For many years, Amazon Web Services—commonly known as AWS—stood as one of the first and largest recurring expenses on nearly every startup’s balance sheet. Startups routinely prioritized AWS within their technology budgets because its scalable infrastructure and pay-as-you-go pricing model allowed fledgling companies to grow quickly without building costly data centers. However, the explosive rise of artificial intelligence is now transforming those established spending patterns. Internal documentation obtained by Business Insider discloses that AWS itself has acknowledged a sweeping and intrinsic change in how young technology firms are directing their cloud expenditures. Increasingly, early-stage founders are choosing to defer the adoption of AWS cloud services, allocating their resources instead toward artificial intelligence models, machine learning inference capabilities, and the tools that support AI software development.
According to these internal analyses, rather than channeling substantial portions of their budgets into traditional cloud offerings—such as computing power, storage capacity, and relational databases—many startups are distributing funds among newer AI-centric technologies that, by design, allow greater flexibility and interchangeability among competing providers. In other words, instead of being locked into a single infrastructure vendor, companies are experimenting with modular AI platforms that can be swapped out or reconfigured with relative ease. One internal report captured this evolving mindset succinctly: company founders explained that their intent is to embrace AWS at a later, more mature stage in their evolution. This sentiment points to a profound transformation across the broader cloud computing ecosystem.
AWS’s previous dominance was built on capturing startups at the earliest phases of growth—offering affordable computing solutions that replaced the expense and complexity of proprietary servers. The emergence of generative AI, however, has initiated a new technological epoch often described as “Cloud 2.0.” This next-generation paradigm consists of a sophisticated network of specialized hardware accelerators, machine learning frameworks, and developer-oriented toolsets that cater specifically to AI-driven processes. As startups increasingly invest in these modern alternatives before turning to AWS’s conventional infrastructure, Amazon’s firm grip on what was once a reliably lucrative segment now faces gradual erosion.
It is important to note that this shift does not amount to a wholesale rejection of AWS. Founders still anticipate leveraging its mature suite of services—especially in later operational stages when demands for regulatory compliance, data security, and enterprise-scale integration intensify. Nonetheless, early spending decisions demonstrate how AI technologies are seizing the initiative, capturing first-mover advantages in startup IT budgets, and potentially solidifying customer relationships long before AWS’s standard offerings enter the equation.
The internal files at the center of this analysis—each stamped with the “Amazon Confidential” designation—date from March and July and were crafted by members of the startup-focused business team within AWS. At least one of the authors collaborates directly with Y Combinator companies, a key pipeline for emerging technology ventures. These reports were vetted by senior executives overseeing Amazon’s relationships with startups and venture capital firms, and Business Insider verified the identities of those involved. Former AWS executive Jon Jones, once vice president overseeing global startup and venture capital initiatives, was listed as one of the key business owners responsible for this division during the March report’s authorship.
Complementing this documentation, at least three current or former employees familiar with the startup division confirmed that the concerns described in these reports remained accurate as of September. They spoke on condition of anonymity due to company confidentiality requirements. In response to inquiries prior to publication, an AWS spokesperson contended that the information was based on outdated data sets and therefore led to obsolete conclusions, asserting that startups continue to rely heavily on AWS for foundational infrastructure. The spokesperson also emphasized that prominent AI innovators, including Perplexity and Luma AI, had recently selected AWS as their primary partner. Their statement reiterated Amazon’s belief that early-stage ventures experiment widely across technology providers but ultimately return to AWS when ready to choose a trusted, long-term platform—one they rely on to safeguard the continuity and future of their organizations.
Additional internal reports highlight that many recently founded AI companies begin their expenditure cycles not with AWS but with AI model providers such as OpenAI or Anthropic, later incorporating developer platforms like Vercel to manage deployment. The consequence is that decisions to purchase AWS services are frequently deferred until startups demand more complex capabilities such as enhanced compliance frameworks or enterprise-grade security tools. Supporting this observation, AWS’s own March report showed that within Y Combinator’s 2024 startup cohort, only 59% used more than three AWS services, marking a notable decline from 2022. By contrast, 88% of those same companies reported relying on OpenAI’s models and 72% on Anthropic’s. The adoption rate for AWS’s own Bedrock developer platform, which provides access to multiple AI models, was a meager 4.3%. AWS responded that these figures were at least a year old and failed to accurately depict current utilization levels.
As AI assumes a central role in technological innovation, AWS has begun examining what it now means for a startup to be “all in” on its platform. An internal July memorandum prepared for CEO Andy Jassy profiled the top 1,000 AI startups, assessing their degree of reliance on AWS infrastructure. Yet, as that document observed, categorizing such relationships has become increasingly ambiguous. Spending across modern cloud architectures extends well beyond the traditional boundaries of compute, storage, and database applications. New AI categories—including GPU-based training, fine-tuning, inference, and AI-as-a-service delivery models—frequently dominate early cost structures, but they tend to be less sticky, allowing startups to quickly migrate between providers.
