Google in Talks to Rent AI Servers from CoreWeave

Here's What That Could Mean

Even the tech giants sometimes require assistance in the ever shifting field of artificial intelligence. Google is now negotiating with CoreWeave, a rapidly rising cloud computing firm that has developed somewhat of a dark horse in the field of AI infrastructure, to lease AI servers, sources say. This is a surprise decision that might reveal much about the present condition of cloud capacity, artificial intelligence workloads, and even the future of the people behind the explosion of AI.

Let us explore what is happening, why it is significant, and what it may imply for the field.

First of all, who (or what) is CoreWeave?

You're not alone if the word "CoreWeave" leaves you baffled. Not very well-known until recently, CoreWeave was. However, among AI circles, the business is slowly changing. Started in 2017 as a crypto mining business—yes, really—CoreWeave shifted into cloud infrastructure as high-performance computing (HPC) became in great request.

CoreWeave now runs a cloud platform made expressly for AI and machine learning applications. Consider great clusters of NVIDIA GPUs, low latency networking, and hardware designed to support all from training large language models to producing sophisticated 3D images.

Business is also thriving. Said to be using OpenAI among its current customers, the business has just raised more than $2 billion in debt financing to grow its GPU-based platform. If the rumors are correct, Google would now be on that ranking list.

Wait, Google must have more artificial intelligence servers?

This is likely the most eyebrow raising section.

Google is not really any technology company after all. This is the Google that more or less designed the contemporary artificial intelligence stack. The Transformer architecture (which underpins ChatGPT and every other LLM today), Google Brain, or DeepMind all came from Google Brain or DeepMind. Their private TPUs, some of the most sophisticated AI chips in existence, span the world, as do their data centers and large internal cloud system.

So why would they require artificial intelligence servers from someone else?

Some interesting ideas are going around.

1. The AI Explosion is breaking Infrastructure

Google is, as everyone else, struggling with an explosion in need for AI compute— and even their extensive network can't keep with it straightforwardly stated.

Particularly large language models (LLMs) are very hardware intensive to train, run, and scale. Simply deploying a model like Bard or Gemini among millions of users can consume incredible GPU processing. And that's before you take into account all the rear stage internal research, corporate cloud clients, and experimental initiatives.

CoreWeave is an expert in precisely this sort of infrastructure. Their cloud has been optimized for AI inference, which is probably of most interest to Google. Using CoreWeave's GPU clusters could be a versatile and rapid answer if Google needs to quickly expand Bard or Gemini services in reaction to a usage increase or to introduce a new function.

2. Sometimes the solution is not TPU chips

Designed specifically for training and running artificial intelligence models, Google has invested much time and money in the creation of their personalized Tensor Processing Units (TPUs). Although quite strong, they are not a onefitall fix.

Especially models developed using frameworks like PyTorch, some run better on NVIDIA graphics cards, which have become the industry standard. CoreWeave also has many NVIDIA chips, especially the H100s, which are presently the most sought after AI GPUs.

If Google is developing AI models or services specifically for NVIDIA hardware (maybe via partnerships or consumer requests), then spinning those up on CoreWeave's infrastructure could be more practical than trying to retrofit them to operate on TPUs.

3. The real client here might be Google Cloud

This may not just concern internal Google initiatives like Bard or Gemini; it could also relate to Google Cloud.

A big component of the business is Google Cloud, and artificial intelligence infrastructure is a major selling feature. If demand from customers is exceeding what Google can currently supply through its own data centers, working with CoreWeave might be a wise solution to plug the holes. It would let Google give customers more GPUbacked instances without having to wait for their own equipment expansions to be available.

This agreement is a means rather to stay competitive; it does not indicate weakness in that sense. Microsoft (Azure) and Amazon (AWS) are also rapidly expanding their AI infrastructure and none of them are reticent to collaborate with or acquire outside talent to get ahead.

4. Spreading out supply chains for AI compute

There is a larger trend here which deserves to be noted: AI compute is fast becoming a strategic resource. Access to high-performance chips is becoming a competitive and sometimes even international issue as more enterprises base their future on artificial intelligence.

Google could be hedging its bets by collaborating with CoreWeave—that is, making sure it has several sources of compute, not just its internal systems. Particularly in a world where chip supply chains can be disturbed by political tensions as well as worldwide shortages, such flexibility is priceless.

What this means for the industry

The possibility of the Google-CoreWeave agreement is quite obviously a sign that we are starting a new stage in the race for artificial intelligence infrastructure—one in which speed, flexibility, and alliances are every bit as important as raw size.

It also shows a growing ecosystem. Just some years ago, there were only a few players even considering providing the compute required to run and train advanced artificial intelligence models. Companies like CoreWeave are now not just in the game but clinching big transactions with OpenAI and perhaps Google.

This is a significant change that might enable still more competition and creativity in the sector. Particularly as artificial intelligence technologies become more varied and more integrated into regular products, smaller businesses with specialized infrastructure may be in great demand.

A new cloud arms race

This might easily be seen as one more step in a bigger arms race inside the cloud and AI sector. Vying for control of the AI infrastructure stack are Google, Microsoft, Tencent, Meta, and newer entrants CoreWeave, Lambda Labs, and even startups including Together.Ai

Continued sales of graphics processing units, tailored chips, and data center expansion are record-breaking. To secure computing resources, businesses are negotiating multiyear agreements worth billions. The person who has a great edge is one able to provide the ideal balance of performance, flexibility, and cost-effective.

In that race, working together might be more important that going it alone. Google's rental from CoreWeave shows they are being astute about where and how they grow, not abandoning their own infrastructure.

Next show for viewing

Though surely one to watch, we still have no idea whether the arrangement is done or under what conditions it is. Should it be greenlit, we could begin to see CoreWeave driven Google services in the wild—or maybe just significantly more Google Cloud GPU capacity.

It also begs the question: who other might be looking to outsource some of their AI workload? Meta? Apple? Tesla? More unexpected collaborations could emerge as demand rises.

This is quite a validation point for CoreWeave. It's no small accomplishment to transition from a crypto mining pivot to a Google infrastructure partner. It also shows more evidence that the AI explosion is far from over for the entire sector—that the competition for compute is quickly becoming heated only further confirms this.

Final thoughts

More than just a funny headline, Google negotiating with CoreWeave to lease AI servers is a picture of where the AI sector is presently: expanding quickly, morphing, and more and more relying on an extensive, dynamic infrastructure ecosystem.

It reminds us that even the top players in technology can't accomplish everything inhouse. Partnering up is sometimes advisable, especially when the stakes (and workloads) are this high.

Should this agreement be finalized, it will be a victory for both firms and a sneak peek into the probably future of the AI scene in years ago.

Stay alive. This one's still just starting.