Close
Close

NIC Lead Times Just Hit 52 Weeks. What China's H200 Move Means for What's Next

 Network interface card lead times just hit a new high, and the reason has as much to do with Beijing as with Nvidia. 

Key Takeaways
ConnectX-7 (MCX7) network interface card lead times have stretched past 52 weeks in parts of the market, driven by large hyperscaler orders in Asia absorbing available capacity.  
Reuters reported on July 7, 2026 that China plans to let its top AI companies buy a limited volume of Nvidia H200 GPUs. Bloomberg followed on July 8 naming Alibaba, ByteDance, and DeepSeek as early applicants. If approved, expect a parallel surge in demand for older Mellanox NIC cards as Chinese hyperscalers move fast on the peripheral networking gear that supports new GPU clusters.  
Nvidia and Mellanox ConnectX-7 and ConnectX-8 spot prices are climbing ahead of an expected official price increase.  
Switches and optical transceivers are tightening on a similar timeline. MQM9790-NS2F lead times jumped from 22 to 42 weeks, and 800G and 1.6T transceiver lead times now run 20 to 26-plus weeks.  

Where network interface card lead times and pricing stand right now

The networking hardware market continues to tighten as AI-driven hyperscaler demand absorbs most of global production. Network interface cards are the clearest example. Nvidia's ConnectX-7 (MCX7) family is now quoting 52 weeks or more on multiple configurations, and the 200G dual-port and 400G single-port variants face the most severe supply constraints in the market today.

Fusion's trading desk, working across our supplier network built over 25 years, is tracking reports from a distributor in Asia that Nvidia is extending lead times across the entire MCX6, MCX7, and MCX8 network interface card lines. The driver is straightforward. Large hyperscalers are placing bulk orders and absorbing capacity that would otherwise reach the open market.

Spot prices for CX-7 and CX-8 cards are rising, and an official price increase from Nvidia and Mellanox is expected. Buyers who lock in volume before that increase lands have a real cost advantage.

Network interface card lead time movement

NIC family

Typical lead time a year ago

Current lead time

Primary constraint

ConnectX-6 (MCX6)

12 to 16 weeks

Extended, allocation-based

Hyperscaler bulk orders absorbing supply

ConnectX-7 (MCX7)

20 to 26 weeks

52+ weeks on 200G/400G variants

200G dual-port and 400G single-port scarcity

ConnectX-8 (MCX8)

Not yet released

Allocation-based at launch

Early hyperscaler reservations

Top requested NIC MPNs across Fusion's network in the past 30 days, in order:

  1. MCX623106AN-CDAT, ConnectX-6 Dx
  2. MCX653106A-HDAT, ConnectX-6 VPI
  3. MCX631102AN-ADAT, ConnectX-6 Lx
  4. MCX755106AS-HEAT, ConnectX-7 InfiniBand/VPI
  5. MCX75310AAS-NEAT, ConnectX-7

How Beijing's H200 approval is reshaping network adapter demand

The story procurement teams should watch closest right now sits outside the NIC market itself. According to Reuters reporting on July 7, 2026, China is preparing to let a limited group of top AI companies buy Nvidia H200 GPUs. Bloomberg reported the next day that Chinese officials had told firms including Alibaba, ByteDance, and DeepSeek that they can apply to purchase a set number of the processors, provided they justify the quantity and intended use.

If the approval moves forward, expect a fast, direct effect on the network adapter market. Every new GPU cluster needs matching network interface cards, switches, and transceivers to connect it. Chinese hyperscalers preparing to deploy H200s are likely to move quickly on older, readily available Mellanox NIC cards rather than wait on newer, allocation-constrained parts. That puts additional pressure on a NIC supply chain that is already tight from US and European hyperscaler demand.

This is consistent with what the broader networking equipment market is already showing. Industry reporting on the sector's Q3 2026 outlook has described demand staying resilient into the traditional third-quarter peak season, even as component shortages and price increases cloud the picture. A confirmed H200 approval would add a second demand source on top of that existing peak-season pull.

Procurement teams sourcing network interface cards and network adapters for AI infrastructure builds should plan for continued tightening through the rest of 2026, not a return to normal lead times.

 

 For more on the capacity constraints behind this demand, see Inside the AI Bottleneck: CoWoS, HBM, and 2-3nm Capacity Constraints Through 2027 and How Hyperscaler Spending Influences Semiconductor Supply Chains

Why switches and optical transceivers are tightening on the same cycle

Network interface cards are not moving in isolation. Switches and optical transceivers, the other two components that make up a complete networking build, are showing the same pattern of extending lead times and rising prices.

