2 days ago
4
Geoffrey Seiler, The Motley Fool
Fri, May 15, 2026 astatine 10:20 AM CDT 5 min read
One of the astir important lessons an capitalist tin larn is to beryllium adaptable. It's casual to get caught up successful a thesis oregon mindset and privation to instrumentality with it nary substance what. However, it is important to beryllium flexible arsenic companies and industries germinate and caller accusation presents itself.
This holds existent for some buying and selling. When the accusation changes, beryllium acceptable to reevaluate and alteration with it. One banal I've done this with precocious is Advanced Micro Devices (NASDAQ: AMD). For the past fewer years, I thought of AMD arsenic a distant No. 2 subordinate successful the graphics processing portion (GPU) marketplace that was improbable to ever go an important subordinate successful the artificial quality (AI) infrastructure build-out.
Will AI make the world's archetypal trillionaire? Our squad conscionable released a study connected a little-known company, called an "Indispensable Monopoly," providing the captious exertion Nvidia and Intel some need. Continue »
However, a batch has changed implicit the past six months. Let's spot however -- and what this means for investors.
Riding almighty trends
AMD has ever been stuck down Nvidia and has faced a steep uphill ascent to vie successful the GPU marketplace owed to its subpar bundle ecosystem that often made its GPUs unusable retired of the container for AI exemplary training. However, aft much than 2 years of investment, the institution has vastly improved its ROCm bundle platform, portion the displacement of bundle developers to open-source AI frameworks has helped unfastened the door, particularly successful inference. Importantly, inference is expected to yet go a overmuch bigger marketplace than ample connection exemplary (LLM) grooming implicit time.
This helped the institution onshore 2 large hyperscaler (owners of ample information centers) customers for its GPUs implicit the past respective months. Both OpenAI and Meta Platforms person agreed to commitments of 6 gigawatts worthy of chips successful the coming years, estimated to beryllium worthy much than $100 billion. While AMD provided some companies with warrants worthy up to 10% of the company, based connected deliveries and AMD banal hitting definite terms triggers, it got the institution into 2 of the biggest AI infrastructure spenders and ensured they would request to instrumentality ROCm wrong their ecosystems. Given the imaginable size of the inference market, this trade-off looks similar a astute move.
AMD's GPUs, meanwhile, are good positioned for the inference market. The institution uses a modular chiplet plan that packs successful much representation capacity. Inference is often much memory-bound than compute-bound, meaning show is constricted by however rapidly information tin determination from representation to the processor, not however almighty the GPU is. As such, AMD's caller MI450 series, which has 1.5 times the representation capableness of Nvidia's upcoming bid of Rubin chips, looks similar it has an vantage successful this area, astatine slightest connected paper.
English (CA) · 
English (US) · 
Spanish (MX) ·