Skip to main content

The Rise of NVIDIA, Part 13: Buying the Plumbing

Buying the Plumbing

🎮 The Rise of NVIDIA — a 20-part series. See all parts »  |  « Part 12: Ray Tracing

By 2019, Jensen Huang had a problem that most CEOs would kill for: NVIDIA's chips were winning. GPUs had become the beating heart of AI training, and the world's fastest supercomputers — Summit and Sierra, both built for the U.S. Department of Energy — ran on NVIDIA silicon. But Huang had noticed something that would reshape the company. The bottleneck in a modern datacenter was no longer the chip. It was the wiring between the chips.

When you lash together tens of thousands of GPUs to train a single model, the machine is only as fast as the network stitching them together. Data has to fly between nodes at staggering speed, and every microsecond of delay is multiplied across the whole cluster. NVIDIA made the muscle. It did not, yet, make the nervous system. So on March 11, 2019, it announced it would buy the company that did.

The company that made the wires smart

Mellanox Technologies was founded in May 1999 in the Israeli town of Yokneam Illit by Eyal Waldman and a group of former Intel and Galileo engineers. For two decades it had been the quiet aristocrat of high-performance networking, pioneering InfiniBand — an interconnect built not for the open internet but for the inside of a supercomputer, where latency is the enemy and raw throughput is king. By the time NVIDIA came calling, Mellanox's InfiniBand and high-speed Ethernet gear was threaded through more than half of the world's fastest machines and inside the datacenters of every major cloud provider.

It was, in other words, the plumbing. Unglamorous, invisible to the end user, and absolutely essential. NVIDIA and Mellanox had already collaborated for years — their combined technology powered over 250 of the TOP500 supercomputers, including the Summit and Sierra systems. Buying Mellanox was less a leap into the unknown than a formalization of a marriage that already existed.

A bidding war for the nervous system

NVIDIA was not the only suitor. Mellanox had put itself in play, and reporting at the time placed heavyweights including Intel and Xilinx among the interested parties. Everyone who understood datacenters understood what Mellanox was worth. NVIDIA won by paying up: $125 per share, all cash, for a total enterprise value of roughly $6.9 billion — at the time the largest acquisition in NVIDIA's history.

Huang framed the logic in the language that would define the next five years of the company. “The emergence of AI and data science, as well as billions of simultaneous computer users, is fueling skyrocketing demand on the world's datacenters,” he said. “Addressing this demand will require holistic architectures that connect vast numbers of fast computing nodes over intelligent networking fabrics to form a giant datacenter-scale compute engine.”

That last phrase is the whole thesis. Huang had stopped thinking of the computer as a box on a desk, or even a server in a rack. In his mind, the unit of computing had become the entire datacenter — and to sell a datacenter, you needed to own the compute, the storage path, and the network. Mellanox founder and CEO Eyal Waldman, who would stay on through the integration, called the deal “a natural extension of our longstanding partnership.”

Clearing the gauntlet, and the payoff

A deal this size, spanning American chips and Israeli networking, does not close on a handshake. It had to survive antitrust review on three continents. The most nervous wait was Beijing, where regulators held real leverage as U.S.–China trade tensions ran hot. Approval finally came, and the acquisition closed on April 27, 2020, with sign-off from the European Union, the United States, and China. Mellanox was folded into NVIDIA's networking division, and within a couple of years the storied brand name quietly disappeared from new products.

What NVIDIA got for its $6.9 billion turned out to be far more than InfiniBand switches. Mellanox brought the SmartNIC technology that NVIDIA would evolve into the BlueField DPU — the “data processing unit” Huang would soon pitch as the third pillar of the datacenter, alongside the CPU and the GPU. It brought the engineering culture that lets you treat thousands of GPUs as one coherent machine, the idea that would later crystallize as NVIDIA's rack-scale systems where the whole rack behaves like a single enormous accelerator.

The timing looks almost prophetic in hindsight. Just as the acquisition closed, the world's appetite for AI compute began its vertical climb, and NVIDIA's data-center business swelled into the largest and most profitable part of the company — eventually dwarfing the gaming empire that had built it. The plumbing NVIDIA bought in 2019 became the thing it was really selling.

Huang had made a quiet, expensive bet that the future of computing wasn't a chip but a fabric. He was right. Next time: the pandemic hits, crypto miners swarm the GPU market, and NVIDIA's supply chain gets stress-tested like never before.


🔗 Explore more from Syncster

Comments

Popular posts from this blog

Cursor AI Review: Is the AI Code Editor Worth It?

I've been using Cursor as my main code editor for a while now, and enough people have asked whether it's worth switching to that a proper review felt overdue. Short version: for me, yes — but with caveats. What is Cursor? Cursor is an AI-first code editor built as a fork of VS Code. That means every extension, theme, and keybinding you already use in VS Code works here, but with AI woven directly into the editing experience instead of bolted on as a plugin. It's made by Anysphere and can run models from OpenAI and Anthropic under the hood. What I like Tab completion is uncanny. Cursor predicts your next edit — not just the rest of the line, but the next change across the file. Once you get used to hitting Tab, going back to a plain editor feels slow. The Composer / Agent mode. You describe a change in plain language and it edits multiple files at once, showing you a diff to accept or reject. For refactors and boilerplate, this saves real time. It unde...

MacBook Pro M5 vs M5 Pro: Which One Should You Actually Buy?

Apple's latest 14-inch MacBook Pro comes in two very different flavors: the base M5 and the step-up M5 Pro . On paper they look similar — same gorgeous Liquid Retina XDR display, same design — but under the hood the gap is bigger than the names suggest. Here's a clear, no-hype breakdown, with concrete use cases so you can match the chip to your work. Quick spec comparison Spec M5 M5 Pro CPU 10-core (4 performance + 6 efficiency) Up to 18-core (6 performance + 12 efficiency) GPU 10-core Up to 20-core Neural Engine 16-core 16-core Memory bandwidth 153 GB/s 307 GB/s (roughly double) Unified memory 16 / 24 / 32 GB 24 / 48 / 64 GB Max storage Up to 4 TB SSD Up to 8 TB SSD Battery (video playback) Up to 24 hours Up to 22 hours Media engines Single encode/ProRes engine More encode/ProRes engines (higher configs) What actually changes between them More cores — the M5 Pro nearly doubles CPU cores and adds GPU cores, so sustained, multi-threaded work finishe...

How I used Google Sheets and Apps Script

Google Sheet is one of the most powerful spreadsheet application that exists online, rivaling with Microsoft's Excel. One of the main strengths is its strong support for collaboration with other users, much easier and popular than collaboration tools with Microsoft Office. Aside from plain spreadsheet, it also supports extensions such as macro. If you are familiar with macros on other office tools, they work almost the same. However, the most extension I use and tinker with is the Apps Scipt . Apps Script Extension One of the challenges I faced recently is how do I track or monitor reports in our department if they are submitted on time or worst, forgotten due to lack of better monitoring tools. So I thought if there can be simple applications that can be deployed or use by a more general user to allow reminding periodically what reports are approaching due dates or those that are past dues. Then I looked for a way, instead of creating a full blown app from scratc...