Original Title: Everybody Wants to Rule the (AI) Stack.
Original Author: David Levy

Editor's Note: As the demand for AI compute power continues to soar, the competition among new cloud players is shifting from "who has more GPUs" to "who can control the full technology stack from hardware to software." As raw compute power rental becomes increasingly difficult to differentiate amidst price wars, a more crucial question emerges: what truly determines the value of AI infrastructure, the GPU itself, or the software layer that schedules, orchestrates, and optimizes GPU performance?

In this article, author David Levy, starting with Nscale's acquisition of Anyscale for around $1.65 billion, examines transactions such as Nebius acquiring Eigen AI, CoreWeave acquiring Weights & Biases, and IREN acquiring Mirantis. These acquisitions all point to the same trend: new cloud companies with GPUs, power, and data centers are collectively extending into the MLOps and AI orchestration layers.

The crux of this article's argument is that what new cloud companies are buying is not just a set of software or a batch of customers, but the ability to shift from "charging by the GPU hour" to "competing based on task outcomes." The scheduling system determines how many GPU hours a task will consume, impacting compute utilization, client costs, and platform profitability. By mastering both software and hardware layers, vendors can engage in synergistic optimization and increase customer migration costs through deeper system integration, thereby retaining the benefits of efficiency gains within the platform.

Simultaneously, integration is also happening in reverse. Lightning AI merges with GPU provider Voltage Park, while inference platforms like Fireworks, Modal, and Baseten are to various degrees building or controlling underlying compute power. Infrastructure companies are moving upward by acquiring software, and inference platforms are expanding downward into hardware. Both are converging from opposite directions towards the same endpoint: holding both compute assets and orchestration software simultaneously.

This implies that the next stage of new cloud competition is no longer just about GPU quantity, power agreements, and delivery speed, but about vying for control over the entire AI workload. The real barrier may lie in who can integrate chips, clusters, scheduling, and inference services into a more efficient, stickier system for customers. However, while the endpoints of these two paths may be the same, the cost structures are starkly different. New cloud companies can fill software gaps through acquisitions, whereas inference platforms seeking to build infrastructure from the bottom up must bear heavier capital expenditures. This also raises a critical question for the latter part—when every company wants to control the full tech stack, who can find cheap enough funding for this expansion?

The following is the original text:

Welcome to your own tech stack, where there is no turning back.

The AI tech stack is moving towards a "one company does it all" direction, with both ends of the industry chain striving to build such full-stack capabilities. New cloud companies are extending upwards through acquisitions, while inference platforms are expanding from the software layer to the infrastructure layer. However, only one side has been able to borrow enough money — but that's a story for later.

Last week, Nscale, a new player in the cloud field, acquired AI orchestration software company Anyscale. This is at least the fifth "AI infrastructure company acquires AI software company" deal in the past year or so.

In simple terms, it means those with a pager managing GPU clusters have started acquiring those hoodie-wearing techies who are still coding at 2 am — the software written by the latter determines how the clusters of the former operate.

Anyscale is the company behind the open-source AI workload orchestration framework Ray. Ray was developed by its founding team during their time at the University of California, Berkeley. According to Bloomberg, Nscale will spend approximately $1.65 billion to acquire Anyscale's commercial platform, engineering team, and clients such as Coinbase, Runway, and Bedrock Robotics.

Ray transitioned to the PyTorch Foundation in 2025 and will remain open source in the future. Anyscale will continue to operate as Nscale's AI orchestration business and retain the Anyscale brand — although after being acquired by Nscale, the "A" and "y" in the name seem somewhat redundant.

This is not an isolated transaction. In May of this year, Nebius acquired Eigen AI for $643 million, IREN acquired Mirantis, CoreWeave acquired Weights & Biases in 2025, and most recently, Qualcomm completed the acquisition of Modular.

These five deals point in the same direction: those who control the underlying hardware all want to further control the software layer that determines how this hardware operates.

By the way, I have previously discussed this trend in this article, and almost verbatim said: "Another AI infrastructure deal — but it's not just an infrastructure deal. To overlook this would be a mistake."

Of course, I talk so much on a regular basis that it's quite normal to be wrong once or twice.

A New Cloud Company Acquires the Ops and Orchestration Layer

The startup tech launched by Porch Capital utilizes the monitoring tool STAX, which tracked the tech stacks of approximately 12,000 venture-backed companies. Among them, the MLOps category data is meager: only 58 companies are using related tools, accounting for about 0.5% of the sample. But look at who's in this 0.5%: Ray and Weights & Biases contribute a total of 60 out of the 68 adoption records in this category. It should be noted that some companies may use multiple tools simultaneously.

As of July 30, these two companies have already found their match: CoreWeave has acquired Weights & Biases, while Nscale has bought Anyscale. In other words, in just 18 months, almost the entire commercially viable MLOps layer in the STAX sample has changed hands.

The transaction itself is news, but the real story is the transfer of control over the entire category.

On a side note, the reason why the MLOps data in STAX is so scarce is that application-layer companies hardly run MLOps tools themselves—at least not visibly from the outside.

This, in itself, is also a noteworthy finding: startups typically just invoke models rather than take responsibility for model deployment and operation.

What They Actually Acquired

Many people see bare GPU rental as a commoditized business: list a price, charge by the hour, and there's no moat.

Such a view is hasty. Power contracts, interconnect network topology, and compute delivery time are all real competitive barriers. Anyone who has truly tried to site a data center understands this.

However, from a broader perspective, this judgment is not entirely wrong—at least every new cloud company is acting on this logic.

The value at the software layer lies in shifting competition from "how much per hour of compute" to "how much to complete a task." Moreover, once customers complete system integration, they won't easily leave just because of another quote.

If a company has expertise in both infrastructure and software layers, it can engage in collaborative design of the two. This means that the benefits of efficiency improvements can be retained in-house rather than passed on to customers through lower bills.

That's why a company selling GPU hours is willing to pay $1.65 billion for a scheduling system.

Because it's the scheduling system that determines how many GPU hours a task exactly needs.

Now Flip the Script

An AI infrastructure company is acquiring software, while an AI software company is, in turn, building infrastructure.

Among them, Lightning AI and GPU infrastructure provider Voltage Park have completed a $2.5 billion merger transaction, with Lightning AI remaining as the surviving company.

Inference and model serving platforms such as Fireworks, Modal, and Baseten are now spread along a spectrum from "fully leased infrastructure" to "increasing proprietary assets."

Some companies, based on strategic principles, adhere to a light asset model, while others are starting to build their own infrastructure.

But their strategic endpoint is no different from Nscale, CoreWeave, and Nebius: they want to own the underlying hardware and control the software layer orchestrating that hardware.

Same Endpoint, Opposite Directions

Everyone wants to master the entire AI tech stack.

New cloud companies extend upward through software acquisitions, while inference platforms expand downward by building infrastructure. Both roads lead to the same endpoint, but one road is far more expensive than the other.

[Original Article Link]

Welcome to join the official BlockBeats community:

Telegram Subscription Group: https://t.me/theblockbeats

Telegram Discussion Group: https://t.me/BlockBeats_App

Official Twitter Account: https://twitter.com/BlockBeatsAsia