In internal documents, Amazon warned that AI startups are delaying spending on AWS

• AI startups spend first on models and tools, delaying traditional AWS cloud spend, document warned.
• Cursor’s traditional cloud budget is below 10% of what it spends on new AI categories: document.
• AWS is facing scrutiny over high prices and a perception that it is lagging behind AI competitors.

For years, Amazon Web Services was among the first, and often the largest, line item on a startup’s budget. The AI boom is upending that spending pattern.

Internal documents obtained by Business Insider reveal that AWS has flagged a “fundamental” shift in how startups are allocating their cloud budgets. Increasingly, they’re delaying AWS cloud adoption and diverting spending toward AI models, inference, and AI developer tools.

Instead of pouring money into traditional cloud services like compute and storage, these companies are spreading costs across newer AI technologies that are easier to switch between, according to the documents.

“Founders tell us they seek to adopt AWS at a later stage,” one of the documents warned.

This suggests a seismic shift is happening in the cloud industry. AWS’s dominance was built on startups that embraced its affordable, scalable computing services as an alternative to running their own data centers. But the generative AI boom has ushered in a new era, a “Cloud 2.0” stack of specialized hardware, software, and tools. As startups use these new AI offerings first, and wait longer to spend on AWS, the cloud giant’s once-firm grip on this lucrative ecosystem could begin to loosen.

To be sure, startups are not abandoning AWS entirely, as they still need its cloud services at later stages. However, their behavior highlights how emerging AI technologies are capturing early IT budgets in new ways and potentially locking in customers before AWS’s traditional offerings come into play.

‘Confidential’ documents

The internal documents cited in this story, marked “Amazon Confidential,” are from March and July. They were written by employees on AWS’ startup business team, including a person who works closely with Y Combinator startups. The documents were also reviewed by AWS executives who manage startup and venture capital relationships. BI verified the identities of these people. Jon Jones, former VP of global startups and venture capital at AWS, is also listed on the March document as one of the business owners responsible for this part of Amazon’s cloud operations.

In addition, at least three current and former employees familiar with AWS’s startup business told Business Insider that the concerns expressed in these internal documents were still valid as of September. These people asked not to be identified as they’re not authorized to talk to the press.

Before publication, an AWS spokesperson said this story was using “old data to reach outdated conclusions.” Startups continue to build on AWS, including leading AI startups such as Perplexity and Luma AI, which “recently chose” AWS, the spokesperson added.

“AWS remains the top choice for startups to build because we offer the best core services as well as the most innovative and powerful generative AI offerings,” the spokesperson wrote in an emailed statement. “Early stage startups experiment with many services and technologies, but when it comes time to choose the provider they trust with the future of their organizations, they overwhelmingly choose AWS.”

Delayed adoption

Many AI startups now make their first technology purchases from AI model providers such as OpenAI and Anthropic, followed by newer developer platforms such as Vercel, according to one of the AWS documents. That means founders are putting off decisions to buy AWS services until later, often when they require advanced capabilities such as compliance and security, the document explained.

Among Y Combinator’s 2024 cohort, 59% reported using more than three AWS services, down by more than four percentage points from 2022, according to the document from March. Meanwhile, 88% of these startups were using OpenAI’s models and 72% were using Anthropic’s. Only 4.3% said they were using AWS’s Bedrock developer tool, which gives access to various AI models.

The AWS spokesperson said this metric is “at least a year old, and it’s not indicative of usage or adoption of AWS.”

Are AI startups “all in” on AWS?

Earlier this year, AWS compiled a list of the top 1,000 AI startups for Amazon CEO Andy Jassy, along with an assessment of whether they were building primarily on the tech giant’s cloud platform, according to the July document. The goal: What does it mean to be “all in” on AWS in the AI era?

With cloud spending rapidly growing beyond “traditional compute,” storage, databases, and analytics, “this question is getting much harder to answer,” AWS employees wrote in the July document.

