TLDR
- Goldman Sachs maintains its preference for Broadcom and Nvidia in the AI compute market despite rising competition from custom chip makers
- Google and Broadcom’s TPU v7 achieved 70% cost reduction per token compared to TPU v6, matching or slightly beating Nvidia’s GB200 NVL72
- Nvidia retains competitive advantages through faster time to market and its CUDA software ecosystem for enterprise customers
- Amazon’s Trainium and AMD chips show only 30% cost improvement and trail behind Nvidia and Google solutions
- Goldman created an inference cost curve analysis comparing chip vendors as the GPU versus ASIC debate intensifies
Goldman Sachs has reaffirmed Broadcom and Nvidia as its top picks in the AI compute ecosystem. The investment bank released analysis comparing different chip vendors as competition heats up in artificial intelligence hardware.
The firm constructed an inference cost curve to evaluate chip performance and pricing. This analysis addresses investor concerns about whether graphics processing units could lose market share to application-specific integrated circuits.
Google and Broadcom’s tensor processing unit is closing the performance gap with Nvidia’s GPU solutions. The TPU v7 delivers approximately 70% lower cost per token compared to the previous TPU v6 generation.
This improvement puts the TPU v7 on par with or slightly ahead of Nvidia’s GB200 NVL72 in terms of absolute cost. The rapid advancement shows how quickly custom chip designs are evolving.
Nvidia’s Competitive Position
Despite the narrowing cost gap, Nvidia maintains key advantages in the market. The company delivers new chip architectures faster than competitors can match.
Nvidia’s CUDA software platform remains a strong competitive barrier. Enterprise customers rely on the mature tools and compatibility that CUDA provides.
These factors help Nvidia maintain its leading position even as rival chips improve their cost efficiency. The company’s established developer ecosystem creates switching costs for customers.
Amazon’s Trainium chips and AMD’s accelerators show different performance characteristics. These solutions have achieved only about 30% cost reduction compared to previous generations.
Both Amazon and AMD chips lag behind Nvidia and Google offerings on absolute cost metrics. This performance gap keeps them at a disadvantage in the current market.
Broadcom’s Growing Role
Goldman sees Broadcom benefiting from multiple trends in AI infrastructure. The company’s partnership with Google on TPU development positions it to gain from custom silicon demand.
Broadcom also stands to profit from AI networking requirements. Advanced AI models need increasingly sophisticated interconnects for data transfer.
The investment bank based its preferences on sustainable AI capital expenditure patterns. Both Broadcom and Nvidia are tied to core infrastructure spending that Goldman expects to continue.
Goldman outlined four potential scenarios for AI evolution in its research. The analysis examined how different technology paths might develop over time.
AMD trades at a P/E ratio of 121.77 according to market data. The company showed 31.83% revenue growth over the last twelve months.
AMD is forecast to grow revenue by 32% in fiscal year 2025. The stock gained 90.87% over the past year despite trailing in the cost comparison.
Riot Platforms recently signed a 10-year agreement with AMD for data center services. The deal involves initial deployment of 25 MW capacity at Riot’s Texas facility.
Wells Fargo named AMD its top stock pick for 2026 citing AI potential. RBC Capital initiated coverage with a sector perform rating and $230 price target.
Get 3 Free Stock Ebooks
Discover top-performing stocks in AI, Crypto, and Technology with expert analysis.
- Top 10 AI Stocks - Leading AI companies
- Top 10 Crypto Stocks - Blockchain leaders
- Top 10 Tech Stocks - Tech giants
