Two Thousand Dollars Per Socket

On August 24, 2021, NVIDIA published a price for artificial intelligence. Not for a chip - for software. NVIDIA AI Enterprise, a suite optimized for VMware vSphere, went on general release at $2,000 per CPU socket for one year with support included, or $3,595 for a perpetual license with support bought separately.

A company that sold silicon was now selling a seat license.

Manuvir Das, head of enterprise computing at NVIDIA, presented the day as a break with what came before. "The first wave of AI has been powered by specialized infrastructure that focused adoption on industry pioneers," he said.

The second wave was supposed to begin that morning. It took two more years.

Dozens, Not Thousands

The same release put a number on the customer base, and the number was modest. Dozens of automotive, education, finance, healthcare, manufacturing and technology companies worldwide were named as early adopters. Dozens - for a product built to bring AI to the mainstream data center, running on servers enterprises already owned, under a hypervisor they already ran.

That is the honest shape of the position Das held: a software business inside a hardware company, priced per socket, aimed at a market that had not yet decided it needed one. The pitch was not that AI worked. It was that AI could be bought the way enterprises buy everything else - as a line item, per socket, per year, with a support contract attached.

For most of two years, most buyers did not.

Then The Race Started

By August 8, 2023 the sentence had changed shape entirely. Announcing NVIDIA AI Workbench, Das described a market that had stopped deliberating: "Enterprises around the world are racing to find the right infrastructure and build generative AI models and applications." The constraint was no longer persuasion. It was supply.

He had spent the intervening period arguing the case in infrastructure terms rather than model terms. In August 2022, appearing alongside Dell Technologies, he put it as a question of how buildings full of servers are laid out: "AI and zero-trust security are powerful forces driving the world's enterprises to rearchitect their data centers".

The AI Workbench pitch was continuous with the 2021 one, and the continuity is the point: make the thing ordinary. The tool, he said, "provides a simplified path for cross-organizational teams to create the AI-based applications that are increasingly becoming essential in modern business". Not a breakthrough. A path. The word he reached for in 2023 was the same word the 2021 price list implied - that AI was about to become something a finance department could requisition.

Same argument, twice, two years apart. Only the weather had changed.

Before The Chips, The Cloud

Das did not arrive at NVIDIA from the model side of the industry. He arrived from storage.

At Dell EMC he ran the unstructured data storage business - the scale-out Isilon NAS platform and the scale-out ECS Object platform, the systems enterprises use when the files stop fitting anywhere sensible. Before that he spent fourteen years at Microsoft, where he was an original engineering leader behind the development of Microsoft Azure. He holds master's and doctoral degrees in computer science from the University of Wisconsin and a bachelor's degree in computer science and engineering from the Indian Institute of Technology Bombay.

Storage, then cloud, then enterprise AI. Three times he has worked one layer beneath whatever the industry was announcing that year - the plumbing rather than the fixture.

It is a useful vantage point and a thankless one. Nobody writes about the object store. But the sequence explains why his 2021 argument took the shape it did. Somebody who spent fourteen years shipping Azure and then ran Isilon does not pitch AI as a revelation. He pitches it as capacity: something that has to be racked, licensed, supported and paid for on a schedule.

A Seat Without Operations

Then, in April 2025, he stopped running things.

Das became an operating partner in the digital infrastructure group at Stonepeak Partners, the private capital firm. It is an advisory seat at an investor rather than a business with a profit-and-loss statement of its own - a deliberate move from building infrastructure to funding it, made at the precise moment the infrastructure trade became the loudest in the market.

The record of what he has done there is not public. Stonepeak does not publish deal-level attribution, and no primary source reviewed here places him on a named transaction. It is the one stretch of a well-documented career where the documentation stops.

What The Board Bought

On May 26, 2026, Western Digital appointed him to its board of directors. The company announced it two days later, noting that the board would then comprise nine directors, eight of them external and independent.

Marty Cole, WD's chairman, framed the appointment around operationalizing AI at scale and around storage - the two halves of the resume, joined. Das's own statement went to where the demand originates: AI, he said, is driving extraordinary growth in data creation and infrastructure investment, and "WD's storage infrastructure is foundational to enabling AI at scale".

It is a claim with an obvious interest behind it, and also an arithmetic one. Every model trained and every inference logged writes something down, and the writing down is not free. The demand he spent four years trying to manufacture at NVIDIA now arrives at Western Digital as a physical requirement, measured in exabytes rather than sockets.

There is a symmetry in that which the press release did not bother to draw. In 2021 he was selling enterprises the idea that they would need AI. In 2026 he sits on the board of a company whose problem is that they do.

On August 14, 2026, he signed Western Digital's annual report on Form 10-K as a director, one name among eleven on the signature page.

Five years after he priced enterprise AI at $2,000 a socket and found dozens of takers, he is back at the layer below - no longer selling the software that runs on the infrastructure, but sitting on the board of the company that makes the place it all lands.