New AI categories of spending “can represent the majority of a startup’s cloud consumption and are much less sticky than traditional services, allowing rapid shifts across an expanding list of providers,” they warned in the document.

The AWS employees cited three newer AI cloud services that were grabbing early spending from these startups: GPU training and fine tuning, GPU inference, and AI-as-a-service, according to the document.

GPUs are special chips that power generative AI. That’s a contrast to traditional cloud services, which run on CPUs. Training and fine tuning are ways to build and improve AI models, while inference is how models are run. AI-as-a-service provides access to models and other AI tools, usually via application programming interfaces (APIs) and subscriptions.

Cursor’s spending

The AWS employees cited AI coding startup Cursor as an example of these issues, according to the July document. Cursor’s spending on “traditional infrastructure” was less than 10% of what it spends on newer AI categories, even though the startup is considered “all in” on the AWS platform.

The “majority” of Cursor’s spending goes toward API calls to external AI models and “neocloud” providers that primarily sell access to GPU servers, the document noted. It didn’t mention specific neoclouds, but some of these newer AI cloud vendors include CoreWeave, Crusoe, Lambda Labs, and Nebius.

The AWS spokesperson told Business Insider that the concerns around “all-in” customers were “false,” without providing further details. Business Insider followed up on Thursday to ask for more specifics, but didn’t get a response. A Cursor representative didn’t respond to a request for comment.

“A step behind”

To be sure, these documents are from March and July, and AWS may have turned things around with AI startups in recent months. However, revenue growth from this key part of Amazon has lagged behind some other cloud providers lately, on a percentage basis.

In the second quarter, Google Cloud and Microsoft’s Azure each grew more than 30%, year-over-year, while AWS grew 18%. Neocloud revenue has soared by more than 200% in the past year, albeit from a much lower base, according to Synergy Research Group.

“With Azure and Google Cloud Platform growing faster than AWS, the once-strong incumbent’s market position may lead to three equal players,” Theory Ventures’ Tomasz Tunguz wrote in a recent blog post.

Still, it’s early in the AI cloud battle, and Amazon has prodigious resources and advantages. The company maintains a close and promising partnership with Anthropic, having invested billions of dollars in the leading AI lab. In July, Morgan Stanley estimated that Amazon could generate $5.6 billion in revenue by 2027 from Anthropic’s use of AWS cloud services. In September, Wells Fargo upgraded Amazon to “buy” based on its belief that Anthropic will enable AWS growth to accelerate in 2026.

Investors also say that AI isn’t necessarily cannibalizing AWS’s cloud revenue. Some of the money startups are channeling into new AI services likely flows back to major cloud providers that supply GPUs and other AI cloud infrastructure. Still, AWS risks losing ground if it can’t capture those early customers as spending shifts, they said.

“AWS is still a step behind Microsoft and Google in driving GPU demand and the ability to sell add-ons to these customers,” Gil Luria, an analyst at D.A. Davidson, told Business Insider.

CB Insights data shows AWS losing some ground among the 1,100 leading AI startups. Between January 2024 and September 2025, AWS captured 30% of that market, trailing Google Cloud’s 38%, but outpacing Microsoft Azure’s 7%. That marks a decline from the prior two-year period (2022 to 2024), when AWS claimed 33%, Google Cloud 34%, and Microsoft 9%. Roughly 25% of the startups said they used more than two cloud providers.

AWS pricing grumbles

AWS’ AI pricing strategy isn’t helping it win startup customers either.

Earlier this year, AWS found that 90% of early-stage startups in Radical Ventures’ portfolio were building primarily on rival clouds, citing AWS’s higher GPU costs compared with competitors, according to one of the documents from March.

Following that discovery, Jassy and AWS CEO Matt Garman met with Radical Ventures’ leadership to craft a new strategy aimed at better targeting the firm’s startup investments and offering a more complete suite of AWS services, according to this document. A representative from Radical Ventures declined to comment.

