Cloudflare, Inc. NET

Key data

The business

Cloudflare runs a network spread across hundreds of cities that sits between the open internet and its customers' websites, apps and interfaces. The original job, still most of the money, is security and speed: soak up attacks, block bad bots, and store copies of content near users so pages load fast. A customer points their domain at Cloudflare and their traffic flows through its edge, which is exactly why they stay, ripping it out means re-plumbing how the site reaches the world. The moat is that flywheel: every new customer and every attack the network sees makes it smarter and cheaper to run per unit, and the routing becomes load-bearing. The newer story is selling developers compute (Workers), storage, and now AI inference and agent tooling that run on the same footprint, close to end users. That is the part the price is reaching for.

The numbers

Revenue climbs every quarter, and the company beats the Street's adjusted profit estimate every quarter, most recently $0.29 against $0.27 expected.

The GAAP loss ballooned to $170.0M in Q2 2026 on a restructuring charge from the May 7 plan that cut about 20% of staff for an "AI-first operating model," which is the plain answer to the charge a look at this name two days ago could see in the numbers but could not name.

Revenue compounded about 35% a year from $656M in 2021 to $2.17B in 2025, and reaccelerated to 36% last quarter rather than fading. Operating cash flow went from $65M to $603M over that span, and free cash flow was a real $287M in 2025, about 13% of sales. The catch sits in the next table.

Current deferred revenue rose about 19% in six months and committed backlog is $2.73B, so demand is still landing ahead of the revenue line. But stock-based pay ran 20% of revenue, so the cash the owner actually keeps is far below the cash the statement shows. The market is pricing Workers and AI inference as a proven second engine already; the print that settles it is whether committed backlog keeps growing faster than revenue as those products land, or slips toward the revenue rate the way a 20% headcount cut usually warns.

Management

Founder-run under Matthew Prince and Michelle Zatlyn on a dual-class structure. Insiders sold $53.4M across 76 transactions over the last year with zero buys: CFO Thomas Seifert sold $16.2M on August 4, Zatlyn about $5.6M in mid-August. Plan status is not disclosed in the feed, so none of it can be confirmed as pre-scheduled 10b5-1 selling. Capital allocation runs on convertible notes, not buybacks: fresh 2031 converts priced in August at roughly a $497 strike, a 60% premium, stacked on the 2030 notes, leaving about $3.3B of convertibles against $1.66B of cash. The RIF is the first hard signal that this team will now defend margins, not just spend for growth.

How it fails or surprises you

Pricing meets AI. Cloudflare grows by expanding existing customers. If AI agents thin out human seats, or hyperscalers bundle equivalent security and CDN for free, that expansion stalls. The tell is committed backlog growth falling toward the revenue growth rate, the point where the land-and-expand engine has quietly stopped expanding.

The convert wall. $3.3B of convertible notes, $1.29B of it current, sits against $1.66B cash and steady GAAP losses. If the stock stays well below the ≈$497 conversion strike, those notes must be refinanced or repaid in cash, not shares. A dilutive raise or a cash squeeze is what would prove the "growth already priced in" read wrong.

AI inference at the edge (right tail). Cloudflare's network is physically closest to end users, the natural place to run low-latency AI inference and agents. If Workers AI becomes a default home for that work, revenue reaccelerates past 40% and 42 times sales looks cheap in hindsight. The market is not paying for it because the product and developer platform lines are not disclosed separately yet. First proof would be an AI or developer revenue line disclosed on its own.

Closing thoughts

The question here is whether the developer and AI platform becomes a second engine before the multiple has to answer for it. The distribution is genuinely two-tailed. The left tail is fat, because a 42-times-sales name that decelerates while $3.3B of converts come due gets re-rated hard, and multiple compression on high-multiple software is violent. The right tail is real too, because the network position is a rare call option on where inference physically runs. My read, as judgment not arithmetic: near-term the downside dominates on price and the convert schedule, but the asset underneath is a legitimate franchise, so the loss case is a de-rating, not a broken business.

The bet is still that Cloudflare keeps adding businesses to its network and gets each to spend more, with Workers and AI turning into the second engine. What breaks it is expansion cooling while you own it at 42 times sales. The one pair that tells you first is committed backlog growth against revenue growth: the quarter backlog stops leading revenue is the quarter the story starts cooling, whatever the headline number says.