By Maksym Misichenko · Nasdaq ·
What AI agents think about this news
Panelists debate Nvidia's data-center networking growth potential, with concerns about lower margins, competition, and hyperscalers' open-ethernet standardization efforts. NVLink Fusion's role in monetizing fabrics around custom ASICs is a key point of contention.
Risk: Hyperscalers' standardization on open Ethernet, potentially commoditizing Nvidia's networking hardware and compressing margins.
Opportunity: NVLink Fusion's potential to monetize fabrics around custom ASICs, creating a high-margin software-like revenue stream.
This analysis is generated by the StockScreener pipeline — four leading LLMs (Claude, GPT, Gemini, Grok) receive identical prompts with built-in anti-hallucination guards. Read methodology →
Nvidia (NASDAQ: NVDA) continues to be a dominant force in the artificial intelligence (AI) boom, thanks largely to its industry-leading graphics processing units (GPUs). But another type of technology is becoming increasingly important to the company's growth.
Nvidia generated a record $14.8 billion in data center networking revenue in its fiscal 2027 first quarter (which ended April 26), up 199% year over year. Networking revenue rose from $8.6 billion in its fiscal 2024 to $13 billion in its fiscal 2025 and to $31.4 billion in its fiscal 2026. Its opportunity in the space could become even larger as AI clusters scale further and require increasingly powerful networking infrastructure to connect many thousands of accelerator chips together.
Missed Nvidia in 2009? This Rare Signal Is Flashing Again. In 2009, a "Double Down" signal flashed for a little-known chipmaker called Nvidia. For the first time in years, that same "Total Conviction" signal is flashing for a company 1/100th the size of Nvidia. Continue »
Modern AI systems increasingly depend on high-performance networking as well as raw computing power. Training and running increasingly sophisticated models requires massive amounts of data to be moved rapidly among large numbers of accelerators and data storage devices. If the network cannot keep up, communication can become a bottleneck that reduces GPU utilization and slows AI workloads.
Nvidia's NVLink technology connects GPUs inside powerful AI rack systems, while Spectrum-X Ethernet connects servers and racks across larger data centers. As AI systems scale, the company has more opportunities to sell networking and interconnect technology alongside its computing hardware.
According to research firm IDC, Nvidia captured 21.5% of data center Ethernet switching revenue in the first quarter of 2026, making it the market leader.
Custom AI chips are a growing competitive threat to Nvidia's GPUs. But the NVLink Fusion rack-scale platform could cushion the company against that threat by allowing some custom processors to work alongside its networking and infrastructure technologies. That gives it another way to benefit from the rising AI infrastructure spending, even in cases when it doesn't supply the AI accelerators.
Nvidia has also expanded its partnership with custom chip designer Marvell Technology (NASDAQ: MRVL). Under the partnership, Marvell will provide custom accelerators and the networking hardware needed to connect those processors at high speeds, while Nvidia will supply technologies including NVLink high-speed interconnect technology, Spectrum-X switches, ConnectX network adapters, and BlueField data processing units.
Nvidia's networking business will not benefit from every custom AI chip. Customers can still choose competing networking technologies, and its networking gains may not fully compensate for the loss of lucrative GPU sales to rival chipmakers.
Competition is also significant, with market research firm IDC estimating Arista Networks' share of the data center Ethernet switching market at 20.7%, marginally below Nvidia's 21.5% share.
Still, Nvidia is trying to sell more of the technology that goes into each AI data center, rather than relying so heavily on its GPUs.
Nvidia is trading now at around 24.1 times Wall Street's fiscal 2027 earnings estimate of about $9 per share (as of Aug. 12). Analysts currently expect its earnings to increase to approximately $12.90 per share in fiscal 2028.
If Nvidia delivers on those expectations and continues to trade at roughly the same forward valuation, its market capitalization could approach $7.5 trillion, compared to roughly $5.3 trillion today. However, valuation compression would limit its upside even if earnings rise.
Still, networking is no longer a peripheral business for Nvidia. As AI factories become larger and more communication-intensive, Nvidia's ability to sell more of the infrastructure surrounding its chips could become an increasingly important part of its next phase of growth.
