Anthropic's Cloud Partners Turn Rivals Ahead of IPO
Anthropic is facing problems with the cloud companies it depends on. The AI startup that makes Claude is struggling to get enough computing power from its partners. According to a new report, these partners increasingly see Anthropic as a direct competitor. The timing is difficult: Anthropic is preparing for a planned IPO in November, its infrastructure costs are rising quickly, and major customers are leaving. What once looked like a close partnership between AI companies and cloud providers is becoming much more complicated.
Anthropic is learning that working with cloud providers can come with conditions. The company, which focuses on AI safety, is reportedly having trouble securing enough computing resources to train and run Claude, its main AI assistant. The reason is that these same cloud providers are developing their own competing AI models. They increasingly see Anthropic as a rival instead of just a customer.
The shortage of computing power comes at a critical time. Anthropic is preparing for an IPO in November, when investors in public markets will closely examine its infrastructure plans. The company is spending much more on infrastructure as it improves Claude, but it is becoming harder to secure the large groups of GPUs needed to compete. Cloud providers once worked hard to attract AI companies by offering generous computing deals. Now they are taking another look at these partnerships because of the competition.
This change reflects a wider shift across the AI industry. Amazon, Microsoft, and Google have all spent billions of dollars developing their own large language models while also selling cloud computing services to AI startups. This arrangement worked when AI companies were simply customers buying services. But as these companies grow into platforms that sell AI products to businesses, they are competing directly with the AI products offered by the cloud giants themselves.
For Anthropic, the timing is especially difficult. The company has just signed a second $45 billion computing deal with infrastructure company Nscale, strengthening a partnership meant to provide long-term access to GPUs. But even a deal this large cannot fully protect the company from wider market pressures. Cloud providers control the basic infrastructure, and they are using that control to give priority to their own AI plans.
The competition is already affecting which companies choose to use Anthropic's products. Developer platform Replit quietly stopped using Anthropic and switched to OpenAI months ago. Its CEO, Amjad Masad, pointed to cost pressures when speaking at TechCrunch Disrupt. The switch shows how high prices and limited access can push business customers toward competitors, even when they prefer what a particular AI model can do.
Anthropic's IPO prospectus will need to explain these infrastructure risks clearly. Investors in public markets will want to understand computing costs, guaranteed access to computing capacity, and the company's plans for dealing with cloud providers that are both partners and competitors. Anthropic's commitment to give a 5% equity stake to a public benefit fund shows that the company is thinking about long-term governance. But getting reliable access to computing power is a more immediate problem.
The wider AI industry is watching closely. OpenAI faces similar problems with Microsoft, its main cloud provider and largest investor. The recent departure of OpenAI's vice president of infrastructure to start a new business points to how complicated these relationships can become as companies grow. Every major AI company faces the same basic challenge: how to build its products using infrastructure owned by companies that are becoming direct competitors.
Anthropic has tried to stand out through its commitment to AI safety. This includes allowing independent AI auditors to examine its work, with an unusually high level of access that extends to individual employees. But a strong record on safety will not solve a shortage of computing power if cloud providers choose to give their GPUs to their own AI models first. Anthropic needs firm promises written into contracts, not just informal partnerships.
The race to build AI infrastructure is getting more intense across the industry. Nvidia recently pledged $250 million over five years for AI safety infrastructure in the UK. This shows that even chip makers are becoming involved in shaping access to computing resources. Every part of the AI industry is trying to strengthen its position, while startups like Anthropic are caught in the middle.
As Anthropic prepares for its November IPO, it will need to convince investors that it can secure the infrastructure required to compete over the long term. This could mean making multi-year agreements with cloud providers, building more of its own infrastructure, or finding other suppliers willing to support an independent AI industry. The Nscale partnership is a step in that direction, but it may not be enough if the major cloud providers continue to make access more difficult.
Anthropic's problems getting computing power reveal a basic conflict in the AI industry: the companies that control the infrastructure are also building competing products. As cloud providers become rivals as well as partners, AI companies face difficult decisions about how to secure the resources they need to survive. For Anthropic, these decisions are becoming especially urgent as it prepares to go public. Investors will be watching to see whether the company can manage these complicated partnerships or whether limited computing power becomes a lasting disadvantage that makes it harder to compete with rivals that have more resources.