Amazon’s push to make AI cheaper — and why it matters to reigniting the stock
Amazon ‘s cloud leadership — and likely the next move in the stock — hinges on whether growth on that side of the business continues to reaccelerate. Custom AI chips are key to that effort. Investors widely view shares of Amazon, which also dominates in e-commerce, as driven primarily by the fortunes of its Amazon Web Services cloud unit, which generates a majority of the company’s operating income. That makes AWS performance the single most important factor for Amazon’s market value, which remains nearly $2.54 trillion despite being the worst Magificent Seven last year. And, AWS will be what Wall Street focuses on Thursday evening when Amazon reports its fourth quarter 2025 results and looks ahead to what to expect this year. A large part of Amazon’s strategy to further ramp up AWS growth is its own artificial intelligence custom Trainium chips. Launched in late 2020, the latest version, Trainium3, was unveiled back in December at the AWS Re:Invent cloud conference. The Trainiums are designed in-house instead of relying solely on third parties. Amazon is not alone. Last week, Microsoft announced its next-generation custom AI chip, the Maia 200, roughly two years after the first iteration. Google has custom AI chips, called tensor processing units, which are getting a lot of attention since the November launch of Gemini 3 as the AI model to beat. Gemini 3 was trained by TPUs, which were co-designed by Broadcom . By building its own chips, Amazon, like its cloud competitors, aims to deliver on two key advantages when running AI workloads: lower computing costs for customers and lower electricity use for data centers. Both matter more than ever because AI models are getting larger and data centers are consuming record levels of electricity. According to AWS vice president David Brown, price performance, or how much compute a customer gets for every dollar spent, is now the key metric customers care about. “If they can find a chip and a processor that allows them to get more performance for fewer dollars, well, that’s a very strategic advantage for their business,” Brown said in an interview with CNBC. That “strategic advantage” goes both ways for Amazon at a time when growth at Microsoft’s Azure cloud and Google Cloud — the second and third largest clouds, respectively, behind AWS — is comparatively on fire. For sure, the law of large numbers is at play here, with calendar year 2025 estimated AWS revenue growth of 19.1% to $177.78 billion, according to FactSet, versus Azure growth of 26.1% to $120.85 billion and Google Cloud growth of 35.8% to $58.71 billion. But there is no denying that Microsoft and Google are gunning for AWS, which does not want to cede its crown anytime soon. Spending guidance balanced against growth will also be something to watch from Amazon, as Alphabet on Wednesday evening reported a great quarter and a huge capital expenditure guide . Last week’s earnings report from Microsoft also indicated massive AI capex . AWS reacceleration optimism follows a 20.2% year-over-year revenue growth rate in the third quarter, which topped expectations of 18.1% and marked the first return to 20% plus growth since 2022. Amazon shares jumped 9.6% in the session after those Q3 numbers were released late on Oct. 30, and another 4% on Nov. 3 to their latest record high close of $254. The stock moved lower from there, losing more than 8.5% as of Wednesday afternoon trading. AMZN MSFT,GOOGL 5Y mountain Amazon, Microsoft, Alphabet: 5-year stock performance AWS bet on custom chips AWS’ custom chip push dates back to its 2015 acquisition of then-startup Annapurna Labs, which became the foundation of its in-house silicon effort. Since then, AWS has built key hardware platforms, including Graviton CPUs (central processing units) and AI accelerators like Trainium and Inferentia (more like Nvidia GPUs) — all designed specifically for cloud and AI workloads. Custom chips, also called application-specific integrated circuits (ASICs), are aimed at reducing hyperscalers’ reliance on Nvidia hardware, which has been the all-purpose AI gold standard. Nvidia’s GPUs (graphics processing units) are in short supply and expensive. For Nvidia’s part, CEO Jensen Huang recently said he was not worried about custom chips. Shortly after the November launch of Google’s Gemini 3, Jensen told Jim Cramer, “What Nvidia does is much more versatile. … Nvidia can address markets that are much, much broader, not just chatbots.” Notwithstanding, Trainium’s cost advantage is meaningful in a market still dominated by Nvidia’s premium chips, according to Brad Gastwirth, global head of research and market intelligence at Circular Technology. “Nvidia is charging astronomical numbers for their silicon,” said Gastwirth, who sold his prior company to Wedbush, where he continued for a couple of years as a chief technology strategist, before branching out on his own again. Since custom chips are much cheaper, AWS can offer lower prices to customers. “That’s a big benefit,” he added. Gastwirth explained why Trainium’s specialization allows it to run certain AI models more efficiently and at a lower cost. “You can have it run a model exactly for what you want to run it for. … If you build something specific