Agentic artificial intelligence (AI) is the next major step in the AI revolution. It allows users to develop AI agents that can perform tasks on their behalf with limited human direction. While the most widely used applications of this technology now involve coding, it could easily expand to other areas over the next few years. Getting in front of trends like this can be a profitable pursuit for investors, so now would be a good time to identify the companies with most the promising chances to capitalize on AI agents before they really hit the mainstream.
The three stocks I'm most bullish on in this space are Nvidia (NVDA +2.59%), Broadcom (AVGO +4.69%), and Alphabet (GOOG 0.20%) (GOOGL 0.41%). Although these three have been huge winners in the AI race already, winners tend to keep on winning, and I think that's exactly what this trio will do.
Image source: Getty Images.
Nvidia and Broadcom
If you've been investing in AI since the accelerated build-out of data centers began in 2023, you're familiar with both Nvidia and Broadcom. Both provide AI computing units, but they're not tackling quite the same segments of the market. Nvidia is clearly the market leader in parallel processing chips, with not only the world's best graphics processing units (GPUs) but also an infrastructure to support them. GPUs can handle a wide variety of computing workloads and are the chips of choice for flexibility and speed.
While GPUs are incredibly impressive, though, users don't always need all of the features they offer. Sometimes, a computing unit dedicated to a specific type of task (for example, AI inferencing) may see only that one variety of workload throughout its service life. In those cases, the flexibility of a GPU would be wasted.
NASDAQ: NVDA
Key Data Points
This is the corner of the market that Broadcom is targeting with its custom AI chips. Broadcom partners with AI hyperscalers and designs application-specific integrated circuits (ASICs) to precisely meet the narrow set of computing needs the chip is expected to face. ASICs are nothing new, but Broadcom (and a few other companies) have started to adapt this concept to the AI world. What an AI-focused ASIC delivers is superior cost performance versus a GPU, in some scenarios. When the workload is properly configured, a custom AI chip from Broadcom can provide higher performance at a lower cost.
NASDAQ: AVGO
Key Data Points
The best example of a custom AI chip is Google's Tensor Processing Unit (TPU). TPUs offer incredible performance, which is one reason why Google Cloud is growing so rapidly. However, Google Cloud doesn't just give users access to TPUs; it also has servers built around Nvidia's GPUs.
As AI agents become more mainstream, they will need more computing hardware from both Nvidia and Broadcom to support them. This will allow both companies to keep growing at their already elevated paces. So, just because these two have grown by a ton and become massive companies, that doesn't mean they're done. I think the future is bright for both of them, and they will continue to deliver incredible shareholder returns.
Alphabet
Alphabet just launched one of the lowest-cost AI agent models: Gemma 4. Its cost? Free. So, how can Alphabet offer a free tool and hope to make money on it? It's all about the future.
NASDAQ: GOOGL
Key Data Points
Alphabet has removed the barrier to entry to creating AI agents. Anyone can do it for free using Gemma 4. However, a user may reach a limit to what these free agents can do, which opens the door to paid-for products. If Alphabet can capture a large portion of the curious audience with its free invitation to build an AI agent, it will get people used to using its tool, and likely capture long-term customers for its paid service when they want to go beyond the basics.
This is a clever approach that sets Alphabet apart from some of its start-up competitors that need their agentic AI offerings to become profitable sooner if they are going to be viable. Alphabet can afford to play a long game here. It's one of the top companies in the AI agent space, and I think it will provide the building blocks that many companies construct their AI agents on.
