Teams Gets Faster by Paying Down Main-Thread Debt

Microsoft cut Teams chat-switch latency by 20% in the first half of 2026 by firing data queries on click and collapsing them into one round-trip. Daily users hit by hangs on macOS and iOS fell 35% in the first quarter. People search on iOS got 25% faster.

Ihor Lukyanenko and four Teams colleagues posted the work on June 10, 2026, after the May 2025 chat and channels redesign. The fixes target blocking work that the 2023 WebView2 rebuild never fully cleared off the main thread.

Warm Switches, Cold Switches, and a 20% Cut

Switching chats is one of the most repeated moves in Microsoft Teams, and Microsoft reduced chat switch latency by 20% on desktop and web in the first half of 2026. The company tracked that figure over a six-week window in May 2026. Older machines and weak networks were the users it named as the ones who would feel it.

Not every switch costs the same. A warm switch returns to a chat already in memory and should feel instant. A cold switch opens a conversation that has not been loaded lately, so Teams has to fetch it before the pane can paint. Cold switches sat at the slow tail, and earlier render work had barely moved them.

CHAT SWITCH TYPES IN 2026

Microsoft listed three reasons cold switches spiked. The conversation query started too late, during the React render, while the main thread was still tearing down the previous chat. Several queries then ran one after another, each with its own trip and render. Non-critical responses could jump the queue and block the content the user was waiting to see.

The Click Handler Now Goes First

The repair is blunt. Teams now fires the data query in the click handler, before rendering starts, so the data layer gets a head start while the main thread paints anything that does not need that payload. All of the data for a switch is packed into one request. One switch now means one data round-trip, which ends the waterfall of small queries that used to stack at the tail.

Rendering pauses when it actually needs that payload, then resumes when the response lands. After a single pass, Teams yields so the browser can paint the frame. Warm and cold switches now sit close enough that the leftover gap is the data layer’s response time. Caching from earlier work shrinks even that wait.

That is a client confession as much as a speed win. GraphQL was supposed to fetch a page in one shot, but engineers had kept REST-era habits, firing lots of small queries and doing business logic on the main thread. The May 2025 chat and channels launch already chased that pattern. It cut switches into Chat from Activity or Calendar by 17% at the 95th percentile and made chat revisits up to 40% faster. The 2026 work is the same bill, collected on the cold path that those render passes could not hide.

Why Teams Froze on Mac and iPhone

A hang is simpler than latency. The app stops taking taps, clicks, or scrolls, even for a moment, and the next move is often a force quit. In the first quarter of 2026, Microsoft went after the most common hangs on macOS and iOS. Daily users hit by those hangs fell 35%.

On the Mac, a health monitor thread pokes the main thread and waits. If the reply is late, it snapshots the call stack and ships that as hang data. A tool called StackDecoder turns those stacks into readable function names, groups matching patterns, and ranks hangs by how often they hit. The ranked list is what they fixed.

THE MACOS FREEZE SOURCES

• Network checks: The native network status tracker made a blocking call on the main thread, and moving it to the background removed the single largest source of macOS freezes.
• WebView2 loading: Loading the WebView2 dynamic library on the main thread caused 6% of all freezes, so the library now preloads in the background.
• Error reporting: A URL request during web app error reporting blocked the main thread; making that call asynchronous cleared 1.7% of freezes.

The iOS side was a longer grind, not one smoking-gun API. The team cached expensive work such as app badging, chat list rendering, and list view-model math, pushed costly jobs off the main thread, and deferred work that did not have to run now. Database access was rebuilt so user taps jump the queue ahead of background jobs, with scenario-aware queuing, scheduling, and cancellation.

The common thread across all of these is simple: move blocking system API calls off the main thread. By systematically applying this principle and validating each fix through health monitor telemetry, we are steadily driving freeze rates down across the entire user population.

Ihor Lukyanenko, Microsoft Teams engineering, Microsoft Teams Blog

A 35% drop in daily users hit by hangs is also an admission. The Mac and iPhone clients were locking up often enough to show in fleet data, not as rare edge cases. Chat switching is a trust problem in the same way. If the last tap froze, the next one gets a double tap or a quit.

WebView2 Left Blocking Work on the Main Thread

Teams did not start here. Development began in 2015 on Electron as the original host, with AngularJS and a pile of custom HTML controls, because that stack shipped a web and desktop client quickly. As meetings, channels, and apps piled on, the client strained device memory and CPU. The 2023 rebuild set out the move from Electron to WebView2, React in place of Angular, Fluent UI controls, and a client data layer worker so fetches would not fight the UI thread.

WebView2 was the resource pitch. On Windows it shares Chromium files with Edge through the WebView2 desktop host layer, which cuts disk use and memory against a private Electron runtime. Macs never got that free lunch, because they do not ship Edge’s runtime. A WebView2 app on macOS still carries its own copy, which is why a library load on the main thread could still freeze 6% of Mac sessions in 2026.

Jeff Teper wrote on March 26, 2024, that new Teams was up to twice as fast and used 50% less memory, and that more than 75% of monthly active users were already on it. Meeting join time was down more than 70% in that same post. Those were real jumps, and they were not the end of main-thread debt. Copilot sits in the 2026 lede as a reason Teams has to be ready on first tap, while a five-year AI meeting archive keeps more history in the same client. Feature work did not pause while the hang list was ranked.

