The full review will be published on Saturday.

Coupang reiterated 20% Y/Y foreign exchange neutral (FXN) growth for the full year. It has not seen consumer weakness stemming from macro volatility. It also reiterated $700M in EBITDA losses for its developing offerings segment.

Coupang trades for 74x forward EPS. EPS is expected to grow by 47% this year and by 117% next year. The firm is currently inflecting to profitability, so the sales multiple chart is the best option.

There’s a lot going on within this product suite and I think understanding the basics is important. This recurring section will be review for some. If it’s not for you, let’s learn:

This is a dominant player in the data observability space. Observability simply refers to the practice of monitoring an entire asset ecosystem to track issues, vulnerabilities and performance. Knowledge is power, and so this organized surveillance has a way of expediting resolutions to challenges. Other players within this area include the hyper-scalers, Splunk, Elastic, CrowdStrike (through M&A) and many more. Datadog splits its observability niche into 3 smaller buckets: infrastructure monitoring, log management and Application Performance Monitoring (APM).

Infrastructure monitoring: provides a holistic view of assets like servers and networks. It automates the collection of traffic and overall usage insights. That means it can more expediently fix and uncover infrastructure problems. This can also help clients and their other vendors uncover where compute capacity is being suboptimally distributed. Fixing those inefficiencies cuts costs and will only get more popular in a world obsessed with controlling exploding GenAI usage/costs.

Log management: collects and manages logs or “timestamped records of events.” This also facilitates faster issue remediation and optimization of performance. This product routinely supports infrastructure monitoring, BUT there’s a key difference between the two. Log management handles event-based data like customer service interactions, while infrastructure monitoring (as the name indicates) handles infrastructure-based metrics.

Application Performance Monitoring (APM): tracks app performance and uncovers/prioritizes performance issues to be remediated.

There’s also a newer, related form of Datadog monitoring called Digital Experience Monitoring. It’s exactly what it sounds like. This product includes real-time user monitoring (RUM) to track precise, observed interactions, and also Datadog Synthetics, which is similar to RUM, but tracks a simulation of expected interactions – thus enriching the overall picture. Datadog delivers detailed churn analysis, engagement metrics and more from these tools. It also provides mobile app and feature testing, as well as actionable user journey visualization reports.

These four product categories, which frequently work together, form its “unified platform.” Other products to know within this overarching offering include Flex Logs (part of log management). The product broadly rolled out towards the end of 2024. These offer a cost-effective means to store and retain large batches of logs by separating storage and query usage. In turn, that separation makes it ideal for long-term data storage and regulatory compliance. Separation also unleashes more data scalability, query customization and cost optimization. Conversely, querying from a flex log is slower than standard logs. That makes Flex Logs better suited for lower priority data and where latency is not a crippling bottleneck.

Because Datadog already handles network viability, security is a wonderfully relevant growth adjacency. A product like Cloud Infrastructure Entitlement Management (CIEM) for example, fortifies strict, minimum access identity controls. There’s a lot of competition with configuration-based cloud tools like this one, but Datadog is no slouch. CIEM diminishes risk of identity attacks in a cloud environment. Its Security Information and Event Management (SIEM) product allows for “long term data log visualization for security investigations.” It’s helpful for broad threat management use cases. This can be done without dedicated staff, making cloud migration and usage easier. Most recently, it added agentless environment scanning (no security agent installation needed) to match with its agent-based product. 

Datadog offers data security and code security tools to detect and prioritize source code blunders. Development, security and operations (DevSecOps) covers the end-to-end, source code-to-package deployment and runtime lifecycle. Datadog has been a large player in the Ops section (and increasingly Sec too). This is it moving into the Dev section more meaningfully, or moving “further left” towards developers.

It also offers a host of products within Cloud Service management. For example, its Kubernetes Autoscaling tool handles resource usage and expansion optimization. It pulls from extensive traffic data to tell customers where they can save on compute capacity and other areas. This is part of its cloud service management push.”

But… this intro would not be complete without its GenAI product work. Toto is the name of its first foundational large language model (FLLM) and Bits AI is its copilot. So far, this can summarize incidents and conversationally field questions. It’s also rolling out autonomous investigations to remove the manual work from uncovering issues with infrastructure, large language models, apps, usage patterns etc.

Much more is coming. And unsurprisingly, it also tweaked and configured its core products to cater to LLM observability.

Usage growth, which powers a significant part of this financial engine, was called similar to Q1. It accelerated for its AI customers and was stable almost everywhere else. They did chat about large enterprises slowing usage growth just a bit Q/Q, but that was called quarterly volatility and the team isn’t concerned about it. Something to track but not to worry about as long as it doesn’t last more than a quarter.

“Year-over-year usage growth with our enterprise customers remains healthy but maybe a bit lower than last quarter. We see that as a product of the volatility that can occur among customers from quarter-to-quarter.”

CFO David Obstler

While more hiring and R&D were expected, GPM was materially worse than the street wanted. This came from a larger-than-expected cloud hosting cost surprise. The heftier expense was related to “large growth spikes from some of its largest customers.” Perhaps that’s where the revenue beat came from, but it was quite strange to juxtapose this with other comments above on large customer usage growth slowing Q/Q. They also said accelerated product innovation and hiring efforts hurt GPM, so perhaps those were the main culprits. It’s focused on clawing back some of this margin via cloud efficiency gains throughout 2025.

“Our view of our long term market opportunities remains unchanged. We continue to believe digital transformation and cloud migration are long term secular growth drivers of our business, as well as critical for every company to deliver value and gain competitive advantage.”

