Pat Gelsinger Just Backed a $100 Million Challenge to Nvidia’s AI Networking Moat. AMD Could Benefit

Habib Ur Rehman, September 23, 2026

The landscape of artificial intelligence infrastructure is poised for a significant shift as Delos Data, a burgeoning startup, has secured over $100 million in funding to tackle what its proponents, including former Intel CEO Pat Gelsinger, identify as the next critical bottleneck in AI development: data movement. The announcement, made on September 15, details Delos Data’s ambition to develop advanced networking chips and software specifically designed for the increasingly heterogeneous AI data center environment, a move that could profoundly impact the competitive dynamics between industry giants like NVIDIA Corporation (NASDAQ:NVDA) and Advanced Micro Devices, Inc. (NASDAQ:AMD).

The Emerging Bottleneck: Data Movement in AI

For years, the conversation around AI acceleration has predominantly focused on the raw computational power of graphics processing units (GPUs) and specialized AI accelerators. Companies like Nvidia have invested billions into developing increasingly powerful silicon, pushing the boundaries of parallel processing. However, as AI models grow exponentially in size and complexity, demanding vast datasets for training and inference, a new challenge has emerged from the shadows of raw compute: the efficient and rapid movement of data.

This challenge, often referred to as "data gravity," dictates that the performance of an AI system is not solely limited by how fast a chip can process instructions, but critically by how quickly data can be fed to that chip and moved between different computational units and memory tiers. Modern AI workloads, from large language models to complex scientific simulations, involve gigabytes, terabytes, and even petabytes of data that must traverse across networks, between GPUs, and in and out of memory. If the network infrastructure cannot keep pace with the accelerators, these powerful chips spend a significant portion of their time idle, waiting for data to arrive. This inefficiency translates directly into wasted computational cycles, increased energy consumption, and higher operational costs for data center operators.

Pat Gelsinger, a general partner at Playground Global, which participated in Delos Data’s funding round, succinctly articulated this problem. Gelsinger described idle compute, awaiting data, as a "major waste of both money and energy." His perspective, rooted in decades of experience at the forefront of semiconductor innovation and data center architecture, underscores the strategic importance of addressing this often-overlooked aspect of AI infrastructure. Gelsinger’s career has been marked by a focus on optimizing entire computing platforms, making his backing of Delos Data a strong signal that the industry’s focus is shifting beyond mere accelerator speed.

Delos Data’s Vision: Open and Heterogeneous AI Networking

Delos Data’s strategic pitch is centered on building a network infrastructure that is hardware-agnostic, capable of seamlessly connecting a diverse array of compute hardware within a single data center. This approach stands in stark contrast to the prevailing paradigm, largely established by Nvidia, where the optimal performance of AI workloads often relies on a tightly integrated stack of proprietary hardware and software. Reuters reported that Delos is designing its network to operate across different types of compute hardware, deliberately moving away from an assumption that every accelerator will be sourced from Nvidia.

The startup’s ambition to develop bespoke networking chips and software suggests a holistic solution that aims to optimize every layer of the data transfer process. This could involve innovations in network interface cards (NICs), switches, and routing protocols, all designed from the ground up to minimize latency and maximize bandwidth for AI-specific traffic patterns. By focusing on the interconnectivity layer, Delos Data aims to unlock the full potential of heterogeneous AI data centers, where specialized accelerators from various vendors can coexist and collaborate efficiently.

Nvidia’s Integrated Ecosystem: A Formidable Moat

Nvidia’s dominance in the AI sector extends far beyond the raw power of its GPUs. The company has meticulously built a comprehensive, tightly integrated ecosystem that spans hardware, interconnects, and software, creating a powerful "moat" around its market position. This ecosystem includes:

1. CUDA: Nvidia’s proprietary parallel computing platform and programming model, which has become the de facto standard for GPU programming in AI. Its widespread adoption means that developers and researchers often find it easier to work within the Nvidia environment.
2. InfiniBand: A high-bandwidth, low-latency interconnect technology acquired by Nvidia through Mellanox. InfiniBand is crucial for connecting thousands of GPUs in large-scale AI training clusters, providing the rapid communication necessary for distributed computing.
3. NVLink: A high-speed, point-to-point interconnect developed by Nvidia to allow direct communication between GPUs and between GPUs and CPUs, bypassing the PCIe bus for even lower latency and higher bandwidth within a single server or node.
4. DGX Systems: Nvidia’s fully integrated AI supercomputing systems, which combine GPUs, NVLink, InfiniBand, and a complete software stack (including CUDA and various AI frameworks) into ready-to-deploy units. These systems offer an "out-of-the-box" solution for AI development, simplifying deployment and ensuring optimal performance.

