These Super Stocks Could Be the Biggest Winners in the AI Inference and Agentic AI Economy

The archetypal signifier of the existent artificial quality (AI) roar was each astir grooming the foundational ample connection models (LLMs) needed to assistance reply queries and execute tasks. Training these AI models required a ton of processing power, which is wherever Nvidia (NASDAQ: NVDA) and its graphics processing units (GPUs) shone.

Not lone are Nvidia's GPUs powerful, but the institution besides smartly seeded its CUDA bundle level successful places wherever aboriginal AI probe was being done. This led to astir aboriginal foundational codification being written connected its bundle and optimized for its chips, giving it a immense moat and making it the biggest AI winner.

While AI exemplary grooming remains important, this AI-focused marketplace is starting to displacement toward inference and agentic AI. New winners volition look arsenic factors specified arsenic memory, cardinal processing units (CPUs), and outgo per inference go much important. Let's look astatine 3 AI stocks that could go the adjacent large AI winners.

AMD: An agentic AI and inference victor

The displacement toward agentic AI and inference plays to Advanced Micro Devices' (NASDAQ: AMD) strengths. First, some necessitate overmuch higher CPU-to-GPU ratios: grooming astatine astir 1:8, inference astatine astir 1:4, and agentic AI astatine 1:1. AMD is the person successful the high-performance information halfway CPU market, truthful this is simply a immense tailwind for the company. On apical of that, information halfway CPU prices should besides beryllium connected the rise, arsenic agentic AI needs much cores (sort of similar idiosyncratic workplaces with the chips), which should rise prices.

AMD's chiplet GPUs, meanwhile, battalion successful much memory, which makes them perfect for inference, which tends to beryllium much memory-bound than processing-bound. With ample GPU inference deals successful place, the institution is acceptable to beryllium a victor successful inference and agentic AI.

Micron: A HBM person

As noted above, inference tends to beryllium much memory-bound than processing-bound, which is large quality for dynamic random entree representation (DRAM) shaper Micron Technology (NASDAQ: MU). The DRAM marketplace is already undersupplied, which is driving up prices, and the marketplace displacement to inference should lone support request booming.

In general, GPUs request to beryllium packaged with a peculiar signifier of DRAM called high-bandwidth representation (HBM) to execute optimally, and inference uses adjacent much of this memory. While static random-access representation is simply a competing technology, it is physically ample and expensive. With AI spot request continuing to surge, the outlook for HBM and DRAM remains strong. Meanwhile, the large 3 DRAM makers -- Micron and Korean companies Samsung and SK Hynix -- person started to fastener successful longer-term (three-to-five-year) deals, marking a large displacement for the representation industry.