Cloudflare (NET) Q2 2026: Cost of Revenue Grew 53% Against 36% Revenue Growth
Cloudflare reported June-quarter revenue of $696.1 million, up 35.9% year over year and $31 million above the ceiling of its own guidance. Growth accelerated from 34% in the March quarter. Non-GAAP EPS came in at $0.29 against $0.21. Current remaining performance obligations grew 35%, the third consecutive quarter of acceleration in that metric. Full-year revenue guidance moved to $2.864–2.870 billion from $2.805–2.813 billion. The stock closed the regular session down 3% at $284.17 and traded up roughly 16% after hours to the $329 area, an all-time high and well through the prior $305 peak.
The number that matters sits one line below the headline. Cost of revenue was $196.5 million against $128.7 million a year ago, an increase of 52.7%. That is seventeen points faster than revenue. Non-GAAP gross margin fell to 73.1% from 76.3%; GAAP gross margin fell to 71.8% from 74.9%. This is not a single-quarter artifact. Non-GAAP gross margin was 78.8% in the third quarter of 2024, 76.3% in the second quarter of 2025, 75.3% in the third quarter of 2025, and 73.0% across the first half of 2026. The business has shed roughly 320 basis points of gross margin per year, steadily, through exactly the period in which AI became the growth narrative.
That trend is the actual disclosure in this print. Caching, DDoS mitigation, and DNS are close to costless at the margin once the network exists; they are the products that produced 79% gross margins. Inference at the edge, R2 storage and egress, Workers execution, and the compute behind the agent and crawler-control products are not. Cloudflare is converting from a company that rents access to a network it already built into a company that sells cycles on it. Cycle businesses do not carry 79% gross margins. They carry something closer to 60%. Nobody on the call has said where the floor is, and the floor is the single most important unknown in the model.
The restructuring makes the point sharper. Cloudflare took a $150.7 million charge in the quarter, at the top of the $140–150 million range guided in May, against the elimination of roughly 1,100 roles — about 20% of a 5,500-person workforce. Operating expenses excluding that charge grew 23.0% against revenue growth of 35.9%, thirteen points of genuine operating leverage. All of it was consumed. Non-GAAP operating margin was 13.8%, down from 14.1% a year ago. The company removed a fifth of its people and its operating margin still went backwards year over year. The savings did not reach shareholders; they went into the cost of serving the incremental dollar of revenue. Matthew Prince framed the cuts in May as a structural reorganization for the agentic era rather than a cost exercise, and the numbers support him more literally than he probably intended. This was not a cost reduction. It was a reallocation from headcount to network.
Free cash flow of $56.4 million, or 8% of revenue against 6% a year ago, deserves less credit than it is receiving. First-half purchases of property and equipment fell 21% in absolute dollars, to $115.2 million from $145.8 million, while revenue grew 35%. Capex dropped from 14.7% of revenue to 8.6%. Depreciation and amortization of $123.2 million now runs above the hardware line it is depreciating. A network company running its installed base harder while workloads shift toward compute is borrowing free cash flow from a future quarter. Either capex normalizes and the free cash flow margin compresses, or the AI workloads are being served on leased and third-party capacity, which is the same gross margin story from the other direction. It is worth noting separately that $37.8 million of non-operating income — interest on a $4.16 billion securities balance — accounts for 35% of the $107.8 million non-GAAP net income. The operating business earned less of the headline profit than the balance sheet did.
The moat is real and is not what the sell-side describes. It is not the CDN, where Akamai and Fastly compete on price and CloudFront competes on bundling. It is position. Cloudflare sits in front of a large fraction of the web with the ability to identify, classify, price, and block machine traffic at the request level, and it holds that position because publishers installed it for reasons that had nothing to do with AI. The crawler-control products, the pay-per-crawl mechanics, the agent identity and wallet products announced days before this print, and the answer-engine visibility tooling all monetize a chokepoint the company acquired for free over fifteen years of giving away DDoS protection. The durable part of the argument is neutrality: Google cannot credibly meter Google’s own crawler on behalf of publishers, and neither can Microsoft or Amazon. That is a structural exclusion, and it is why the toll-booth thesis is not obviously replicable by a hyperscaler with more capital. The vulnerability is the inverse — the chokepoint is only worth what the AI labs will pay to pass through it, and the labs are the counterparties with the most incentive and the most capital to route around it.
Valuation now assumes the toll booth works. At $329 the shares carry roughly $117 billion of market capitalization against $892 million of net cash after $3.27 billion of convertibles, for an enterprise value near $116 billion. That is 40.5 times the midpoint of full-year revenue guidance and roughly 39 times forward ARR on the third-quarter run rate. The stock has cleared essentially the entire published target range in a single session: BofA and Citizens at $330, Morgan Stanley at $322, Mizuho at $310, KeyBanc at $300, RBC and Jefferies at $290. Consensus was near $252 three weeks ago. Every one of those numbers has to move before the shares can advance on anything other than momentum.
Base case is $300 to $360 over twelve months, on 32–35% revenue growth, gross margin settling in the low seventies, and non-GAAP operating margin holding near 14%; that is a Rule-of-50 profile and it justifies the current multiple without expanding it. Bull case is $420 to $480, and requires the crawler and agent-payment products to become a disclosed revenue line rather than a narrative — a per-request take rate on machine traffic would reprice this as a transaction network rather than infrastructure software, and the catalyst is the fourth-quarter print or the 2027 guide. Bear case is $190 to $220 and has nothing to do with Cloudflare execution. It is a cohort derating: the high-multiple AI infrastructure complex trading back to 20–25 times sales on a hawkish Fed or a hyperscaler capex pause takes forty percent out of this name with revenue growing 35% the whole way down. At forty times sales, the multiple is the position.
Cloudflare guided the September quarter to $736 million. Against $562.0 million a year ago, that is 31.1% growth — five points below the quarter that just moved the stock 16%. The implied fourth quarter is slower still, at 29%. The company beat its own second-quarter ceiling by 4.7%. Apply the identical beat to the third-quarter guide and growth lands at 37%. The entire case for owning this at forty times sales is that the sandbag stays exactly the size it was last quarter.