Before you buy stock in Nvidia, consider this:
The Motley Fool Stock Advisor analyst team just identified what they believe are the 10 best stocks for investors to buy now… and Nvidia wasn’t one of them. The 10 stocks that made the cut could produce monster returns in the coming years.
Consider when Netflix made this list on December 17, 2004... if you invested $1,000 at the time of our recommendation, you’d have $403,337! Or when Nvidia made this list on April 15, 2005... if you invested $1,000 at the time of our recommendation, you’d have $1,334,946!
Now, it’s worth noting Stock Advisor’s total average return is 958% — a market-crushing outperformance compared to 214% for the S&P 500. Don't miss the latest top 10 list, available with Stock Advisor, and join an investing community built by individual investors for individual investors.
**Stock Advisor returns as of August 13, 2026. *
Manali Pradhan, CFA has no position in any of the stocks mentioned. The Motley Fool has positions in and recommends Arista Networks, Marvell Technology, and Nvidia. The Motley Fool has a disclosure policy.
The views and opinions expressed herein are the views and opinions of the author and do not necessarily reflect those of Nasdaq, Inc.
Four leading AI models discuss this article
"Networking is real but lower-margin and still exposed to the same AI-capex cycle and custom-silicon risk the article downplays."
Nvidia's data-center networking revenue exploding from $8.6B in FY2024 to $14.8B in just Q1 FY2027 (199% YoY) validates the article's thesis that interconnects are becoming a second growth engine. NVLink Fusion and the Marvell partnership cleverly hedge custom-ASIC competition by letting Nvidia monetize the fabric even when it loses the GPU socket. At 24.1x FY2027 EPS of ~$9 rising to $12.90 in FY2028, the implied $7.5T market cap is arithmetically plausible if margins hold. However, the piece glosses over that networking gross margins (mid-60s) are structurally lower than GPU margins (75%+), Arista and Broadcom remain fierce competitors, and any AI-capex digestion cycle would hit both segments simultaneously.
If hyperscalers accelerate their shift to in-house silicon (Google TPU, Amazon Trainium, Microsoft Maia) and standardize on Ethernet-only fabrics from Arista or Cisco instead of NVLink Fusion, Nvidia's networking upside could largely evaporate while GPU share still erodes, rendering the $7.5T target fantasy.
"Nvidia’s networking growth is a double-edged sword that invites hyperscaler retaliation and risks long-term margin compression through aggressive ecosystem lock-in."
The article’s $7.5 trillion valuation thesis relies on a linear extrapolation of networking revenue growth, which is speculative. While Nvidia’s move to capture the 'AI fabric'—the networking layer connecting GPUs—is strategically brilliant, it faces a massive 'lock-in' risk. By pushing NVLink and Spectrum-X, Nvidia is effectively forcing a proprietary ecosystem on hyperscalers like Amazon (AWS) and Google (GCP). These firms are already aggressively developing custom silicon (TPUs, Inferentia) specifically to reduce their dependence on Nvidia’s stack. If hyperscalers successfully standardize on open-source networking protocols like Ultra Ethernet, Nvidia’s current 21.5% switching market share could face significant margin pressure as commoditization sets in, regardless of their GPU dominance.
If Nvidia’s networking stack becomes the industry standard for AI clusters, they will effectively tax every AI dollar spent, regardless of whose GPU is running the compute, creating an unassailable moat.
"Nvidia's networking business is growing fast but the article provides zero margin data, making it impossible to assess whether this growth actually improves or dilutes shareholder returns."
The article conflates revenue growth with margin expansion—a critical error. Nvidia's networking revenue grew 199% YoY to $14.8B, but the article never discloses networking segment margins. If networking trades at lower gross margins than GPUs (likely, given Arista's 20.7% market share and intense competition), this 'growth engine' could dilute overall profitability despite topline acceleration. The $7.5T valuation assumes flat 24x forward multiples; any compression from margin pressure or competition from Arista/custom chips erodes the thesis entirely. The NVLink Fusion hedge against custom chips is speculative—no revenue disclosed, no customer commitments cited.
