU series, which is expected to hit the market around the same time as Nvidia’s Rubin architecture products. AMD confidently claims that the MI450 will surpass Nvidia’s products in both training and inference performance, and analysts at least expect it to be competitive enough to capture significant market share from Nvidia.
Meanwhile, Nvidia’s major customers have not stopped their own independent research and development efforts. Microsoft is planning significant investments in its next-generation custom AI chip, Maia 300, and intends to redesign it using more advanced process technology this summer. Meta is also actively expanding the application of its custom MTIA chips in generative AI scenarios. Furthermore, OpenAI itself is reportedly collaborating with Broadcom, planning to invest up to $10 billion to develop a custom AI accelerator.
