AWS CEO Matt Garman Says AI Inference Isn’t ‘Even 1%’ of What It Will Become. AMZN Stock Is 1 of 3 Stocks to Buy.

In a recent interview, Amazon (AMZN) Web Services CEO Matt Garman said that inference, the computing used to run AI models, is less than 1% of what it will eventually become. “I don’t think it's even 1%,” he said. Garman stated that every time inference prices have dropped by half, usage has jumped about tenfold. So instead of price cuts hurting AWS' revenue, they’re doing the opposite. Cheaper AI makes companies use it for many things, so the overall market keeps growing.

How Amazon, AMD, and Broadcom Win From Inference

Amazon: Inference is a massive part of Amazon. CEO Andy Jassy said inference already makes up well over half of AWS' AI revenue, and that business is booming. AWS is already at a $184 billion run rate, up 39.3%. I laid out the math recently, and at that pace, it would put the business close to the trillion-dollar mark Jassy is targeting within about five years. The AWS backlog has now reached a whopping $578 billion. The company's Trainium chips handle most of the inference traffic on its Bedrock AI service. Garman said that they can save customers 20% to 30% on inference costs, and they're sold out through the end of next year. This shows how strong the demand is.

AMD: Some of AMD's (AMD) biggest customers mainly use its chips for inference. OpenAI and Meta (META) each committed to 6 gigawatts of AMD chips, while Anthropic (ANT) agreed to deploy up to 2 gigawatts. On AMD's Q1 call, CEO Lisa Su said that while its GPU deals cover both training and inference, “the largest deployments are for inference.” Its MI450 chip has about 50% more memory than Nvidia's upcoming Vera Rubin. More memory lets one chip hold a large AI model on its own, instead of splitting it across several chips. This makes answering requests faster and cheaper. AMD also benefits through CPUs. AI agents run on inference, and while GPUs run the models, CPUs handle the surrounding work. As I covered when Lisa Su said the real CPU demand wave hasn't started, that's already pushing data centers toward one CPU for every GPU, up from one for every two to four. And AMD earns close to half of all server CPU revenue.

Broadcom: Broadcom (AVGO) designs custom AI chips with the biggest AI companies, including Google's (GOOG) (GOOGL) TPUs, OpenAI's Jalapeno, Meta's (META) MTIA, and chips for Anthropic. These custom chips are built for one specific job, and that makes them a great fit for inference, since it's the same task repeated millions of times. A chip built just for that can answer each request more cheaply than a general-purpose GPU. CEO Hock Tan said OpenAI's first custom chip, Jalapeno, outperforms Nvidia's Grace Blackwell GPU on inference. Meta's custom chip is also built for inference at scale, and Broadcom is expected to start shipping it in the fiscal fourth quarter.

AMD Has Run Furthest, Broadcom Has Lagged

All three stocks have hit their all-time highs this year, but they've moved very differently since. AMD has been the clear winner, surging about 162% in 2026, far ahead of the S&P 500's ($SPX) roughly 11% gain. Strong demand for its server chips from AI agents and big deals with Meta and OpenAI pushed it to a record in October. Since then, it has dropped about 15% after a series of chip selloffs. AMZN stock is up about 16% this year and sits just 7% below its high, helped by AWS posting its fastest growth in years. AVGO stock has been the weakest of the three, up less than 2% this year and 29% below its peak. The stock slid after its June earnings, and worries about AI spending, along with the recent chip selloff, added to the pressure. So if Garman is right about inference, Broadcom might be the cheapest way to get in right now.

Broadcom Looks Like the Best Value of the Three

Amazon's forward price-to-earnings (P/E) of 27.80x looks far below its 5-year average of 158.13x, but that average is inflated by years when it barely made a profit. Its forward price-to-sales (P/S) ratio of 3.48x sits about 21% above average. EPS is expected to dip in 2027 after this year's boost from Anthropic gains, but the P/E still falls to about 15x by 2029.

AMD is the most expensive of the three. Its forward P/E of 62.89x is about 56% above its 5-year average, and its P/S ratio of 15.97x is about 81% above. The EPS helps explain that premium. Analysts expect earnings to grow 94% this fiscal year and more than double next year despite being compared to a really high base. That brings the P/E down to about 29x in 2027.

Broadcom looks like the best value. Its forward P/E of 27.39x is right in line with its 5-year average, and its forward P/S of 14.39x is just 9% above its average. The modest premium looks attractive next to its EPS trajectory. Analysts expect earnings to grow over 55% in each of the next three fiscal years. That brings its P/E down to about 10.5x by fiscal 2028. With strong growth and not much of a premium, I think Broadcom stands out among the three.

What Do Analysts Expect for AMZN, AMD, and AVGO Stocks?

As per Wall Street analysts, all three of these stocks hold a “Strong Buy” rating. AMZN stock’s mean price target of $328.33 indicates 23% upside, which is smaller than it is for the other two, since AMZN sits much closer to its 52-week high than AMD or AVGO.

AMD has a mean price target of $660.37, indicating 19% upside. Its highest price target of $1,250 indicates a massive 126% upside, while the lowest price target of $460 is only 11% below where the stock currently sits.

AVGO stock’s mean price target of $525.68 indicates the biggest upside of the three, at 48%. Even its lowest price target of $380 is well above the current share price. This is in line with what I said about AVGO stock standing out among the three. 

On the date of publication, Jabran Kundi did not directly have a position in any of the securities mentioned in this article. All information and data in this article is solely for informational purposes. For more information please view the Barchart Disclosure Policy here.