Wi-Fi 7 has one feature I would put at the center of an enterprise business case, and it isn't the one on the slide.

According to IDC, Wi-Fi 7 grew from less than 1% of enterprise dependent access-point revenue in the first quarter of 2024 to 44.5% in the first quarter of 2026. Dell'Oro separately reported that Wi-Fi 7 represented 37% of indoor access-point units shipped in that quarter. Revenue share, unit share, and client adoption are three different things. The purchasing shift is real, but it doesn't mean most users will see anything close to the number on the datasheet. (https://www.idc.com/resource-center/blog/wi-fi-7-captures-44-of-enterprise-wlan-dependent-access-point-revenue-as-1q26-market-grows-15-9-to-nearly-2-7-billion/) (https://www.prnewswire.com/news-releases/wlan-market-continues-double-digit-growth-in-1q-2026-despite-memory-shortage-according-to-delloro-group-302790707.html)

I'm not trying to argue against Wi-Fi 7. I've been running wireless through three generations of this same cycle. Peak rate gets the attention, while the uplink, client mix, security design, or channel plan usually decides what users actually feel. With Wi-Fi 7, Multi-Link Operation is where I would start the business case.

Three generations, same lesson

Wi-Fi 5 Wave 2 sold downlink multi-user MIMO, or MU-MIMO, as the density fix. In production, the gain depended on compatible clients, usable client groups, channel design, and a wired path that could carry the traffic. Many access points still terminated on 1 GbE.

Wi-Fi 6 was marketed around a 9.6 Gbps theoretical maximum. A typical laptop had one or two spatial streams, not eight. Cisco's testing guide models about 1.5 Gbps of achievable throughput for a two-stream Wi-Fi 6 client at the highest modulation and channel width under its stated assumptions. Useful performance, certainly, but still a different number from 9.6 Gbps. (https://www.cisco.com/c/en/us/support/docs/wireless-mobility/wireless-lan-wlan/212892-802-11ac-wireless-throughput-testing-and.html)

Wi-Fi 6E was sold on 6 GHz. In 2021, Aruba cited a 650 Group forecast of more than 350 million 6 GHz-capable devices entering the market in 2022 and more than 200% unit growth for enterprise Wi-Fi 6E access points. Those were forecasts, not a count of anybody's managed fleet. Plenty of us lit up 6 GHz radios that remained lightly used while the endpoints caught up. (https://www.businesswire.com/news/home/20210525005243/en/Aruba-Introduces-Industry%E2%80%99s-First-Enterprise-Grade-Wi-Fi-6E-Solution)

Each time, the bottleneck was somewhere the brochure wasn't looking.

What Wi-Fi 7 changes

Multi-Link Operation, or MLO, lets a compatible client and access point establish more than one link inside a multi-link association. Depending on the client and MLO mode, those links can be used concurrently or one at a time. That gives the connection another prepared path when one link is busy. It doesn't eliminate roaming between access points, and it only helps when both ends support the needed mode.

Here is the catch nobody puts on the slide: Wi-Fi 7 support alone doesn't tell you which MLO behavior a client will negotiate.

Simultaneous transmit and receive, or STR, can use two links concurrently. Enhanced multi-link single radio, or eMLSR, monitors multiple links but carries data on one selected link at a time. STR is optional for client devices. Qualcomm documents high-band simultaneous MLO in its FastConnect 7800, and Cisco used a Qualcomm 7800-based STR client in a published test. MLO is a device-by-device capability, not a box to check on a fleet spreadsheet. (https://www.qualcomm.com/wi-fi/products/fastconnect/fastconnect-7800) (https://blogs.cisco.com/networking/wi-fi-7s-multi-link-operation-mlo-dissection-from-packets-to-performance)

Channel width needs the same reality check. With the full 1200 MHz allocation in low-power indoor operation, there can be up to three non-overlapping 320 MHz channels. In Standard Power operation, Cisco Meraki documents regulatory-plan maxima of one in the United States and two in Canada, with the location-specific Automated Frequency Coordination response determining what each access point may actually use. A 320 MHz channel can make one test client look excellent while leaving fewer channels to reuse across the building. In many dense designs, 80 MHz or 160 MHz will make more sense. (https://documentation.meraki.com/Wireless/Design_and_Configure/Architecture_and_Best_Practices/Wi-Fi_7_%28802.11be%29_Technical_Guide)

