Anthropic Declines Investment at Valuation Over $800 Billion, While AI Chip Market Accelerates

Anthropic has demonstrated that in the new phase of the AI boom, money is no longer the main deficit: even when offered investment at valuations above $800 billion, the company can afford to decline. Against this backdrop, Meta is deepening its stake in proprietary chips alongside Broadcom, while ASML is raising its sales forecast because demand for AI infrastructure continues to accelerate the entire semiconductor market. The main story in this trio is Anthropic's reaction to investor interest.

The company rejected attempts to invest in it at valuations exceeding $800 billion. This fact in itself is more important than any single funding round: it shows how sharply the appetite of capital for developers of foundational AI models has grown. A few years ago, such companies were struggling for access to computational resources and funding, but now they can choose not only partners, but also the moment when they even need external money at all.

The rejection doesn't necessarily mean Anthropic won't raise capital later, but it clearly underscores its negotiating power. Investor interest in Anthropic is easy to explain. Companies creating large AI models are at the center of a market where several critical issues are decided at once: who controls the best models, who brings them to market fastest, who acquires corporate clients, and who can sustain the pace of spending on training and deployment.

In such a market, valuation is not just a reflection of current financial metrics, but a bet on future market share in a critical technological platform. If investors are willing to discuss valuations above $800 billion, this means the market believes not only in growing demand for generative AI, but also in the limited number of companies that can realistically claim leadership. In parallel, a second story is developing—the battle for computational independence.

Meta is expanding its multibillion-dollar partnership with Broadcom to design and release custom chips. For major platforms, this is no longer a side project, but part of core strategy. The deeper AI is embedded in advertising systems, search, recommendations, enterprise tools, and user services, the more painful the dependence on general-purpose accelerators and external suppliers becomes.

Proprietary chips offer a chance to better control the cost of computation, tailor architecture to specific workloads, and reduce supply shortage risks. This is especially important at a moment when demand for AI compute is growing faster than the industry can add supply. The third signal came from ASML—a key supplier of advanced equipment for semiconductor manufacturing.

The company raised its sales forecast for the full year, citing growth driven by AI demand. For the market, this is an important indicator, because ASML sits deep in the foundation of the entire value chain. When model developers and the largest platforms increase budgets, the next step becomes orders for new chips, factory capacity expansion, and growing demand for the most complex lithography machines.

When optimism strengthens at such a supplier, it looks not like local news about one company, but as confirmation that the investment cycle in AI is spreading across all infrastructure—from software to equipment. Together, these stories form a coherent picture. The center of gravity of the AI race is shifting from demonstrations of capability to control over the entire stack economy: capital, compute, supply chain, and equipment.

Anthropic shows that strong model developers can dictate terms to investors. Meta demonstrates that the next level of competition is proprietary hardware and more efficient cost structures for running AI services. ASML confirms that the boom is not limited to applications and chatbots, but is already materializing in orders for the semiconductor industry.

For the market, this means one simple thing: winners of the AI cycle will be determined not only by the quality of their models, but also by their ability to control the resources those models run on.