[Anthropic](https://www.anthropic.com) is learning that cloud partnerships come with strings attached. The AI safety-focused company is reportedly struggling to lock down enough computing resources from the very cloud providers it relies on to train and run Claude, its flagship AI assistant. The reason? Those same providers are building their own competing AI models and increasingly view Anthropic as a rival rather than a customer.
The compute crunch comes at a critical moment. Anthropic is gearing up for a November IPO that will put its infrastructure strategy under intense scrutiny from public market investors. Infrastructure spending has ballooned as the company scales Claude’s capabilities, but securing the massive GPU clusters needed to stay competitive is getting harder. Cloud providers who once eagerly courted AI labs with generous compute deals are now reassessing those relationships through a competitive lens.
The shift reflects a broader realignment in the AI industry. [Amazon](https://www.amazon.com), [Microsoft](https://www.microsoft.com), and [Google](https://www.google.com) have all poured billions into developing their own large language models while simultaneously selling cloud infrastructure to AI startups. That dual role worked when AI labs were customers buying services. But as those labs mature into platform companies offering enterprise AI products, they’re competing directly with the cloud giants’ own AI offerings.
For Anthropic, the timing is particularly awkward. The company just [inked a second $45 billion compute deal with infrastructure firm Nscale](https://www.techbuzz.ai/articles/anthropic-nscale-eye-second-45b-compute-deal), doubling down on a partnership designed to secure long-term access to GPUs. But even massive deals like that can’t fully insulate the company from broader market dynamics. Cloud providers control the underlying infrastructure, and they’re using that leverage to prioritize their own AI ambitions.
The competitive tensions are already showing up in customer defections. Developer platform [Replit](https://replit.com) [quietly dumped Anthropic for OpenAI months ago](https://www.techbuzz.ai/articles/replit-dumps-anthropic-for-openai-citing-cost-pressures), with CEO Amjad Masad citing cost pressures at TechCrunch Disrupt. That switch underscores how pricing and availability issues can push enterprise customers toward competitors, even when they prefer a particular model’s capabilities.
Anthropic’s IPO prospectus will need to address these infrastructure risks head-on. Public market investors will want clarity on compute costs, capacity commitments, and the company’s strategy for navigating relationships with cloud providers who are simultaneously partners and rivals. The [5% equity stake Anthropic committed to a public benefit fund](https://www.techbuzz.ai/articles/anthropic-commits-5-equity-stake-to-public-benefit-fund) signals the company is thinking about long-term governance, but infrastructure access is a more immediate concern.
The broader industry is watching closely. [OpenAI](https://www.openai.com) faces similar tensions with [Microsoft](https://www.microsoft.com), its primary cloud provider and largest investor. The recent [departure of OpenAI’s VP of Infrastructure](https://www.techbuzz.ai/articles/openai-vp-infrastructure-sachin-katti-exits-to-build-new-venture) to start a new venture hints at the complexity of managing these relationships at scale. Every major AI lab is navigating the same fundamental challenge: how to build on infrastructure controlled by companies that are becoming direct competitors.
Anthropic has tried to differentiate itself through safety commitments, including [opening its doors to independent AI auditors](https://www.techbuzz.ai/articles/anthropic-opens-doors-to-independent-ai-auditors) with unprecedented employee-level access. But safety credentials won’t solve compute scarcity if cloud providers decide to allocate GPU capacity to their own models first. The company needs ironclad contractual commitments, not just handshake partnerships.
The infrastructure arms race is intensifying across the board. [Nvidia](https://www.nvidia.com) recently [committed $250 million to UK AI safety infrastructure](https://www.techbuzz.ai/articles/nvidia-commits-250m-to-uk-ai-safety-infrastructure) over five years, highlighting how even chip makers are getting involved in shaping compute access. Every player in the AI stack is jockeying for position, and startups like Anthropic are caught in the middle.
As Anthropic heads toward its November IPO, the company will need to convince investors it can secure the infrastructure needed to compete long-term. That means either locking down multi-year commitments from cloud providers, building more of its own infrastructure, or finding alternative suppliers willing to bet on an independent AI ecosystem. The Nscale partnership is a step in that direction, but it may not be enough if the cloud giants keep tightening the screws.
Anthropic’s compute struggles reveal a fundamental tension in the AI industry: the companies that control infrastructure are also building competing products. As cloud providers shift from partners to rivals, AI labs face tough choices about how to secure the resources they need to survive. For Anthropic, those choices are coming into sharp focus just as it prepares to go public. Investors will be watching to see whether the company can navigate these messy partnerships or if compute scarcity becomes a structural disadvantage that limits its ability to compete with better-resourced rivals.