Neocloud providers, such as CoreWeave, which specialize in GPU-based computing, may be starting to emerge as serious competitors for AWS. One of the documents noted growing demand among customers for access to “small increments of GPU capacity” with pay-as-you-go flexibility, an area where AWS is at a “disadvantage” compared with neoclouds.

Frustration with AWS’s pricing has also surfaced publicly. Gavin Baker of Atreides Management recently wrote on X that AWS had raised prices for Nvidia’s Blackwell chips. In response, Chamath Palihapitiya of Social Capital wrote that Amazon had become “too expensive,” adding that his portfolio company, 8090, had switched to using chips from Groq, another one of his investments.

The AWS spokesperson told Business Insider that the company is “always working to optimize products and services for our customers,” adding that it recently cut the price of EC2 Nvidia GPU-accelerated instances by 45%.

AWS also acknowledged in one of the documents that there’s a notable shift toward “industry-specific AI adoption,” exemplified by the growth of startups such as Harvey in legal tech and Lila Sciences in biotech. AWS expected this trend to accelerate as AI moves deeper into specialized applications and intelligent agents, the document noted.

The AWS spokesperson said the company is seeing growth across “all AI startups,” from foundational models to vertical and horizontal solutions leveraging AI.

‘Playing catch up’

Another challenge for AWS is its lagging reputation in AI, which is making it increasingly difficult to secure speaking slots for executives at venture-capital events and other industry conferences, according to one of the documents. The issue is particularly pronounced in the Bay Area, where many of those opportunities are concentrated, it stated.

“2.5 years post the launch of ChatGPT, AWS is still viewed as playing catch up in AI by many public/private investors, founders, and industry influencers,” the document warned.

The AWS spokesperson told Business Insider this is “false,” without providing further details.

This skepticism isn’t limited to private investors. During Amazon’s July earnings call, Morgan Stanley analyst Brian Nowak questioned Jassy about perceptions that AWS is falling behind in AI and losing market share to competitors. Jassy’s answer did little to reassure investors, and Amazon’s stock slipped further during the call.

AWS is also struggling to identify and engage early-stage startups that later evolve into major cloud customers. One of the internal documents noted that AWS’s VC-driven discovery model isn’t well-suited to the rise of “AI-native solopreneurs and bootstrapped teams.” To address this, the company plans to develop a data-driven prediction model to better surface promising startups early on.

“This blind spot poses increasing risk to cloud market share,” it stated.

The AWS spokesperson said the company continues to engage founders as early as possible, in collaboration with VCs through programs such as AWS GenAI Accelerator and AWS Activate.

Several Amazon employees told Business Insider that AWS’s startup leadership team lacks deep experience in the venture ecosystem, which may be compounding the problem. The challenge was underscored by the abrupt departure of Jon Jones, AWS’s vice president of startups and venture capital, who resigned last month after a year in the role.

Here’s the full statement from the AWS spokesperson:

“Business Insider is using old data to reach outdated conclusions that don’t reflect the way that startups are operating today. AWS remains the top choice for startups to build because we offer the best core services as well as the most innovative and powerful generative AI offerings. That’s why leading startups like Perplexity, Luma AI, Writer AI, poolside, Latent Labs, Datology, and others all recently chose AWS. Early stage startups experiment with many services and technologies, but when it comes time to choose the provider they trust with the future of their organizations, they overwhelmingly choose AWS—which is why over 85% of the CNBC Disruptor 50, over 85% of the Forbes AI 50, over 75% of startups from the CBI AI 100, and over 70% of the TechCrunch US AI Startups lists all choose AWS.”

Have a tip? Contact this reporter via email at [email protected] or Signal, Telegram, or WhatsApp at 650-942-3061. Use a personal email address, a nonwork WiFi network, and a nonwork device; here’s our guide to sharing information securely.

Read the original article on Business Insider

The post In internal documents, Amazon warned that AI startups are delaying spending on AWS appeared first on Business Insider.