This integrated approach makes large AI clusters easier to deploy and manage, and it gives customers fewer compelling reasons to mix architectures from different vendors. The synergy between these components ensures that Nvidia’s GPUs can operate at peak efficiency, creating a powerful value proposition that has historically been difficult for competitors to challenge.

The Potential Ripple Effect: AMD’s Strategic Advantage

While Advanced Micro Devices, Inc. (NASDAQ:AMD) is not an investor in Delos Data, the startup’s success could indirectly benefit AMD significantly. AMD has been vigorously enhancing the performance of its Instinct series of AI accelerators, such as the MI250 and MI300X, and has been actively winning AI deployments. However, a persistent "friction point" in the market has been the difficulty of seamlessly integrating AMD hardware with other vendors’ ecosystems, particularly when Nvidia’s tightly integrated stack is the dominant standard.

Hardware-agnostic networking, as envisioned by Delos Data, could substantially lower this friction. If data centers can efficiently connect a mix of accelerators, regardless of vendor, then AMD’s Instinct GPUs would be able to compete on a more level playing field. The perceived complexity and interoperability challenges of mixing AMD hardware with other components often serve as a protective barrier for Nvidia. By abstracting the networking layer and making it truly open, Delos Data could enable data center operators to choose accelerators based purely on performance, cost-effectiveness, and power efficiency, rather than being constrained by ecosystem compatibility. This would open up new opportunities for AMD to secure larger AI deployments and expand its market share in the lucrative AI chip sector.

Chronology of a Strategic Shift:

• Early 2020s: Rapid growth in AI model sizes and complexity begins to highlight the limitations of traditional networking infrastructures in large-scale AI data centers. Discussions around "data bottlenecks" intensify within the tech community.
• Pat Gelsinger’s Strategic Moves: Following his return to Intel as CEO in 2021, Gelsinger emphasizes a focus on platform leadership and efficient data movement, consistent with his long-held views on integrated systems. His later involvement with Playground Global reinforces this strategic lens.
• Delos Data’s Genesis: While the exact founding date is not publicly detailed in the provided information, the company likely began operations prior to its major funding round, developing its core technology and strategic vision.
• September 15, 2026: Delos Data officially announces it has raised over $100 million in funding, with Playground Global, where Pat Gelsinger is a general partner, joining the round. This public declaration signals the company’s readiness to execute its vision.
• Ongoing: Nvidia continues to invest billions in R&D for its integrated AI platform, while AMD ramps up production and adoption of its Instinct accelerators, intensifying the competition for AI market share.

Challenges and the Incumbent’s Strength

The counterargument to Delos Data’s ambitious undertaking is formidable and highlights the immense challenge of disrupting an established market leader like Nvidia.

1. Nvidia’s Scale and Resources: Nvidia is a multi-billion dollar company that invests extensively in research and development across its entire AI stack, from silicon design to software ecosystems like CUDA, and its robust networking solutions. A startup, even with $100 million in funding, is attacking an incumbent that commands vastly greater financial and engineering resources.
2. Mature and Proven Ecosystem: Nvidia’s networking products, particularly InfiniBand, are mature, widely deployed, and have a proven track record in the most demanding AI supercomputing environments. Customers are familiar with Nvidia’s solutions and trust their reliability and performance. Delos Data must not only match this performance but also demonstrate superior value proposition to convince customers to switch.
3. Software Integration: Hardware is only one part of the equation. Nvidia’s software ecosystem, led by CUDA, provides seamless integration between its GPUs, interconnects, and AI frameworks. Delos Data’s networking solutions must be able to integrate effectively with existing software stacks, or provide compelling new software layers, without introducing new complexities for developers.
4. Commercial Deployment and Proof of Concept: Delos Data faces the critical task of proving its claimed performance improvements and cost efficiencies in real-world commercial deployments. Moving from lab benchmarks to large-scale, enterprise-grade data centers requires rigorous validation, extensive support, and a robust product roadmap.
5. Market Inertia: Enterprises often exhibit a degree of inertia, preferring to stick with established, reliable vendors rather than taking risks on new, unproven technologies, especially for mission-critical infrastructure like AI data centers. Overcoming this inertia will require a compelling combination of performance, cost savings, and ease of integration.

