Anthropic Plans to Manufacture Its Own Chips, Altering the Industry Landscape
There's a signal that rarely fails in analyzing corporate structure: when a company scaling at breakneck speed starts talking about making what it previously bought, it’s not announcing a victory. It's admitting a vulnerability. On April 9, 2026, Reuters reported that Anthropic is evaluating the design of its own artificial intelligence chips. Currently, conversations are in the preliminary phase, there isn't a dedicated team for this purpose, and the company hasn’t ruled out continuing to rely solely on external suppliers. But the mere fact that such discussions exist at the managerial level says a lot about the structural fragility underlying growth that, on paper, seems impressive.
Contextual numbers matter: Anthropic’s recurring annual revenue surpassed $30 billion, approximately tripling the $9 billion recorded at the end of 2025. An acceleration of this magnitude in such a short time is not just a metric of success. It also signals that computational demand is growing proportionally, or even faster, than revenues. And therein lies the problem.
The Cost of Depending on Those Who Sell You Gasoline
Anthropic currently operates with a multi-supplier strategy: chips from Nvidia, computational capacity from Amazon and Google. The company recently formalized a long-term agreement with Google and Broadcom to secure 3.5 gigawatts of computational capacity starting in 2027, including access to Google’s Tensor Processing Units (TPUs). From the outside, this seems like a diversified and solid position. From the inside, it’s a growing fixed cost structure anchored to decisions made by third parties.
The reliance on Nvidia as the dominant supplier in the AI processor market isn’t new, but its consequences become more acute as a company scales. When you’re small, you pay the market price and move on. When your language model generates tens of billions in revenue and each query requires intensive computation, the market price becomes the factor determining whether your operating margins make sense. No long-term profitability analysis can survive structural dependence on a resource you don’t control and one that is chronically scarce.
The semiconductor shortage isn’t temporary. The global supply chain for advanced processors has been under sustained pressure since 2021, and specific demand for AI workload has turned that pressure into a permanent strategic constraint for any company competing in this space. Anthropic isn’t considering manufacturing chips because it’s a good idea in the abstract. It’s considering it because the alternative—continuing to pay whatever manufacturers charge—has a financial viability ceiling that’s approaching faster than growth headlines suggest.
The Trap of Scaling Without Controlling Critical Inputs
The pattern describing Anthropic's situation has a direct parallel in managing asset portfolios with high risk concentration in a single supplier. Imagine a hedge fund generating consistent returns but whose strategy relies on accessing market data controlled by only one vendor. As long as that provider doesn’t raise prices or restrict access, the model works. However, the fund’s valuation is inflated because the market isn’t adequately discounting that concentration risk. Anthropic faces the same problem, only in silicon.
What makes this discussion particularly relevant from a structural risk angle is that Anthropic hasn’t decided anything yet. And that indecision, in itself, is an operational signal. Companies with robust financial architectures and controlled supply chains don’t have embryonic conversations when they’re already exceeding $30 billion in ARR. They have them years earlier, when the cost of exploration is low and the risk of making a mistake doesn't threaten operational continuity. The fact that this discussion is happening now, with that level of revenue and without a formed team, suggests that growth has outpaced infrastructure planning.
Meta and OpenAI have long been advancing similar initiatives for developing their own chips. They aren’t doing this for technological whims. They do it because at a certain scale, every dollar paid to an external supplier for the most critical input in your operation is a dollar that doesn’t translate into your competitive advantage. Vertical integration in semiconductors isn’t a bet on the future; it’s a tardy response to a vulnerability that accelerated growth has made urgent.
When Controlled Betting Becomes Structural Necessity
Designing proprietary chips isn’t a low-cost exploration with positive asymmetry. It’s a capital-intensive bet, with development horizons measured in years, requiring the hiring of highly specialized talent or acquiring already-formed teams, producing commercially viable results only if the internal usage scale justifies the investment. Google began developing its TPUs in 2016 and took several cycles to translate them into tangible operational advantages. Amazon built Trainium and Inferentia in similar timeframes. If Anthropic decides to move forward, it will be buying time towards a more controlled cost structure, but paying for that time with a level of organizational complexity it currently lacks.
The concrete financial question is whether the $30 billion ARR reflects margins that support such investment or if part of that growth is being subsidized by the capital of its investors, including Alphabet and Amazon, who have their own interests in how Anthropic manages its computing infrastructure. The fact that Anthropic’s primary investors are also its main computing suppliers creates a governance dynamic that warrants more attention than it receives in standard news coverage. This isn’t a conflict of interest in ethical terms; it’s a structural constraint: any decision made by Anthropic regarding chips directly affects the profit and loss accounts of those that injected capital.
If Anthropic moves towards designing its own processors, it won’t be to stop purchasing chips from its current suppliers in the short term. It will be to build, years in advance, a base of variable costs that doesn’t solely depend on what third parties choose to charge for the input that powers its product. This is the only move that makes economic sense at the scale this business is reaching. What the preliminary discussions reveal is that the window for making that transition in an orderly and non-urgent manner is closing.