The throughput numbers also need their test conditions attached. Intel lists a 5.8 Gbps maximum physical-layer rate for its two-stream BE200 at 320 MHz and 4096-QAM. In Cisco's STR test, using 20 MHz at 2.4 GHz plus 40 MHz at 5 GHz, Wi-Fi 7 averaged 747 Mbps, compared with 506 Mbps when MLO was disabled and the client ran in Wi-Fi 6 mode. Both numbers are legitimate. Neither predicts a mixed-client office. (https://www.intel.com/content/www/us/en/products/sku/230078/intel-wifi-7-be200/specifications.html) (https://blogs.cisco.com/networking/wi-fi-7s-multi-link-operation-mlo-dissection-from-packets-to-performance)

There is no useful global percentage for Wi-Fi 7 clients in an enterprise fleet. Cisco says many large enterprises report that about 30% of their clients are 6 GHz-capable, but that broader group includes Wi-Fi 6E and Wi-Fi 7. Pull the actual Wi-Fi 7 and MLO capabilities from your controller and endpoint inventory before funding a feature your clients may not negotiate. (https://www.cisco.com/c/en/us/products/collateral/networking/wireless/wifi7-future-of-wireless-dg.html)

Buy now, or wait?

Buy now if:

  • You're at a natural refresh. Dell'Oro reported in June 2026 that average selling prices had begun to rise, several vendors had raised list prices because of higher memory costs, and lead times were expected to remain variable. That makes timing worth watching, but it doesn't replace a design. (https://www.prnewswire.com/news-releases/wlan-market-continues-double-digit-growth-in-1q-2026-despite-memory-shortage-according-to-delloro-group-302790707.html)
  • You run a genuinely dense venue, such as an arena, lecture hall, or conference center, where 6 GHz capacity, scheduling, multiple resource units, or preamble puncturing addresses something you've measured.
  • You have a reliability-sensitive or latency-sensitive use case, such as automated guided vehicles, closed-loop control, clinical telemetry, or wireless audiovisual production, and you can pilot the exact endpoints and applications.
  • Your access layer already has the multigigabit ports, Power over Ethernet, cabling, controller capacity, and licensing the selected access points require.

Wait if:

  • Your Wi-Fi 6E equipment has useful life left, and the complaints are about coverage or roaming. A new access point in the same location won't repair a poor radio-frequency design. (https://www.cisco.com/c/en/us/products/collateral/networking/wireless/wifi7-future-of-wireless-dg.html)
  • Your inventory shows too few Wi-Fi 7 clients, or the wrong MLO modes, for the use case.
  • You have 1 GbE access ports and no switching or cabling budget. The wired edge can become the ceiling before the radio does.
  • The bottleneck is the WAN, firewall, or software-as-a-service path. Faster Wi-Fi won't remove it.

What the access-point quote leaves out

The access point may be the easiest line item.

