Skip to content
Townsend Networks
Supply Chain

The AI Buildout Is Squeezing Enterprise Switch Supply: What It Means for Your Refresh

Townsend Networks Team

Hardware engineering & procurement desk

The AI Buildout Is Squeezing Enterprise Switch Supply: What It Means for Your Refresh

If you've priced a switch refresh recently, you've probably noticed two things: quotes are firmer than they used to be, and delivery dates come with more hedging. Neither is your VAR's fault. The biggest infrastructure buildout in the industry's history is happening right now, and enterprise buyers are sharing a supply chain with it.

The demand shock, in numbers

IDC's Worldwide Quarterly Ethernet Switch Tracker put the global Ethernet switch market at $14.5 billion in Q2 2025, up 42.1% year over year, with the data-center segment growing 71.6% while campus and branch grew 12.5%. The third quarter held the pace: $14.7 billion, up 35.2%, with data-center revenue up 62%. Read those growth rates side by side: the AI side of the market is growing five times faster than the enterprise side, on shared manufacturing capacity.

Dell'Oro Group reports that the majority of AI back-end network ports reached 800 Gbps during 2025, and forecasts AI back-end switch sales alone to exceed $100 billion in cumulative spending over 2025-2029.

The vendors confirm it from their side of the table:

  • Cisco crossed $1 billion in AI infrastructure orders from webscale customers a full quarter ahead of its fiscal-2025 target and finished FY2025 with more than $2 billion, on roughly $56.7 billion in revenue. The pace then tripled: $5.3 billion in AI infrastructure orders through the first three quarters of fiscal 2026 (reported May 2026), with the company now expecting roughly $9 billion for the year.
  • Arista closed 2025 at roughly $9.0 billion in revenue, up about 25%, hit its $1.5 billion AI-networking target, and in May 2026 raised full-year 2026 guidance to about $11.5 billion. On the same earnings call, management told investors it expects roughly 52-week lead times on critical silicon through 2027, and that it has grown its own purchase commitments to $8.9 billion to lock in supply.
  • Broadcom began shipping Tomahawk 6, a 102.4 Tbps switch ASIC, in June 2025, and every hyperscaler wants systems built on it.

Why your campus refresh feels it

Enterprises aren't buying 800G AI fabrics. But the same silicon foundries, the same optics manufacturers, and the same vendor production slots serve both markets. When hyperscalers pre-buy transceiver capacity (and analysts at Dell'Oro and LightCounting both flagged tight supply for high-speed optics through 2025 and into 2026), everything downstream gets tighter.

Optics are the clearest example. An 800G transceiver and a 100G campus transceiver aren't the same part, but they compete for the same fab lines, lasers, and packaging capacity, and the 800G buyer signed a bigger check. TrendForce projects 800G-class transceiver shipments jumping from roughly 24 million units in 2025 to roughly 63 million in 2026, against a global shortage of the EML lasers inside them; LightCounting estimated transceiver demand running about 30% ahead of supply into 2026. The overflow lands on everyone else's delivery dates.

Three effects show up in enterprise quotes:

  • Lead times. Typical quoted lead times in the channel for mainstream enterprise switches now run several weeks to a few months, and stretch further on high-density PoE models and optics-heavy configurations. That's far better than the 2021-2023 crisis, when some Catalyst SKUs quoted 300+ days, but it's no longer "ships Thursday" either.
  • Price firmness. Vendors pushed through list-price increases in 2025 partly attributed to tariffs, then again in early 2026 as DRAM contract prices spiked, increases concentrated in memory-heavy hardware. With demand outrunning supply, there's little discounting pressure on current-generation gear.
  • Allocation risk. When a vendor has to choose between a hyperscaler's standing order and a mid-size enterprise's one-time refresh, you know who waits.

Planning a refresh in this market

Start the clock earlier. If your refresh window is Q4, get quotes in Q2. Budget approvals that used to tolerate a two-week procurement cycle now need a quarter of runway on new equipment.

Split the bill of materials. The parts of your network that genuinely need new (say, Wi-Fi 7 access points or a core with the latest software features) are a fraction of most refreshes. Access-layer switching, distribution, spares, and optics are commodities where the previous generation does the same job. A Catalyst 9300 moves the same packets whether it shipped from a distributor last week or from our bench yesterday; the difference is a lead time measured in days instead of a fiscal quarter.

Protect the project schedule, not just the budget. The expensive failure mode in 2026 isn't overpaying for a switch: it's a cutover weekend slipping because one optics line item quoted twelve weeks. Walk the bill of materials and flag every part with a vendor lead time longer than your project buffer; those are the lines to source from stock, whoever's stock it is.

Use the secondary market as a pressure valve. Refurbished enterprise switches ship from stock (no allocation, no quarter-long lead time) at a fraction of list price. Every unit we sell is bench-tested by our own technicians and backed by a 3-year warranty, which outlasts the support window on plenty of new gear.

Buy optics like they're strategic. They are now. Compatible and refurbished optics are one of the largest line-item savings in any build, and stocking a shelf of spares costs little compared to a project stalled waiting on transceivers.

The AI buildout isn't slowing down: Cisco, Arista, and Broadcom are all guiding higher for 2026. Enterprises that treat supply as a planning input, not a surprise, will refresh on schedule and under budget.

Need switches or optics without the lead time? Browse our tested switch inventory and optics, or request a quote. A human answers, usually the same day.

Sources

Quick answers

Why are enterprise switch lead times longer in 2026?

The AI buildout is consuming shared manufacturing capacity. Data-center Ethernet switching grew 62-72% year over year through 2025 (roughly five times faster than campus), and hyperscalers pre-buy silicon, optics, and vendor production slots. Enterprise quotes inherit longer lead times, firmer pricing, and allocation risk.

How can I keep a network refresh on schedule despite lead times?

Start quotes a quarter earlier than usual, flag every line item whose vendor lead time exceeds your project buffer, and source commodity tiers (access switches, optics, and spares) from tested refurbished stock that ships from the shelf in days instead of a fiscal quarter.

Need help planning your next move?

Tested, warrantied hardware with real availability, and cash offers for decommissioned gear. Quotes answered by a human.

Keep reading