Anthropic, Google and Broadcom: next generation gigawatts partnership for AI

Anthropic has announced a significant expansion in its partnership with Google Cloud and Broadcom, with the aim of ensuring gigawatt scale computing capacity for its future artificial intelligence models. The agreement provides for the expansion of the use of Google's Processing Units (TPUs), which are specialized chips designed to accelerate machine learning workloads. The initiative aims to meet the growing demand for computational resources in both training and inference of the company's AI systems.

Why is the gigawatt scale essential for the AI partnership?

The artificial border intelligence models, such as those developed by Anthropic, require huge volumes of parallel processing. Training a single version of a large language model can consume thousands of hours of GPU (or TPU) and energy equivalent to that of small cities. The reference to gigawatts — unit equivalent to billions of watts — indicates that the planned infrastructure will be comparable to that of large hyperscale data centers. Without this capacity, the advancement in the reasoning and security capacity of models can stagnate.

Google Cloud and TPUs: pillars of gigawatts partnership in AI

Google TPUs are specific application integrated circuits (ASICs) designed exclusively to accelerate machine learning frameworks such as TensorFlow and JAX. By expanding the use of these chips, Anthropic gains access to an optimized architecture for large distributed workloads with high energy efficiency. Google Cloud, in turn, strengthens its AI ecosystem by attracting one of the industry's most demanding customers. The partnership also includes the joint development of new network and storage capabilities to support the multi-gigawatt scale.

Broadcom and the infrastructure of the gigawatts partnership for AI

Broadcom is one of the leading suppliers of semiconductor solutions and network infrastructure for data centers. In collaboration with Anthropic and Google, the company is active in providing high-speed interconnection components and network chips that enable efficient communication between thousands of TPUs. Without this network layer, the collective performance of the cluster would be limited by latency bottlenecks. Broadcom also contributes to chip packaging expertise and cooling systems for high computational density.

Impact of gigawatt partnership on the AI ecosystem

Anthropic's move reinforces the trend that border AI companies need to ensure access to specialized hardware long before their models are launched. For Brazilian startups that develop applications based on large models, this can signal that the input cost will remain high, but also that providers like Google Cloud are investing in regional infrastructure. In addition, the dispute for computing capacity is driving innovations in alternative chips — such as those that [Cerebras Systems is leading to the stock exchange] (/noticias/ipo-da-cerebras-systems-chips-gigantes-para-ia-chegam-a-bolsa-de-valores).

The partnership involves a long-term commitment, with investments in new facilities and multiannual contracts. Anthropic did not reveal exact values, but the gigawatt scale indicates figures in the billions of dollars house over several years. The news was confirmed in the company's official press release, which can be read in full on the Anthropic website.

Frequently asked questions about the gigawatts AI partnership

Is Anthropic replacing GPUs with TPUs?

Not completely. The partnership expands the use of Google TPUs, but Anthropic also uses other architectures, such as NVIDIA GPUs, depending on the workload. The idea is to diversify computational resources to avoid excessive dependence on a single supplier.

What changes for developers who use the Anthropic API?

In the short term, there are no direct changes for API users. In the long run, greater computational capacity should allow Anthropic to offer faster models, with greater accuracy and less latency, in addition to supporting future versions of Claude with more parameters.

How can Brazil benefit from this partnership?

Indirectly, the expansion of Anthropic's global infrastructure can lead to increased availability of AI services in the region if Google Cloud decides to allocate part of this capacity to data centers in South America. In addition, the demand for professionals specialized in optimization of models for TPUs and high performance networks opens up opportunities for Brazilian engineers.