AWS is more or less on track

Welcome to Runtime! Today: AWS narrowly misses analyst estimates if you're not into the rounding thing, Stanford and UW researchers think they've found a way to out-DeepSeek DeepSeek, and the latest enterprise moves.

(Was this email forwarded to you? Sign up here to get Runtime each week.)

Round numbers

Technically speaking, each one of the Big Three cloud infrastructure providers missed Wall Street's expectations for their fourth-quarter 2024 revenue numbers. But after several years during which it seemed to be far behind its challengers, AWS came the closest to meeting those expectations.

AWS reported quarterly revenue of $28.8 billion on Thursday, an increase of 19% over 2023's fourth quarter. That was in line with expectations according to CNBC's Amazon earnings preview posted before the results were released, but Jim Cramer Inc. decided to go to a second decimal place for an AWS-centric post-release report, noting that $28.786 billion in revenue is not as good as the $28.84 billion those analysts were apparently really expecting.

• AWS is now on a $115 billion annualized revenue run rate, Amazon CEO Andy Jassy said during a call with analysts after the release of the results.
• "Though we expect growth will be lumpy over the next few years as enterprise adoption cycles, capacity considerations, and technology advancements impact timing, it's hard to overstate how optimistic we are about what lies ahead for AWS's customers and business," Jassy said.
• Still, investor reaction — inspired in part by a weaker-than-expected forecast for that whole other retail part of Amazon's business — sent the stock down more than 4% in after-hours trading.

AWS growth remains well behind its Big Three counterparts, both of whom recorded growth slightly over 30% during the last quarter, but AWS is growing off a much larger number. Jassy suggested that like Microsoft and Google, AWS is running into external headwinds that limit that growth.

• All of the hyperscalers are struggling with data-center construction as power supplies remain "constrained," in Jassy's words, and local opposition heats up.
• But Jassy also threw a little shade at either Nvidia or Intel during the Q&A session with financial analysts, suggesting that the chip makers have run into production problems in recent months.
• "It's true that we could be growing faster if not for some of the constraints on capacity, and they come in the form of chips from our third-party partners coming a little bit slower than before, with a lot of midstream changes and taking a little bit of time to get the hardware actually yielding the percentage healthy and high quality service we expect," Jassy said.
• That shade was wrapped around a pitch for AWS's homegrown silicon, so apply as many grains of salt as desired, but The Information reported last month that servers powered by Nvidia's newest Blackwell chips have run into overheating problems, and Intel announced during its earnings call last week that it was pushing out its next-generation Xeon CPU and killing its next-generation data-center GPU.

Amazon spent $26.3 billion on capital expenditures during the fourth quarter, which CFO Brian Olsavsky said was "primarily" related to AWS but also includes Amazon's vast network of retail fulfillment centers. "We think that run rate will be reasonably representative of our 2025 capital investment rate [and] similar to 2024 the majority of the spend will be to support the growing need for technology infrastructure," he said.

• Capital expenditures are at the top of Wall Street's mind after DeepSeek's low-cost models blew everyone's mind last month, but Jassy appeared to agree with Microsoft CEO Satya Nadella that low-cost models will only increase the opportunity for generative AI over time.
• "One of the interesting things over the last couple weeks is sometimes people make the assumption that if you're able to decrease the cost of any type of technology component — in this case, we're really talking about inference — that somehow it's going to lead to less total spend in technology. And we have never seen that to be the case," he said.
• However, there's still a long way to go: in his prepared remarks, Jassy acknowledged that "there aren't that many generative AI applications at large scale yet," and all the hyperscalers will be chasing those large-scale generative AI apps in 2025.

Looking at you, Cal

If recent efforts to build low-cost but high-performance large-language models have staying power, foundation model companies are in for a reality check this year. Researchers from Stanford, the University of Washington, and AI2 published a paper last week detailing how they built a model for less money than it takes to go to one of either school's football games.

TechCrunch reported that the s1 model (are we forever stuck with these naming conventions?) compares favorably to some of the newer "reasoning" models from OpenAI and DeepSeek, and the researchers made the code and training data available on GitHub. However, s1 wasn't built from scratch; it was "distilled" from Google's Gemini 2.0 Flash Thinking Experimental, according to the report.

Distilled models aren't going to push the LLM industry forward, by definition, but they do suggest that breakthrough open-source models could be replicated by companies for less money than they're spending on coffee after imposing return-to-office mandates. Closed-model makers are likely to try and find a legal way to fight distillation of their models, but as open-source alternatives get good enough, it might not matter.

Enterprise moves

Robin Washington is the new president and chief operating and financial officer at Salesforce, a "newly created role" for the longtime member of Salesforce's board of directors.

Hemanth Vedagarbha is the new president of Firebolt, joining the data warehouse company after serving in sales leadership roles at Confluent and Oracle.

Lyndsey Valin is the new chief customer officer at SugarCRM, following similar roles at PROS and Vertafore.

The Runtime roundup

Softbank is in talks to acquire Arm server chip designer Ampere for around $6.5 billion, according to Bloomberg.

Cloudflare beat Wall Street's estimates for revenue and profit, and its stock soared nearly 12% in after-hours trading.

Workday laid off 1,750 people, or about 8.5% of its workforce, as it shifts toward a more AI-centric strategy.

Databricks acquired BladeBridge, which helps companies move between data warehouses, for an undisclosed sum.

Federal government agency CIOs will be reclassified as political appointees as part of the Trump administration's push to install loyalists across the civil sector, NBC reported.

Thanks for reading — see you Saturday!