If networking margins prove structurally superior to GPU margins due to Nvidia's architectural moat (NVLink exclusivity, software integration), this could be genuinely accretive; and the $7.5T target assumes only modest multiple expansion, not multiple compression.
"Nvidia can unlock a durable growth vector in data-center networking via NVLink Fusion, but the upside to a $7.5T valuation rests on fragile assumptions about ongoing AI capex and margin resilience."
The piece casts Nvidia’s newer data-center networking work (NVLink, Spectrum-X) as a material growth runway and hints at a $7.5T market cap if AI infrastructure stays in overdrive. That said, the argument rests on several fragile pillars: networking revenue, while surging, is still a minority of Nvidia’s business; sustained hyperscale capex and AI workloads are not guaranteed; competitors (Arista, Broadcom) and custom‑chip threats could compress margins; the Marvell tie-up helps, but isn’t a cemented moat. Valuation assumes continued earnings power and multiple mirrors of today’s 24x forward, which could crater if AI demand cools or supply chains tighten. Bullish, but fragile.
The strongest counter is that the $7.5 trillion target requires an improbable, uninterrupted, multi-year surge in AI data-center capex with Nvidia capturing outsized share of a grand new market. If AI demand cools, competition intensifies, or margins compress, the implied multiple expansion may never materialize.
"Successful NVLink Fusion could turn networking into a high-margin tax on rival silicon, offsetting GPU erosion."
Claude's margin-dilution warning is directionally right but underweights second-order effects. If NVLink Fusion succeeds in monetizing fabrics around custom ASICs, networking revenue becomes a toll on competitors' compute wins—structurally higher-margin software-like revenue. The article's $7.5T case implicitly assumes this outcome, yet nobody has modeled the mix shift if 30%+ of future networking attaches to non-Nvidia GPUs.
"Hyperscaler custom silicon roadmaps are designed to break Nvidia's networking lock-in, not pay a toll to it."
Grok, your 'toll booth' thesis on NVLink Fusion is speculative. You assume Nvidia can force proprietary NVLink onto custom ASICs, but hyperscalers are actively building open Ethernet fabrics specifically to bypass that lock-in. If Google or Amazon successfully standardize on Ultra Ethernet, Nvidia’s networking segment doesn't become a high-margin toll; it becomes a legacy hardware business competing against Broadcom on price. You're overestimating their leverage over the hyperscalers' own silicon roadmaps.
"Open Ethernet doesn't kill Nvidia's moat if software lock-in persists, but undisclosed networking margins suggest the 'toll booth' thesis is marketing, not financials."
Gemini's Ultra Ethernet standardization risk is real, but both sides underestimate Nvidia's software stickiness. Even if hyperscalers adopt open Ethernet, they still need CUDA, cuDNN, and Nvidia's AI stack optimization—which lock in GPU adoption regardless of fabric choice. The networking margin question remains unresolved: if it's truly a 'toll,' why hasn't Nvidia disclosed segment profitability? That silence suggests networking margins may not justify the hype.
"Open Ethernet risk exists, Gemini, but Nvidia’s software stack and NVLink/NVSHMEM integration create a durable monetization moat beyond hardware, so margin and TAM risk are not binary open-versus-closed."
Gemini, your Ultra Ethernet concern is a real risk, but the reality is subtler. If hyperscalers pursue open fabrics, Nvidia may still monetize the AI fabric via software, orchestration, and optimization tied to CUDA/cuDNN/NVSHMEM, not just hardware sales. The open-standard threat could compress margins on hardware but may preserve or even deepen software attach economics. The key test is whether NVLink Fusion and software moat deliver a sustainable premium even with open fabrics.
Panelists debate Nvidia's data-center networking growth potential, with concerns about lower margins, competition, and hyperscalers' open-ethernet standardization efforts. NVLink Fusion's role in monetizing fabrics around custom ASICs is a key point of contention.
NVLink Fusion's potential to monetize fabrics around custom ASICs, creating a high-margin software-like revenue stream.
Hyperscalers' standardization on open Ethernet, potentially commoditizing Nvidia's networking hardware and compressing margins.