Third quarter 2026 earnings call, November 4, 2026
A sharp sequential deceleration is built into the Q4 guide
Datadog delivered third quarter revenue of $1.197 billion, up 35% year-over-year and above the high end of its guidance range. The fourth quarter outlook is more sobering. Management guided revenue to $1.19 billion to $1.2 billion, or 25% to 26% growth, which implies essentially flat sequential revenue. Full-year 2026 revenue guidance of $4.515 billion to $4.525 billion implies about 32% growth. The company said that excluding its largest customer, the guide implies Q4 growth in the low to mid-30s percent range, so the headline deceleration is concentrated in one account and in the way the company chooses to guide it.
The CFO, David Obstler, said the Q4 guide "contemplates their revenue at their minimum commitment." The customer renewed in Q3 and, as disclosed in August, cut its usage, but it still spent "substantially above" that minimum. Datadog does not control where the customer lands above the floor. Guiding to the floor builds in a mechanical beat if usage stays elevated, and a miss is not possible if the customer stays at its contractual minimum. Obstler acknowledged that Q3's upside came partly from this dynamic but declined to quantify it. Investors should treat Q4 as a deliberately conservative setup for this account, not as a read on underlying demand. They should also note that the customer's behavior in 2027 is unresolved. Management will give 2027 guidance on its year-end call and said it will continue to guide to contractual minimums.
Early renewals by AI customers are a near-term revenue headwind
The more consequential new disclosure is that several large AI customers renewed far earlier than Datadog expected, trading higher commitments for better pricing. CEO Olivier Pomel said "we saw renewals way earlier than we thought we would," and noted that another occurred in October that the company did not discuss. He was direct about the cost: "these are all good news, but they do create headwind into the short term into the revenue, and that's one change from last quarter."
Management framed the mechanics as ordinary volume pricing. Obstler stressed that these are "renewals at higher levels than their previous commit, which is how they're getting the volume discounts." Pomel added that some accounts are optimized during the process, with customers trimming parts of their usage or being steered to different retention tiers. All of this takes place against a backdrop of customers planning to double, triple or quadruple their footprint. Barclays and Goldman Sachs both probed whether this becomes a recurring annual step-down in sequential growth. Management did not commit to a pattern. Pomel said the dynamic is new because some customers are scaling "to a level that we've never seen in the industry or any industry really before," and argued that as the AI cohort broadens, revenue will become more stable. He characterized the variance as "a few percentage points quarter-to-quarter" relative to expectations.
Obstler also reminded investors that Q4 carries a seasonal drag from late-December usage lulls and code freezes, which compounds the early-renewal effect and the largest-customer treatment. The sequential slowdown therefore has three overlapping causes, and only one of them is within the company's visibility.
Non-AI growth keeps accelerating, and new logos are carrying more of the load
The healthier signal is in the base business. Revenue growth from non-AI customers accelerated for the fifth straight quarter to the high 20s percent, up roughly 1 point sequentially and up from about 20% a year ago. Excluding the largest customer, total revenue growth also accelerated. Growth was described as broad across segments, spending bands and industries.
New customers accounted for about 30% of year-over-year revenue growth, up from 25% a year ago. Net revenue retention was in the low 120s percent, flat sequentially, and gross retention held in the mid- to high 90s. Customer count reached about 33,500, and customers with ARR above $100,000 rose to about 4,860 from about 4,060, representing roughly 91% of ARR. RPO grew 38% to $3.85 billion, and current RPO grew in the mid-30s percent, although management reiterated that revenue is the better indicator.
Pomel declined to forecast whether non-AI growth will keep rising, saying "it's a very high growth rate already... it's hard to tell where it's going to go," while expressing confidence in the underlying trend. He pointed to a secondary driver. More AI-generated code is increasing pressure on CI, deployment and testing, which is pulling Datadog's CI products and other offerings beyond core observability into customer deployments. The company also cited new large-customer lands that have not yet ramped, including a 5-year, 7-figure annualized deal with a U.S. federal defense agency, built on its FedRAMP High certification, and 7-figure deals with two of Japan's largest banks. Pomel noted these "don't even register into the revenue yet."
