Cisco adds Supermicro compute as AI Factory targets rack-scale Rubin era

Cisco Systems, Inc. (NASDAQ: CSCO) has expanded its Secure AI Factory with NVIDIA through a new partnership with Super Micro Computer, Inc., adding rack-scale compute systems to an architecture that already combines Cisco networking, security, observability and NVIDIA accelerated computing. Beginning in October 2026, Cisco plans to offer Supermicro liquid-cooled and air-cooled systems as part of the portfolio, including infrastructure capable of supporting NVIDIA Vera Rubin NVL72 and NVIDIA HGX Rubin NVL8 platforms. The expanded architecture targets enterprises, neocloud operators and sovereign-cloud customers that want high-density AI infrastructure delivered through a more integrated deployment model rather than assembling networking, compute, cooling and software independently. The strategic move arrives after Cisco recorded $9.3 billion of hyperscaler AI infrastructure orders during fiscal 2026, but it also deepens Cisco’s exposure to a hardware category where rapid revenue growth is already creating pressure on product margins.

The August 25 announcement represents a significant broadening of Cisco’s role in the artificial intelligence stack. Cisco historically supplied the networking equipment connecting servers, but the addition of Supermicro systems means customers can now buy a Cisco-led architecture spanning GPU compute, front-end and back-end networking, liquid cooling, security, operational tooling and infrastructure certification. Cisco said the expanded platform will comply with NVIDIA Cloud Partner requirements for neocloud and sovereign-cloud deployments, while its Enterprise Reference Architectures are also being extended to the latest NVIDIA infrastructure. Availability beginning in October gives the partnership a near-term commercialization window rather than leaving it as a future architecture concept.

Why is Cisco moving deeper into rack-scale AI compute instead of remaining a networking supplier?

Artificial intelligence infrastructure is increasingly sold as a system rather than a collection of independent components. An AI cluster built around dozens or hundreds of GPUs needs servers, Ethernet or InfiniBand-class networking, optical connections, cooling, power delivery, security, orchestration and observability to work together at extremely high utilization. A networking vendor can capture substantial value from that environment, but a vendor controlling more of the validated architecture can influence a much larger customer decision.

Cisco’s partnership with Supermicro is therefore less about becoming a server manufacturer and more about controlling the integration layer. Supermicro supplies dense GPU systems and liquid-cooling expertise, NVIDIA supplies accelerated computing platforms and Cisco wraps those components with its networking, management, security and services portfolio. Cisco said customers will be able to deploy rack-to-fabric liquid cooling using Cisco liquid-cooled AI networking systems alongside Supermicro liquid-cooled servers, supporting workloads ranging from trillion-parameter model training to high-throughput inference.

The model also reduces procurement complexity for neocloud operators. Emerging AI cloud providers often need to deploy new GPU capacity rapidly but may not have the engineering depth of established hyperscalers to validate every combination of server, switch, cooling system and software component independently. A pre-integrated architecture can shorten commissioning time if it performs as advertised, particularly when access to expensive GPUs and memory makes deployment delays costly.

Cisco is adding a services layer around that proposition through Cisco Validated Infrastructure Services, which is aligned with NVIDIA Infrastructure Services. The company is also building a dedicated large-scale AI laboratory to develop testing software and certification tools, suggesting that Cisco wants deployment assurance itself to become part of the commercial value rather than simply shipping hardware.

How does the expanded Cisco AI Factory differ from a conventional GPU server deployment?

The most important difference is the networking architecture surrounding the compute. Cisco said its NVIDIA Cloud Partner compliant offering will use Cisco Silicon One-based switches for front-end networking and Cisco switches based on NVIDIA Spectrum-X for the back end, with both environments unified through Cisco Nexus One. Cisco describes itself as the only NVIDIA technology partner using its own networking switches and network operating system inside an NCP-compliant solution.

Back-end networking has become a major AI infrastructure battleground because GPUs must exchange enormous quantities of information during distributed training. If the network introduces congestion or latency, expensive accelerators can sit idle waiting for data, reducing the economic output of hardware that may cost millions of dollars per rack. Cisco is therefore attempting to position networking performance as a direct component of AI return on invested capital rather than a supporting IT function.

The architecture also combines AI workloads with conventional enterprise infrastructure. Cisco said customers can manage dense AI clusters alongside non-AI applications instead of operating completely separate technology environments, while NVIDIA AI Enterprise and Cisco Cloud Control provide software for operations and observability. The ability to correlate compute job health with network interface cards, optics, compute and network performance could become especially valuable as enterprises move beyond pilot projects and begin troubleshooting large production AI systems.

Security is another differentiator Cisco is emphasizing. The Secure AI Factory concept is designed to apply controls from infrastructure through AI agents, reflecting the fact that sovereign and enterprise customers may need stricter control over models and data than public cloud deployments provide. That can create an opening for on-premises and private AI infrastructure even when hyperscale cloud remains economically attractive for other workloads.

Why are sovereign clouds and neocloud providers becoming strategically important to Cisco?

The artificial intelligence cloud market is no longer limited to Amazon Web Services, Microsoft Azure and Google Cloud. Neocloud companies specializing in accelerated computing are building GPU clusters for customers that want capacity without competing directly against hyperscalers for scarce hardware, while governments are investing in sovereign AI infrastructure where sensitive data and models remain under domestic control.