Despite these challenges, the substantial investment in Delos Data, coupled with the strategic backing of industry veterans like Pat Gelsinger, underscores a growing conviction that the current AI infrastructure paradigm is ripe for disruption. The demand for flexible, cost-effective, and energy-efficient AI solutions is escalating, creating an opening for innovators who can address the fundamental bottlenecks of data movement.

Broader Implications for the AI Industry

The success of companies like Delos Data could have profound implications for the broader AI industry:

• Increased Competition and Innovation: A more open and hardware-agnostic networking layer would foster greater competition among accelerator vendors. This could drive further innovation in chip design, pushing companies to differentiate on raw performance, efficiency, and specialized capabilities, rather than relying on ecosystem lock-in.
• Lower Costs for AI Development: By enabling more efficient use of computational resources and potentially reducing reliance on single-vendor solutions, Delos Data’s technology could lead to lower infrastructure costs for AI training and inference. This would democratize access to advanced AI capabilities, making them more accessible to a wider range of businesses and researchers.
• Greater Flexibility for Data Center Operators: Data center operators would gain unprecedented flexibility in designing and scaling their AI infrastructure. They could mix and match accelerators from different vendors to optimize for specific workloads, budgets, and power envelopes, rather than being limited to a uniform architecture.
• Advancement of Open Standards: Delos Data’s strategy aligns with a broader industry trend towards open standards and disaggregated architectures. If successful, it could accelerate the adoption of open networking protocols and architectures in AI, moving away from proprietary solutions.
• Energy Efficiency: By reducing idle compute time and optimizing data movement, Delos Data’s solutions could contribute significantly to improving the energy efficiency of AI data centers, a critical concern as AI workloads consume increasingly vast amounts of power.

Investor Sentiment and Market Dynamics

The financial community is closely watching these developments. Insider Monkey’s analysis of hedge fund activity in Q2 showed increased exposure to both incumbent leaders. A total of 285 hedge funds held positions in Nvidia, an increase from 275 in Q1. Notable moves included Arrowstreet Capital raising its position by 10% to approximately 34.7 million shares, indicating strong confidence in Nvidia’s continued growth trajectory. Similarly, AMD saw its holder count climb to 164 from 134, with Marshall Wace modestly increasing its stake to approximately 3.9 million shares. This suggests that while there is growing interest in potential disruptors, investors are simultaneously reinforcing their bets on the established players, recognizing their current market dominance and robust financial performance.

Despite the positive sentiment, AMD also recorded approximately 41.7 million shares sold short as of August 31, representing roughly 2.6% of its float and about 2.5 days to cover. This indicates that a segment of the market perceives potential downside risks or believes AMD’s current valuation might be stretched, possibly due to the intense competition in the AI accelerator space and the challenges of challenging Nvidia’s entrenched ecosystem.

Conclusion

The investment in Delos Data by Pat Gelsinger and Playground Global represents a significant strategic wager on the future of AI infrastructure. It posits that the next battleground for AI supremacy will not be solely about raw compute power, but about the efficiency and flexibility of data movement within complex, heterogeneous data centers. While Nvidia’s integrated ecosystem presents a formidable challenge, Delos Data’s vision of hardware-agnostic networking could unlock new levels of performance, efficiency, and competition, indirectly benefiting companies like AMD and ultimately shaping the trajectory of AI innovation for years to come. The coming years will reveal whether Delos Data can successfully navigate the complexities of market entry and technical execution to establish itself as a pivotal player in the evolving AI landscape.