Profitability stepped up, with Q4 margin guided lower on marketing spend
Non-GAAP operating margin reached 28% in Q3 against 23% in both the prior quarter and the year-ago period, as OpEx growth slowed to 25%. Gross margin was 81.2%, up from 79.6% in Q2, which management characterized as stability around 80%. Free cash flow was $251 million, a 21% margin, and the trailing 12-month margin was 26%. The company ended with $5.2 billion in cash and securities.
Q4 operating margin is guided to about 24%, with operating income of $280 million to $290 million, reflecting heavy marketing investment including AWS re:Invent. Full-year operating income guidance of $1.095 billion to $1.105 billion also implies a 24% margin. Full-year non-GAAP EPS is guided to $2.73 to $2.75. Management said it continues to invest aggressively in go-to-market, citing uncovered geographic and end-market white space such as the federal sector, and in R&D across AI and headcount.
Infrastructure Monitoring passes $2 billion in ARR, though GPUs are not yet a driver
Infrastructure Monitoring now exceeds $2 billion in ARR, growing in the mid-30s percent. Management said the company is the observability market share leader but holds a share still "in the teens." In response to a Bank of America question about the AI contribution, Pomel said the growth is not directly tied to GPUs. GPU monitoring is used mostly for training workloads and "we don't monetize those products widely yet." The number of GPU devices monitored is up 8x this year, so the revenue opportunity from GPU monitoring is upside to current numbers.
Scout and Bits AI show Datadog moving into proprietary models
Datadog introduced Scout, its own reasoning model optimized for site reliability tasks and trained with in-house distributed reinforcement learning infrastructure. Pomel said it performs "on par or even better than frontier models on alert investigations at a fraction of the inference cost," and it now powers Bits Investigations. The lower cost matters commercially, since it lets Bits investigate more alerts, start earlier and go deeper. The Adaptive ML team, which joined in June, is accelerating work on reinforcement learning and post-training.
On model strategy, Pomel described a deliberate mix of top-lab models for tasks only they can handle, open-weight models for cost-efficient commodity tasks, and proprietary post-trained models where Datadog's own data creates an advantage. He expects that mix to stay roughly stable. He was cautious on the broader training opportunity, noting that a year or two ago "basically nobody was doing training" outside a few frontier labs, and that while the number of large-scale trainers has grown by an order of magnitude, "it's still early to call it a mass market."
Bits AI is described as the fastest-adopted product in the company's history, and Bits Testing became generally available in October. Datadog moved from charging specifically for investigations to a model of AI credits spanning more use cases. Pomel said pricing stays "fairly close to inference costs" so that efficiency gains can be passed to customers, because volume is expected to rise by orders of magnitude as unit prices fall by orders of magnitude. That implies Bits is likely to be a thin-margin product, and its real payoff is in platform usage. Pomel said he doesn't much care which route becomes the dominant monetization, "as long as we get more usage." No revenue contribution was disclosed.
AI adoption metrics are broadening, and distribution through the labs is new
Datadog now has over 850 AI customers, up from about 750 last quarter, including 39 spending more than $1 million annually, and all of the top 15 AI leaders. About one third of customers use at least one AI product, including three quarters of customers above $100,000. Agent Observability spans are up 9x this year, MCP tool calls are up 50x, and more than one third of Cloud Cost Management customers now track AI costs, with the number of customers doing so up more than 11x. AI providers have become the largest category of spend that customers monitor in the tool outside the three major clouds. Pomel said non-AI-native customers are increasingly consuming Agent Observability, which he described as "crossing the chasm," although no hard numbers were given.
Datadog also made Agent Observability available in the Anthropic marketplace and is a launch partner for OpenAI's new enterprise marketplace, which lets customers purchase directly through the frontier labs they already use. Deal examples reinforced the AI exposure, including an 8-figure expansion with the AI division of a large technology company that nearly tripled its commitment just two quarters after landing, and an expansion for an AI coding company already at an 8-figure annualized level that renewed early.
Platform breadth continues to widen. 61% of customers use four or more products, 41% use six or more, and 16% use ten or more, versus 8% a year ago.
The quarter's quality is not in question, as non-AI growth and margins both improved. The debate now is how much volatility comes with a customer base in which a handful of hypergrowth AI accounts renegotiate pricing early and one dominant account sits above its floor. Management's answer is that this is a usage-based model with better mid-term than short-term visibility, which is accurate but offers little help in modeling the next two quarters.