The internal records even named GPU-dependent services as leading drivers of this new allocation pattern. Unlike conventional CPUs that power most cloud workloads, GPUs are essential for the complex matrix computations underlying generative AI. Training and fine-tuning processes adapt these models, while inference handles live execution. These sophisticated yet fluid segments of AI infrastructure are typically sold with flexible subscription or API-based models, emphasizing portability over long-term lock-in.
As an illustrative case study, one internal AWS document referenced the AI coding startup Cursor. Although Cursor publicly identifies as an “all-in” AWS partner, less than 10% of its computing spend is devoted to traditional AWS infrastructure components. The majority, instead, funds API calls to external AI models and “neocloud” providers offering GPU-heavy services. While the document did not name those third-party vendors outright, known players in this emergent arena include CoreWeave, Crusoe, Lambda Labs, and Nebius. AWS’s representative disputed the characterization that such customers represent a growing concern but declined to provide data refuting it.
Even as AWS works to strengthen its AI strategy, evidence suggests its financial momentum has slowed relative to competitors. In the second quarter, Google Cloud and Microsoft Azure each reported year-over-year revenue growth exceeding 30%, whereas AWS expanded by 18%. Meanwhile, neocloud revenue surged more than 200% within the same timeframe, albeit from a small starting base. Industry analysis, including commentary from Theory Ventures’ Tomasz Tunguz, predicts that these dynamics could eventually yield a more balanced three-way competition among AWS, Azure, and Google Cloud.
Despite these headwinds, Amazon remains well positioned given its extensive capital resources and its partnership with leading AI research firms, notably Anthropic—a relationship that could yield billions in upcoming revenues according to financial forecasts. Yet analysts caution that AWS may nonetheless lose strategic footing if it fails to capture startups during their earliest growth phases. For instance, D.A. Davidson’s Gil Luria observed that AWS trails Microsoft and Google in driving customer demand for GPUs and in attaching complementary services to those AI workloads.
Pricing considerations further compound AWS’s challenge. Internal findings indicated that as many as 90% of early-stage startups connected to Radical Ventures were opting for cheaper, rival clouds, citing AWS’s relatively high GPU pricing. Following this revelation, Amazon executives including Jassy and AWS CEO Matt Garman reportedly met with Radical Ventures’ leadership to recalibrate strategy and offer improved term structures. In parallel, market criticism has surfaced publicly: prominent investors such as Gavin Baker and Chamath Palihapitiya have argued that AWS’s GPU costs have become prohibitive, citing instances where portfolio companies shifted workloads to alternative chip providers.
Amazon’s representatives maintain that the company continuously refines its services for optimal customer value, pointing out that it recently implemented a 45% price reduction on NVIDIA GPU–accelerated EC2 instances. Nevertheless, internal notes acknowledged that AWS faces disadvantages compared with neoclouds when it comes to offering smaller, pay-as-you-go GPU increments, which customers increasingly prefer for flexibility.
Furthermore, AWS has recognized a growing movement of industry-specific AI adoption, with startups focusing on niche applications in fields such as law and biotechnology—symbolized by companies like Harvey in legal tech or Lila Sciences in biotech. This specialization trend is expected to accelerate as AI matures into more granular domains where intelligent systems supplement human expertise.
Yet internal commentary also suggested that AWS’s public image within the AI sector lags its rivals. Corporate insiders remarked that, more than two years after the debut of ChatGPT, many investors and founders continue to perceive AWS as catching up rather than leading in AI innovation. This perception has even hindered opportunities for AWS executives to secure key speaking positions at high-profile venture capital events, particularly in Silicon Valley. While AWS dismisses such claims as erroneous, these perceptions persist among stakeholders and recently surfaced in analyst questions during Amazon’s quarterly earnings call. CEO Andy Jassy’s responses left some investors unconvinced, prompting short-term dips in the company’s stock price.
The reports further cautioned that AWS’s methods for discovering and nurturing early-stage startups may not be keeping pace with the changing nature of entrepreneurship. The traditional venture capital–driven discovery pipeline tends to overlook independent, AI-native founders or small, bootstrapped teams that operate outside established funding networks. To address these weaknesses, AWS is developing a data-driven prediction engine designed to identify promising ventures more effectively. The internal assessment concluded that failure to do so could pose an increasing risk to AWS’s long-term cloud market share. Some insiders attribute part of the difficulty to a lack of venture ecosystem experience within the AWS startup leadership team, a deficit made more visible following the recent departure of Jon Jones, the division’s vice president, after only a year in the position.
In its official statement, Amazon reiterated that claims based on earlier data no longer reflect current realities. The company insisted that AWS continues to be the preferred choice for building and scaling startups due to its unmatched breadth of foundational cloud services and advanced generative AI tools. As supporting evidence, the spokesperson noted that leading innovators—including Perplexity, Luma AI, Writer AI, Poolside, Latent Labs, and Datology—have all recently selected AWS. The company also emphasized that overwhelmingly large proportions of elite startup cohorts, such as those listed in the CNBC Disruptor 50 and Forbes AI 50 rankings, rely on AWS. This continued alignment, the spokesperson concluded, reaffirms AWS’s enduring status as the foundation upon which emerging companies trust their futures.
Sourse: https://www.businessinsider.com/amazon-ai-startups-delaying-aws-spending-2025-10