These customers can be particularly attractive to Cisco because many need complete infrastructure architectures rather than individual components. A hyperscaler may design its own networking or servers, but a newer cloud operator can benefit from validated reference designs and global support. Cisco’s Supermicro partnership expands the amount of that architecture it can provide without having to vertically integrate server manufacturing itself.

Sovereign AI adds a policy dimension. Governments increasingly view domestic computing capacity as strategic infrastructure, particularly where national data, defence workloads, regulated industries or public-sector models are involved. Cisco’s ability to combine networking, compute, security and management while maintaining customer control over data makes the architecture relevant to those buyers even when raw compute economics remain more attractive in public cloud.

The challenge is competition. Dell Technologies, Hewlett Packard Enterprise, Lenovo, Supermicro itself and numerous specialized integrators are all pursuing enterprise AI infrastructure, while NVIDIA increasingly provides reference architectures that reduce differentiation among hardware vendors. Cisco needs its networking, security and operational software to create enough additional value that customers choose a Cisco-led stack rather than purchasing similar GPU systems through another integrator.

How does Cisco’s $9.3 billion of AI infrastructure orders change the significance of this announcement?

Cisco’s AI infrastructure business is already operating at a scale far beyond experimental deployments. The company received approximately $4 billion of hyperscaler AI infrastructure orders in its fiscal fourth quarter alone, bringing fiscal 2026 orders to $9.3 billion. Cisco generated about $4 billion of AI infrastructure revenue during the year and expects that figure to reach approximately $7.5 billion in fiscal 2027.

That forecast means AI infrastructure could represent more than 10% of Cisco’s fiscal 2027 revenue if the company reaches the midpoint of its $72.2 billion to $73.4 billion full-year guidance. This is already large enough to influence group-level growth and margins, making expansion into rack-scale compute strategically more consequential than a routine product extension.

Cisco’s broader financial momentum is also strong. Fiscal fourth-quarter revenue increased 18% to $17.3 billion, while GAAP net income rose 51% to $3.9 billion. Networking product orders increased 40%, total product orders increased 35%, and fiscal 2026 revenue reached $63.3 billion, up 12%.

The less comfortable number is gross margin. Cisco’s non-GAAP product gross margin fell to 64.8% from 67.5% a year earlier even as revenue and orders accelerated. High-growth AI infrastructure can carry a different margin structure from Cisco’s traditional software and networking products, so becoming more deeply involved in server-heavy deployments may strengthen revenue while simultaneously changing the economics underneath that growth.

Can Cisco make enough profit from AI infrastructure to justify the faster hardware growth?

This is becoming the central investor question. Cisco’s fiscal 2026 AI infrastructure orders and revenue demonstrate genuine demand, but additional server and GPU-related content can carry lower gross margins than proprietary networking or software. If Cisco merely becomes a reseller of expensive compute hardware, revenue can rise sharply without producing the same operating leverage investors normally associate with high-value infrastructure software.

The Supermicro partnership is designed partly to avoid that outcome. Cisco can allow Supermicro to provide server engineering while concentrating its own differentiation around Silicon One, Spectrum-X networking, Nexus software, security, observability, support and certification services. The more revenue attached to those proprietary layers, the stronger the economic case becomes.

Cisco’s fiscal 2027 outlook provides room for the company to prove the model. Management expects revenue of $72.2 billion to $73.4 billion and non-GAAP earnings per share of $5.05 to $5.11, compared with fiscal 2026 revenue of $63.3 billion and non-GAAP EPS of $4.33. If AI infrastructure helps drive that revenue increase while overall operating margins remain resilient, concerns around hardware mix may ease.

Cisco also ended fiscal 2026 with $46.7 billion of remaining performance obligations, up 7%, giving it substantial contracted visibility outside the immediate AI hardware cycle. This matters because the company’s investment case remains broader than accelerated computing even as AI becomes one of its fastest-growing infrastructure categories.

What does Cisco’s August 25 share price say about investor sentiment after the AI order boom?

Cisco shares closed at $111.11 on August 25, up 0.8% in the session but about 0.4% below the August 18 close of $111.61. Compared with the July 24 close of $114.17, the stock was down approximately 2.7%, while its 52-week trading range stood at $66.13 to $130.37.

The more revealing comparison is with August 12, when Cisco closed at $123.88 before the market digested its fiscal fourth-quarter results and outlook. By August 25, the shares were roughly 10% below that level despite record revenue, strong earnings growth and $9.3 billion of annual hyperscaler AI orders. Investors appear to be distinguishing between the strength of AI demand and the margin profile of supplying that demand.

The Supermicro expansion gives Cisco another opportunity to improve the economics by attaching more of its networking, security and services portfolio to every AI infrastructure deployment. It also increases the importance of execution because Cisco is moving deeper into a market characterized by extremely rapid hardware cycles, intense competition and enormous capital requirements.

The strategic direction is increasingly clear. Cisco no longer appears content to be the network connecting somebody else’s AI servers. It wants to become the architecture through which enterprises, neoclouds and sovereign customers deploy entire AI factories, and fiscal 2027 will show whether that ambition produces both revenue growth and acceptable margins.