In the field of artificial intelligence chips, Nvidia (NVDA) has built a near-monopoly advantage with its GPUs, commanding an 85% to 92% share of the data center GPU market. However, such an extraordinarily high market share inevitably attracts challengers. Now, two tech giants, Google and Amazon, are launching a full-scale offensive to try to shake the dominance of this chip powerhouse.
Alphabet (GOOGL) has long been active in the AI chip arena, with its self-developed Tensor Processing Units (TPUs) initially used for internal compute support. Earlier this year, the company launched its eighth-generation chips, designed respectively for AI training and inference scenarios. Its latest strategic moves clearly signal an intent to capture market share: Google sponsored a large-scale data center project led by TeraWulf, providing USD 3.2 billion in financing and taking a 14% equity stake, converting the facility from cryptocurrency mining to AI computing. The data center will deploy thousands of Google TPUs, and the majority of its compute capacity has already been reserved by AI startup Anthropic. Through this flagship project, Google aims to directly demonstrate its TPUs’ ability to compete with Nvidia GPUs and has already planned to sell the chips directly to “select” customers.
Meanwhile, Amazon (AMZN) is also accelerating its chip commercialization efforts. Its cloud division, AWS, has long been developing custom chips with AI capabilities, with the latest-generation Trainium3 and Inferentia2 dedicated respectively to training and inference. Previously, Amazon used such chips only internally within AWS cloud services, but the company’s CEO recently revealed plans for external sales. According to media reports, Amazon has announced that it will directly sell its custom AI chips to customers, and the company’s AI head confirmed that relevant commercial discussions have already begun. This move marks a critical step by Amazon in officially challenging Nvidia.
Despite the shifting competitive landscape, Nvidia’s earnings growth remains robust. In its most recent fiscal quarter, the company’s revenue grew 85% year-over-year, while diluted earnings per share soared 214%, and management expects revenue to grow 95% in the next quarter, indicating accelerating growth. The continued expansion of the AI market means the arena is broad enough to accommodate multiple winners.
A prevailing view in the current market is that the recent softening of GPU compute rental prices signals a peak in demand. For example, rental prices for the flagship B200 have fallen from recent highs, and prices for the H200 have also seen significant declines within just three weeks. However, deeper analysis suggests this interpretation may overlook key supply-demand fundamentals. A more noteworthy signal is that forward contract prices are telegraphing substantial increases: one service provider has revealed that its B200 GPU rental prices will surge 94% upon contract renewal in October of this year. At the same time, delivery lead times for new orders have extended to a staggering 12 to 15 months, with schedules pushed out to 2027, clearly pointing to persistent supply constraints.
The recent massive compute rental agreement signed between Google and SpaceX further corroborates, from a different angle, that the market’s hunger for Nvidia GPUs has not waned. Analysts believe that such long-term agreements effectively alleviate short-term market concerns about Nvidia’s share being eroded by Application-Specific Integrated Circuits (ASICs). Although competition from Google and Amazon is becoming a reality, AI adoption is still in its early stages. From a valuation perspective, Nvidia’s current forward sales multiple is even lower than that of its challengers Alphabet and Amazon, which makes Nvidia stock still attractive for buying from a long-term viewpoint.
