AI chip showdown
Nvidia faces fierce competition from industry giants and startups alike. Learn how the Rubin Superchip is its secret weapon in the AI chip wars.
The era of artificial intelligence (AI) supremacy is defined by raw compute power, software ecosystems and manufacturing scale. Nvidia currently claims about 80% of the AI training market and up to 95% of inference workloads.
Yet Intel, Advanced Micro Devices (AMD), hyperscale cloud providers, and innovative startups are chipping away at Nvidia’s dominance..
Nvidia's dominance faces challenges from four primary fronts:
Competitor | Key players | Competitive edge |
---|---|---|
Established chipmakers | Intel (Gaudi 3) AMD (Instinct MI300X) |
Cost-effective alternatives, edge AI focus, software ecosystem development |
Tech giants | Google (TPU, Trillium) Amazon (Inferentia) Microsoft (Maia, Cobalt) Meta, Apple |
Reducing Nvidia dependency, tailored AI chips for own services and devices |
Startups | Graphcore, Cerebras, Groq, SambaNova, D-Matrix | Innovative architectures, venture-backed disruption potential |
Market trends | Apple, Qualcomm, OpenAI Triton | Edge AI processing, software alternatives to CUDA |
Nvidia's response to these challenges is the Rubin architecture, named after astronomer Vera Rubin. This groundbreaking platform combines a next-generation graphics processing unit (GPU) built on Taiwan Semiconductor Manufacturing Company (TSMC)'s 3 nanometre (nm) process with Nvidia's first custom central processing unit (CPU), Vera, creating an integrated 'superchip.'
Chip model | Peak performance (petaflops) | Power draw (watts) | Petaflops per watt |
---|---|---|---|
Nvidia H100 | 4 (FP8 with sparsity) | 700 | 0.0057 |
Nvidia Blackwell | 20 (FP4) | 900 | 0.0222 |
Intel Gaudi 3 | 1.8 (BF16/FP8) | 600 | 0.0030 |
AMD MI350 | 2.6 (estimated, FP8) | 600 | 0.0043 |
Nvidia Blackwell leads in energy efficiency, delivering roughly four times the petaflops per watt of the H100.
TSMC’s 3 nm capacity is nearly booked through 2026. Nvidia holds a large share of those wafers for Blackwell and Rubin, with AMD, Intel and Apple also securing booked volumes. Rising wafer costs (up 3 to 6% for 3 nm) and Taiwan-centric risks add execution challenges.
Nvidia’s volume reservations grant scale advantage but expose it to geopolitical tensions in semiconductor supply.
“Custom silicon and open software ecosystems are the foundation of AI’s future. By tightly integrating hardware and software we unlock unprecedented performance and efficiency that power the world’s most ambitious AI workloads.”
- Jensen Huang, Nvidia CEO
“Energy efficiency is not just about the chip; it’s about the entire system - architecture, packaging, software and data centre infrastructure working in harmony.”
- Mark Papermaster, AMD CTO
In a fiercely competitive landscape, Nvidia's unrelenting innovation in H100, Blackwell, and Rubin architectures, combined with its robust ecosystem, solidifies its position as the leader in AI chips. While rivals like Intel and AMD continue to make strides, and emerging threats from startups and open-source initiatives keep Nvidia on its toes, the company's supercharged roadmap and ecosystem advantages set a high bar for competitors to overcome.
As the AI revolution unfolds, Nvidia appears well-positioned to maintain its reign in the AI chip market for the foreseeable future.
This information has been prepared by IG, a trading name of IG Australia Pty Ltd. In addition to the disclaimer below, the material on this page does not contain a record of our trading prices, or an offer of, or solicitation for, a transaction in any financial instrument. IG accepts no responsibility for any use that may be made of these comments and for any consequences that result. No representation or warranty is given as to the accuracy or completeness of this information. Consequently any person acting on it does so entirely at their own risk. Any research provided does not have regard to the specific investment objectives, financial situation and needs of any specific person who may receive it. It has not been prepared in accordance with legal requirements designed to promote the independence of investment research and as such is considered to be a marketing communication. Although we are not specifically constrained from dealing ahead of our recommendations we do not seek to take advantage of them before they are provided to our clients.