On switches, Nvidia's MQM9790-NS2F saw lead times jump from 22 to 42 weeks, with open market prices rising alongside the delay. Broader Nvidia switch lead times are reaching 24 weeks or more across multiple customer accounts, and some buyers are now evaluating alternate switch brands to avoid the wait. Nvidia's Q3400 switch line now requires transceivers to be bundled together at the time of purchase, a change that adds cost and complexity to a straightforward switch order. Trade sources also point to a potential price increase, possibly a doubling, on the SN5610.

On optical transceivers, a shortage of digital signal processor chips is the key bottleneck. Broadcom's DSP supply has been constrained by limited 3nm capacity, a pattern that has been reported across the broader optical networking supply chain and is pushing 800G and 1.6T transceiver lead times to 20 to 26 weeks or more. Multimode transceivers are the hardest hit category. Multiple customers have reported two-month delivery delays on 800G twin-port OSFP transceivers specifically.

Top requested transceivers across Fusion's network in the past 30 days, in order:

  1. MMS4A20-XM800
  2. 980-9IAT0-00XM00
  3. MMA1Z00-NS400

Related reading: GPU Shortage and Price Increases in 2026, on the same demand pressure driving switch and transceiver pricing. 

What's happening with RAID controllers and other data center components

RAID controllers round out the current networking-adjacent supply picture, and they show the sharpest reported pricing move of any category covered here. Trade sources in Korea have flagged a Broadcom MegaRAID price increase of more than 20 percent. Lead times on multiple RAID controller part numbers are stretching into Q4 2027 at best, with some customers reporting quotes of over a year.

Top requested RAID controllers across Fusion's network in the past 30 days, in order:

  1. 05-50077-00
  2. 05-50076-00
  3. 05-50077-01

What procurement teams should do now

The common thread across network interface cards, switches, transceivers, and RAID controllers is the same: demand from AI infrastructure builds is outrunning supply, and official price increases are expected to continue through the rest of 2026. Waiting is the most expensive option available right now. Procurement teams should take three steps.

  1. Place orders on long-lead NIC, switch, and transceiver parts ahead of confirmed price increases, rather than after them.
  2. Identify approved alternate MPNs and brands now, before a shortage forces a rushed substitution.
  3. Work with a distribution partner that holds independent inventory positions and can source outside standard franchise allocation.

Fusion tracks lead time and pricing movement across the networking hardware market in real time through Scout, our inventory and market intelligence platform, and sources components through a supplier network built over 25 years. All parts move through Fusion's AS9120B, AS6081, and ISO 9001 certified quality process before they ship. Contact Fusion's team to check current availability and lead times on the parts named in this update.  

  1. What is a network interface card (NIC)?
  2. What is a NIC card used for?
  3. What is an optical transceiver?
  4. What connectors do SFP transceivers use?
  5. What does transceiver mean in networking?
  6. What types of NIC cards are available?

A network interface card, or NIC, is the hardware component that connects a server, storage system, or other device to a network. It handles the physical connection and the data transfer between the device and the network, whether that connection runs over Ethernet or InfiniBand.

A NIC card lets a device send and receive data over a network. In a data center, network interface cards connect servers to switches so that data can move between compute, storage, and networking equipment at the speeds a given workload requires.

An optical transceiver is a module that converts electrical signals into optical signals for transmission over fiber, and converts them back to electrical signals on the receiving end. Transceivers plug into switches, routers, and network interface cards to extend network connections over fiber optic cable.

SFP transceivers connect to network equipment through an SFP port and connect to the network through fiber optic connectors, most commonly LC connectors for single fiber pairs. The exact connector type depends on the transceiver's data rate and reach.

In networking, a transceiver is any device that both transmits and receives signals. The term combines transmitter and receiver. Network transceivers convert data between the electrical format used inside equipment and the optical or electrical format used on the network cable.

NIC cards vary by network type (Ethernet or InfiniBand), speed (from 1GbE up through 800Gb/s and beyond), port count, and form factor, including standard PCIe cards, OCP form factor cards, and mezzanine cards. Data centers running AI workloads typically use high-speed models such as Nvidia's ConnectX-6, ConnectX-7, and ConnectX-8 families.