AI Chips from Nvidia, AMD, Google, and Tesla: A Comprehensive Review

This article provides a review of the AI chips developed by AMD, Google, and Tesla, exploring their potential performance and market impact.

One of the standout contenders is Tesla with its strong Dojo 3 chip, which could surpass AMD in terms of both performance and chip volume, possibly ranking it as the second-best AI chip.

The significance of AI compute is prominently highlighted on Tesla's Abundance slide, which lists AI alongside Optimus, Robotaxi, and Full Self-Driving (FSD). This emphasizes the importance of the Dojo 2 and Dojo 3 AI training chips, integral to refining FSD technology and training the Optimus robot. Furthermore, the AI5 and AI6 inference chips will play a role in efficient performance for Optimus and Robotaxi.

Looking into AMD, it is estimated that they shipped between 300,000 and 400,000 Instinct MI300 units in 2024. This led to approximately $5 billion in revenue for AMD in that year.

The Average Selling Price (ASP) of AMD's chips can be calculated as follows:
$5 billion divided by 300,000 units results in around $16,667 per chip,
while $5 billion divided by 400,000 units equals approximately $12,500 per chip.

For 2025, it's projected that AMD will sell about 500,000 AI chips, generating an estimated $7.5 billion in revenue.

Nvidia AI Chips in 2025

Nvidia's data center revenue serves as a crucial indicator of its AI chip sales. Analysts expect that Nvidia will achieve $110.36 billion in data center revenue in 2024, and based on this growth, a 2025 revenue of approximately $120 billion is anticipated.

The total number of chips sold is influenced by the ASP. Nvidia's H100 GPUs reportedly sell for between $20,000 and $40,000 each. Using an average ASP of $30,000, we can estimate the number of chips:
$120 billion divided by $30,000 equals approximately 4 million chips.

Google TPUs in 2025

Google's Tensor Processing Units (TPUs) are primarily developed for internal use, with production focused on its data centers rather than for widespread sale. In 2024, the global shipment of self-developed cloud AI ASIC accelerators, including TPUs, is projected to be 3.45 million units. Google is expected to hold around 74% of this market share, translating to about 2.55 million TPUs. Anticipating a 20% growth in 2025, total shipments could rise to 4.14 million units, allowing Google to maintain its share:
4.14 million multiplied by 0.74 results in around 3.06 million TPUs.

In terms of performance, the current TPU v4 delivers 275 teraFLOPs, while TPU v5e provides 197 teraFLOPs. Looking ahead, the upcoming TPU v6, which is planned for 2025, is expected to achieve around 400 teraFLOPs.

Tesla Dojo 2 in 2025

Tesla's Dojo 2 AI training chip is projected to enter large-scale production by late 2025. According to Tesla, the Dojo 1 chips, which offer 367 teraFLOPs, equate to about 5% of the processing power of 50,000 to 100,000 Nvidia H100 chips, suggesting the existence of around 15,000 to 30,000 Dojo 1 chips in operation.

Tesla has invested approximately $500 million annually in building out its Dojo supercomputers. Assuming an ASP of $10,000 per chip, similar to high-end AI chips, this translates to about 50,000 chips purchased yearly.

The Dojo 2 is expected to deliver 10 times the performance of its predecessor, thereby achieving around 3-4 petaflops, which could put it at a performance level exceeding that of Nvidia H100 chips.

Furthermore, the Dojo 3, which Tesla plans to produce in 2026, could see an additional 10x increase in performance, leading to an estimated 40 petaflops—potentially competitive with Nvidia B300s and less expensive than Nvidia's Rubin chips.

Tesla is also expected to create a million-chip AI data center by 2026. If the Dojo 3 performs successfully, and with XAI and Tesla as its primary customers, it could propel Tesla into the second position, surpassing AMD in the AI chip market.

The AI chip landscape is evolving rapidly, with innovations from companies like AMD, Google, and Tesla shaping the future of technology.

AI, Chips, Technology, Nvidia, AMD, Google, Tesla