LLMs & Cloud Hyperscalers

1. The Hyperscalers

The cloud computing market is dominated by three major providers, often referred to as Cloud Hyperscalers:

• Amazon Web Services (AWS)

• Microsoft Azure

• Google Cloud Platform (GCP)

 

Together, these giants command an impressive total annual revenue nearing $200 billion. The majority of large enterprises opt for one of these three providers, primarily due to the trust and breadth of services they offer. Each of these platforms provides around 200 different services, ranging from virtual machines (VMs) to databases, API gateways, and virtual private cloud setups. Most enterprises have already configured their organizational structures, permissions, spending and credit agreements, virtual private clouds/VPNs, and security setups to align with these cloud providers. This integration makes the hyperscalers not just large, but also extremely ingrained in their clients' operations. It becomes prohibitively expensive for a large enterprise to switch to another provider, despite the availability of multi-cloud solutions, such as Infrastructure-as-a-Code.

​

2. Closed Source LLMs

The immense potential of Large Language Models (LLMs) and Generative AI has not gone unnoticed by the Hyperscalers. Each has invested in different ways to secure a foothold in this emerging field. Amazon and Microsoft have chosen to invest in leading startups – Anthropic and OpenAI respectively – while Google has developed its own LLM, continuing its tradition of in-house innovation. The current partnerships stand as follows:

• AWS and Anthropic (Claude)

• Azure and OpenAI (ChatGPT)

• GCP and Palm (soon to be Gemini)

​

These closed-source LLMs are among the best in the market. Each hyperscaler has strategically positioned themselves to ensure they offer a top-tier closed-source LLM, aiming to retain and attract customers.

​

3. Open Source vs. Closed Source

The LLM community is currently divided over a crucial question: will open-source or closed-source models prevail? Llama and Falcon are exemplary open-source LLMs, and more are being released regularly. Although there is a noticeable quality gap between these and the likes of Claude, ChatGPT, and Palm, this gap is narrowing steadily. It's reasonable to be optimistic about the future of open-source LLMs. However, a common misconception needs addressing: open-source does not necessarily mean free. Operating these models still requires significant GPU resources, which is where the strategic plans of the Hyperscalers come into play. Simply, Hyperscalers don't care which one wins as long as the customers continue to run their LLMs on their clouds.

​

4. Hyperscalers’ Strategy

Consider a large enterprise using AWS. This enterprise has its entire organization's infrastructure integrated with AWS – including permissions, budgets, alerts, logging, and monitoring. They have committed to multi-year spending agreements with AWS and have set up their virtual private cloud securely within the AWS ecosystem. Now, suppose this enterprise wants to incorporate LLMs into their system. AWS simplifies this by offering:​

• Easy integration of Anthropic Claude with just one click, as though adding another VM in the private cloud.

• Similar one-click integration for open-source models, seamlessly adding another layer to the cloud infrastructure.

​

This convenience is highly attractive. Consequently, the prospect of leaving AWS for alternatives like ChatGPT or Gemini starts to seem prohibitively expensive and complex. While this example is specific to AWS, same logic applies to GCP and Azure as well.

​

5. The Future of Hyperscalers and LLMs

The future looks exceptionally bright and active for AWS, Azure, and GCP. Their continued growth appears almost assured, driven by their capacity to offer high-quality, integrated AI solutions like ChatGPT, Claude, and Gemini. For many large enterprises, the decision to stick with their existing cloud provider is straightforward. They benefit from leveraging their established infrastructure and avoiding the complexities of switching to new systems. However, this dynamic is different for midsize enterprises and startups. These smaller and more agile organizations have the freedom to choose any cloud provider without facing significant switching costs. This flexibility allows them to select the most suitable cloud environment for their specific needs, potentially even experimenting with multiple providers. This diversity in choice and adaptability is set to foster a dynamic LLM infrastructure ecosystem.