for your needs, you can save a tremendous amount of money instead of buying a high-powered GPU that does way more than what your needs are.” At companies as big as Amazon, according to Gastwirth, the economics improve over time. While custom chips require upfront investment, they become cheaper to operate at scale. There are very few things that a GPU is able to do that something like a Trainium accelerator can’t do. AWS Vice President David Brown AWS’ Brown reinforced this point, including that the performance gap between GPUs and custom accelerators is narrowing. “When it comes to AI, there are very few things that a GPU is able to do that something like a Trainium accelerator can’t do,” Brown said, pointing to the growing ability of purpose-built chips to handle large-scale training and inference tasks customers care about. Customers — including AI startup Anthropic, behind the chatbot Claude — using the latest version are aiming to reduce training and inference costs by up to 50%. In turn, that not only supports higher usage on AWS, but it also improves Anthropic’s economics, which seems to be showing up in its forecasts. The Information reported last week that Anthropic increased its internal revenue projections to at least $17 billion in 2026, up from the previous $15 billion estimate. For 2027, the AI startup estimates $46 billion, up from the prior estimate of $39 billion. As AWS’s largest cloud partner, Anthropic’s cost-efficient growth should help AWS reaccelerate growth. What the Street is watching In an interview with CNBC, Roth MKM analyst Rohit Kulkarni said AWS is “positioned to do quite well in this new world of AI cloud,” as a growing number of customers seek out AWS in-house silicon. He cited Amazon disclosing a multibillion-dollar revenue run rate for the Trainium chip line, with more than one million chips in production and over 100,000 customers. Roth MKM also wrote in a note last month that Trainium chips are “focused on a category of compute that has arguably a larger addressable market opportunity.” In turn, the analysts, who have a buy rating on Amazon stock, raised their price target to $295 per share from $270. They said they would see it as a positive if cloud revenue exceeds 22% in the upcoming quarter. The FactSet consensus estimate calls for 21.7% growth in Q4. That’s up sequentially, and well above a year-ago’s advance of 18.9%. In a similar light, Mizuho said Trainium availability “should drive more AI revenue,” adding the expansion of inference workloads — which Trainium chips are designed for — “should drive a broader revenue base for AWS.” The Mizuho analysts forecasted AWS revenue growth to accelerate 23% in 2026. They cited the resurgence of AWS as the “primary driver” of why Amazon shares are positioned for multiple expansion this year. Mizuho has a buy rating on Amazon stock with a $285 price target. To be sure, AWS’ custom silicon strategy, which is still in its early innings, has also come with some bumps in the road, according to Baird. The firm said in a note late last month that AWS’ focus on in-house chips “could have long-term benefits from higher margins as well as less dependency on third parties,” while noting that scaling the strategy across customers and workloads takes time. The Baird analysts cited early friction surfacing as customers diversify workloads, citing “growing pains” such as “forced adoption” of AWS proprietary stacks, and Anthropic also placing some workloads on Alphabet ‘s Google Cloud. Baird also said that AWS is “stepping up orders for off-the-shelf chips,” including Nvidia’s Blackwell, to help alleviate near term capacity bottlenecks as custom silicon production ramps. The Baird analysts have an outperform rating on Amazon stock and a price target of $285. Bottom line It’s likely that AWS quarterly results could either make or break the stock in the near term as investors watch closely to see whether the AWS reacceleration story continues. While the cloud giant has the capacity to deliver the growth investors are anticipating, Jim Cramer is “not sponsoring Amazon right now,” he said Tuesday during the Morning Meeting. He added the cloud and e-commerce giant could be “lumped in with software” — a sector that’s hurting due to fears that AI will disrupt enterprise software business models. Ahead of Thursday evening’s earnings, the Club has an Amazon price target of $275 per share. While we have a buy-equivalent 1 rating on the stock, Jim’s wait-and-see approach so close to the print is just good discipline. (Jim Cramer’s Charitable Trust is long AMZN, NVDA, GOOGL. See here for a full list of the stocks.) As a subscriber to the CNBC Investing Club with Jim Cramer, you will receive a trade alert before Jim makes a trade. Jim waits 45 minutes after sending a trade alert before buying or selling a stock in his charitable trust’s portfolio. If Jim has talked about a stock on CNBC TV, he waits 72 hours after issuing the trade alert before executing the trade. THE ABOVE INVESTING CLUB INFORMATION IS SUBJECT TO OUR TERMS AND CONDITIONS AND PRIVACY POLICY , TOGETHER WITH OUR DISCLAIMER . NO FIDUCIARY OBLIGATION OR DUTY EXISTS, OR IS CREATED, BY VIRTUE OF YOUR RECEIPT OF ANY INFORMATION PROVIDED IN CONNECTION WITH THE INVESTING CLUB. NO SPECIFIC OUTCOME OR PROFIT IS GUARANTEED.
<