  1. Power by model. Many full-capability tri-radio access points need IEEE 802.3bt for unrestricted operation. The standard can deliver up to 51 W to a Type 3 powered device and 71.3 W to Type 4. The result still depends on the model. Zyxel documents full 4T4R operation for the WBE660S on 802.3bt, reduced 2T4R operation on 802.3at, and disabled Wi-Fi radios on 802.3af. (https://ethernetalliance.org/wp-content/uploads/2019/12/WP_EA_Overview8023bt_V2p1_FINAL.pdf) (https://support.zyxel.eu/hc/en-us/articles/14843610154642-WBE660S-WIFI7-Access-Point-Power-Supply-special-features-and-device-behavior-in-case-of-power-supply-problems)
  2. The wattage label isn't enough. Cisco's Wi-Fi 7 power reference says its Wi-Fi 7 access points require standards-based PoE and don't accept proprietary Cisco UPoE as a substitute. It also says the AIR-PWRINJ7 can provide full power but tops out at 5 Gbps, while the CW-INJ-8 supports 10 Gbps. Check the access point, switch, injector, and software release together. (https://www.cisco.com/c/en/us/support/docs/wireless/catalyst-9100-access-points/215467-cisco-access-point-power-requirements-qu.html)
  3. Switch budget, not port count. A switch can have enough PoE-capable ports and still lack the total power budget. Include startup, failover, attached devices, and operating headroom.
  4. Uplink and cable. IEEE 802.3bz brought 2.5GBASE-T and 5GBASE-T to much of the installed Cat 5e and Cat 6 base at distances up to 100 meters, subject to the cable and rate. For a new run that must support 10GBASE-T to 100 meters, Cat 6A is the safer baseline. Existing Cat 6 may support 10GBASE-T to 37 meters, or as far as 55 meters when alien crosstalk permits it. Test the plant before replacing it, and before promising it will work. (https://archive.nbaset.ethernetalliance.org/wp-content/uploads/2019/02/FAQ_022119_v2.pdf) (https://www.flukenetworks.com/knowledge-base/applicationstandards-articles-copper/10gbase-t-field-testing-requirements)
  5. MLO changes client identity. A multi-link device has an MLD MAC address and per-link MAC addresses. Cisco keys its multi-link statistics to the MLD address while retaining per-band detail. Zyxel documents translating a radio address to the MLD address on the wired side. Authentication, accounting, monitoring, and enforcement tools may not all present that client the same way. (https://www.cisco.com/c/en/us/td/docs/wireless/controller/9800/17-15/config-guide/b_wl_17_15_cg/m_wi-fi-7_operations.html) (https://community.zyxel.com/en/discussion/25379/understanding-multi-link-operation-mlo)
  6. Security is part of the feature. Cisco Meraki documents WPA3 or Enhanced Open with Protected Management Frames as the requirement for a client to use Wi-Fi 7 data rates and MLO. That doesn't automatically make the entire service set identifier, or SSID, WPA3-only. Cisco documents mixed security configurations that can retain older clients while qualified clients use MLO. The exact result depends on the platform, software, security method, and client. (https://documentation.meraki.com/Wireless/Design_and_Configure/Architecture_and_Best_Practices/Wi-Fi_7_%28802.11be%29_Technical_Guide) (https://www.cisco.com/c/en/us/td/docs/wireless/controller/9800/17-18/config-guide/b_wl_17_18_cg/m_wi-fi-7_operations.html)
  7. Client refresh. The chipset, driver, operating system, security mode, and power policy affect what a device will negotiate. This is the line item nobody wants in the wireless business case, and the one that decides whether MLO pays off.

MLO changes what authentication, accounting, inventory, logging, captive portals, and incident response may see. Test those paths before enabling it broadly.

The mild take

Wi-Fi 7 is a good standard being sold on its weakest argument. Buy it when MLO, client readiness, radio-frequency design, and refresh timing support the case. Use the channel width the design supports, and plan around the modes the fleet actually negotiates. Treat 46 Gbps as a standard-level ceiling, not a user expectation. (https://blogs.cisco.com/networking/how-wi-fi-7-is-about-to-transform-business-as-we-know-it)

What nags at me is that Wi-Fi 8 is explicitly being developed around Ultra High Reliability. TP-Link has announced an October 2026 launch for a pre-standard consumer platform, and Qualcomm says commercial products using its Wi-Fi 8 platforms are expected late in 2026. The IEEE task group still projects final approval in May 2028. None of that makes a justified Wi-Fi 7 refresh obsolete. It does show where the industry's attention is moving. (https://www.tp-link.com/in/press/news/22370/) (https://www.qualcomm.com/news/releases/2026/03/qualcomm-debuts-ai-native-wifi-8-portfolio-unifying-client-and-n) (https://www.ieee802.org/11/Reports/tgbn_update.htm)

Has anyone measured a user-visible improvement from MLO in a mixed-client production environment? Not a lab number. A ticket that stopped getting filed. I'd like to hear it.

Also available on LinkedIn

Continue with this article and its discussion on LinkedIn.