Pat Gelsinger, the former CEO of Intel, has backed a new startup that is trying to break Nvidia’s grip on the AI data center networking market. Delos Data announced a $100 million funding raise on September 15 to build networking chips and software, and Gelsinger is involved through his role as a general partner at Playground Global. The bet is simple: build a network that works across any accelerator, not just Nvidia’s.
The move puts AMD in an interesting spot. The startup is designed to work with any kind of compute hardware, which means it could eventually connect systems built around AMD’s silicon. AMD is not an investor in Delos, but the company’s design could end up benefiting the chipmaker anyway.
Gelsinger’s Stake in Playground Global
Gelsinger is now a general partner at Playground Global, the venture firm behind the investment. His involvement signals that the former Intel CEO sees value in a challenger to Nvidia’s dominant position in the AI infrastructure market.
Delos Data is building chips and software aimed at AI data centers. The startup’s pitch is straightforward: it is designing a network that connects to different kinds of compute hardware, rather than assuming every accelerator comes from Nvidia. That directness is the whole point.
The Problem Gelsinger Wants to Solve
Gelsinger described idle compute waiting for data as a major waste of both money and energy. The idea is that if a computer core sits idle while waiting for data to arrive, the entire system loses efficiency.
Delos Data’s approach tries to fix that by making the network itself more flexible. Instead of tying everything back to Nvidia’s proprietary gear, the startup is building a system that can work with any accelerator, regardless of who made it.
Nvidia’s Advantage and How Delos Bets Against It
Nvidia’s strength comes from selling a tightly integrated system. The company offers accelerators, networking, interconnects and software all built together. Customers buy the whole package because it works as one unit.
Delos bets that the market will become less uniform as inference processing expands. Inference workloads are expected to grow significantly, and Delos believes those workloads will spread across many different accelerators. Its open-networking approach is meant to reduce friction between vendors, so data moves freely no matter which hardware is doing the work.
The startup’s argument is that the market is moving toward more diversity in inference processing, and its open-networking approach is designed to handle that shift. The company is betting that inference workloads will fragment across many accelerators, and that open networking can reduce friction between vendors.
Inside the Hedge Fund Numbers
Insider Monkey tracked 285 hedge funds holding Nvidia in Q2, up from 275 in Q1. Arrowstreet Capital raised its position 10% to roughly 34.7 million shares. Nvidia’s holder count grew to 285 from 275.
AMD saw similar movement. Its holder count rose to 164 from 134. Marshall Wace held approximately 3.9 million shares after a modest increase.
About 41.7 million AMD shares were sold short as of August 31, equal to roughly 2.6% of float and 2.5 days to cover.
The Hedge Fund Picture
The hedge fund data shows that both Nvidia and AMD have active investor attention. Nvidia drew 285 insiders following its stock in Q2, up from 275 in Q1. Arrowstreet Capital increased its Nvidia stake by 10%, bringing its position to roughly 34.7 million shares.
AMD saw a smaller but still notable rise in interest. Its insider following climbed to 164 from 134. Marshall Wace added a modest amount to its AMD position, holding approximately 3.9 million shares.
Short interest in AMD stood at 41.7 million shares as of August 31, representing roughly 2.6% of the company’s float and 2.5 days to cover.
The Bet Against Nvidia’s Moat
Delos Data is trying to create a moat of its own. By building a network that works across any accelerator, the startup hopes to reduce the friction that currently favors Nvidia’s tightly integrated stack.
The comparison between the two approaches is direct:
| Approach | Focus |
|---|---|
| Nvidia | Tightly integrated system of accelerators, networking, interconnects and software |
| Delos Data | Open networking that works across different accelerators |
Why AMD Could Benefit
AMD is not an investor in Delos Data. But the startup’s design means its network could eventually connect systems built around AMD’s silicon. That is the indirect path: Delos builds a neutral layer, and AMD becomes a natural fit for customers who want options beyond Nvidia.
The bet is that inference processing will fragment across many accelerators, and that open networking can reduce friction between vendors. AMD could benefit from that fragmentation, even if it is not directly investing in the startup.
What Happens Next
The $100 million raise gives Delos Data the capital to build its chips and software. The question is whether the market agrees with its assessment of inference processing fragmentation.
Nvidia is not standing still. Its mature software and established networking products give it a strong base, and scale means customers already know what works.
The market will ultimately decide which approach wins, and the next few years of data center shipments will show which path customers actually choose.
Where the paper stands
The paper backs Delos Data’s approach of building a network that works across any accelerator, not just Nvidia’s, and is against letting big companies like Nvidia lock out competitors through dominance over AI data center networking. The danger is not the technology itself but the concentration of power around it, and Nvidia’s tightly integrated stack — accelerators, networking, interconnects and software all built together — is exactly the kind of moat the paper warns against. Licensing regimes and compliance costs only giants can afford are a moat, not a safeguard.
Delos Data’s open-networking approach targets that friction directly, and the paper supports challengers willing to do so. The paper opposes broad rules that hand the market to the incumbents, and it is against pauses, slowdowns and federal licensing of AI: they freeze today’s leaders in place and lock out whoever would have challenged them. When the biggest firms ask to be regulated, the paper asks who those rules would lock out.
Readers should watch how the market treats inference processing fragmentation, and whether Delos Data’s $100 million raise turns into actual adoption. The next few years of data center shipments will show which path customers actually choose.
Source material: “Pat Gelsinger Just Backed a $100 Million Challenge to Nvidia’s AI Networking Moat. AMD Could Benefit,” Yahoo Finance.
Get the Notebook.
The day's best stories and every fresh verdict, in plain English, in your inbox by seven. One email a day, no more.