THE TEAMS PERFORMANCE BILL

1. 2015: Teams ships on Electron and AngularJS so web and desktop can move together.
2. March 27, 2023: Microsoft outlines the WebView2, React, and data-layer rebuild, after a two-framework stretch that made classic Teams slower while React pieces landed inside it.
3. March 26, 2024: New Teams is described as up to twice as fast with 50% less memory, and more than 75% of monthly active users are on it.
4. May 2025: The combined chat and channels experience reaches general availability, with 17% faster switches into Chat at p95 and revisits up to 40% faster.
5. November 2025: A follow-up cites video rendering, memory use, and platform-level work, then the 2026 post picks up from there.
6. First quarter of 2026: Hang hunts on macOS and iOS cut daily users impacted by 35%.
7. First half of 2026: Chat-switch latency falls 20% on desktop and web, and iOS people search is 25% faster.

Each row is a partial payment. Query waterfalls, nested React updates, and blocking native calls kept showing up after the “new Teams” label had already been sold as the fix.

The Hang Rate on Apple Hardware

Windows still gets the shared Edge runtime. The hang chapter is written for Apple hardware because that is where the health monitor found the worst main-thread blocks, and because a freeze there is public. In a conference room, a locked Mac is a stalled meeting, not a private stutter. Mixed-platform Teams Rooms hardware kits do not save you if the laptop sharing the screen has stopped taking clicks.

Schools and agencies, which the product still lists as core Teams customers, also live on older Macs and iPads and on networks that make a cold switch obvious. Microsoft said the 20% chat-switch cut would show up most on slower networks and older devices. That is the fleet that never got a new-PC refresh when classic Teams got heavy, and it is the fleet that still notices when a network status check runs on the UI thread.

The iOS Hang Rate work is the same idea in Apple’s own metric: how often, and for how long, the app ignores input in real use. Caching, database queue jumps, and deferred work are unglamorous, and they are what a phone user feels when the chat list finally scrolls with the thumb.

People Search on iOS Is 25% Faster

Search and @mentions are the other high-frequency tap on a phone. Microsoft modernized the people picker in chats and channels on iOS and says people search results are 25% faster. The plumbing is a cleaner query pipeline, less database queue congestion, and higher throughput, using the same task queuing that now puts user-started work ahead of background jobs.

That number is narrower than the chat-switch cut. It is iOS only, and it is search, not the whole app. It still sits on the same spine. If the database queue is jammed, a name lookup feels like a hang even when the process is technically alive. Clearing that queue is how a 25% search gain and a 35% hang-user drop can come from the same quarter of work.

Microsoft Says Performance Is Never Finished

Lukyanenko’s close is that performance is never finished, and that some of the next gains will be visible while others stay behind the scenes. The honest remainder is already in the chat-switch write-up. Once the waterfall is gone, the wait you still feel is the data layer. Microsoft cannot paint its way around a slow network response.

The 2025 crew already put performance gates on pull requests, live alerts when latency crosses a line, and regular profiling, including on low-end devices. Those gates exist because the 2023 rebuild did not freeze the client in amber. New panes, Copilot surfaces, and archive defaults keep adding work to the same main thread the hang monitor is watching.

For now the published score is specific. Chat switches on desktop and web are 20% quicker than they were at the start of 2026. Daily hang victims on Mac and iOS are down 35% from the first-quarter hunt. Name search on iOS is 25% quicker. The leftover difference between a warm chat and a cold one is the time the data layer takes to answer, which is the part of the bill that now sits off the device.

Frequently Asked Questions

What Is a Warm Switch Versus a Cold Switch in Teams?

A warm switch reopens a chat that is already in memory, so it should paint almost at once. A cold switch opens a conversation that has not been loaded recently, so Teams must fetch it first. Microsoft says that at higher percentiles, cold switches dominate the latency distribution, which is why the 2026 work targeted that path instead of another round of warm-path rendering tweaks.

When Did Microsoft Replace Electron in the Teams Desktop App?

Teams began in 2015 on Electron and AngularJS. The commercial rebuild was described on March 27, 2023, with a Windows preview first and Mac, web, education, and VDI support to follow later that year. Microsoft built the consumer client on the new stack first, then extended that codebase to commercial features, after an awkward stretch when React pieces ran inside classic Teams and made the old app heavier.

What Does Apple Hang Rate Mean for Teams on iOS?

Hang Rate is Apple’s metric in Xcode Organizer for how often and how long an app stops responding to taps, scrolls, or typing in real-world use. Microsoft treated it as the iOS scoreboard for the first-quarter work, alongside the Mac health-monitor stacks, rather than as a lab benchmark run on a single high-end phone.

How Much Faster Was the 2023 New Teams App?

Microsoft said new Teams delivered up to two times faster app performance and used 50% less memory, and that it cut meeting join time by more than 70%. For anonymous browser join in some third-party meetings, known as Direct Guest Join, the bundle dropped from 120MB to 10MB. Those figures come from the March 26, 2024, Teams blog post, not from the 2026 hang and chat-switch update.

How Does Microsoft Try to Stop Teams From Slowing Down Again?

After the 2025 chat and channels work, pull requests run through performance gates that replay core scenarios and block merges if latency, responsiveness, or memory gets worse than a baseline. New builds also go through a 50/50 experiment before general availability, so core metrics in the treatment group can be compared with a control group before a feature is on for everyone.