CEO Oliver Pomel

Cross-selling & a budding Platform Play:

Datadog continues to successfully cross-sell more products to juice retention rates, lifetime value, top-line growth and margins. That can be seen in its positive 8+ module customer momentum, which I don’t think should be overlooked. I think it’s somewhat easier for security-first companies to expand into observability, as they’re already handling all of the data that goes into proper monitoring. Branching out from observability to security is harder, as it requires less adjacent and less overlapping capabilities to do well. Positive module momentum is partially a byproduct of successful security cross-selling, which I think bodes very well for this long-term growth engine.

Simply put, cross-selling for enterprise software platforms is what separates the generational compounders from the dime-per-dozen point solutions. This quarter, Datadog cited success in cross-selling two key products. Flex logs crossed $50M in ARR just 6 quarters after launching, which is a new company record. And that $50M is coming with compelling margin; that was perceived as a bit of a risk, considering the highly attractive consumer pricing of the offering. It’s also driving usage growth for all of its other products as an emerging secret weapon for platform-level adoption. 

The other area called out was its database monitoring suite. This is approaching $50M in ARR with 60% Y/Y growth. To build on this momentum, it is purchasing Metaplane. Per Datadog, this is an “end-to-end data observability company.” It features extensive, automated help with prevention and resolution of data vulnerabilities, quality control and improper permissions.

“[Metaplane] helps make sure the data is being extracted from the right place, transformed the right way and is being fed into the right AI models on the other hand.”

CEO Oliver Pomel

“We've seen for several years now that better freshness and quality were critical for applications and business analytics, and we believe that they are now becoming key enablers of the creation of new enterprise AI workloads.”

CEO Oliver Pomel

Let’s put some numbers behind these ideas. Datadog signed eleven $10M+ contracts during the quarter vs. just one Y/Y. Some of these wins included:

Observability & AI Product Traction:

AI is moving the financial needle for DDOG. 4,000 customers are now using an AI integration, representing 100% Y/Y growth. AI native customers are now 8.5% of ARR vs. 3.5% Y/Y and this cohort added a full 6 points to Y/Y growth rates vs. just 2 points Y/Y. One customer is having an outsized impact on this, but growth for all other clients in this group is strong.

While the app explosion stemming from AI, which is where DDOG should shine, hasn’t really happened yet, it’s enjoying great success with retrofitting existing products for AI assets. LLM observability is the main example today, as customers for this product doubled in two quarters. It’s also gearing up to add more use cases in workload automation and app-building within Bits AI. Even if the app layer of the opportunity is still not ready to boom, it’s vital that DDOG has these products in place whenever that happens.

As leadership told us last quarter, it still expects usage and cost optimization for its GenAI customers as they rapidly grow consumption first and figure out optimizations later. That always happens when customers rapidly grow usage over a short period of time and it’s baked into DDOG guidance.

Security:

A lot of the security products mentioned in this report were included in the 101 section of this article. For some updates, its code security tool, for both first and third-party assets, has 1,000 customers just months after launching. AI continues to automate source code creation, which means a lot more projects and packages for DDOG to monitor and protect. And? Code writing explosions inherently coincide with data usage explosions. Along those lines, the sensitive data scanner is also building great traction, as it allows customers to better understand the hygiene and sensitivity of their overarching ecosystem.

More M&A:

Datadog is acquiring Eppo for its “feature flagging and experimentation platform.” We’ve seen how incredibly impactful Palantir’s digital split-testing engine has been for its own AI products. We know how hard Nvidia and AMD are working on enabling ontology (digital twin split testing) at scale. This should help Datadog get more capable in that regard. Eppo should augment the scale that Datadog customers can tinker with and safely deploy software.

“We see automated experimentation as a key part of modern application development with the rapid adoption of AI generative code, as well as more and more of the application logic itself being implemented with non-deterministic AI models.”

CEO Oliver Pomel

Global Expansion:

Datadog is building a new data center in Australia. They see that as a great market to prioritize international expansion.

Macro:

The team has experienced zero sales cycle elongation or macro headwinds from the recent trade wars. Stability has been the case for every high-quality enterprise sales firm like this one.

Average quarter. I fully understand why people didn’t love the negative gross margin surprise. I really don’t mind hiring more and spending more on R&D, as that raises the long-term value creation ceiling… but I don’t want to see that gross margin hiccup recur. It probably will be a blip on the radar… the team is excellent and the products are valuable. But? At its lofty valuation heading into the report, things like this aren’t easily forgiven and that’s likely why many other positive data points from the print were shrugged off. That’s important for the active trader and less so for the long-term investor like most of us are.

I don’t think the investment case soured from this one small miscalculation. Datadog continues to fortify its platform status across cloud security and observability. It’s creating a GenAI-inspired lift to financials that most companies in its world aren’t matching. It continues to find steady growth and enough margin help elsewhere to still maintain its full-year EPS guide (despite the EBIT weakness). This is a rock-solid company that should be a rock-solid company for a very long time. While this wasn’t their best showing, it was still a report that most other companies can only dream of delivering. And it was still fine.

Needed Industry Definitions:

GPU: Graphics Processing Unit. This is an electronic circuit used to process information and data. The accelerated compute needed for GenAI apps and models pulls from next-gen GPUs. It thinks its “MI” series of GPUs (part of the “Instinct” product family) boasts best-in-class memory and bandwidth, which Nvidia would certainly disagree with. AMD also thinks its 2025 Instinct release will compete with Nvidia’s world-class Blackwell platform.

CPU: Central Processing Unit. This is a different type of electronic circuit that carries out assignments and data processing. CPUs fall in the general compute bucket. General compute CPUs are still optimal for static, step-series and instruction-based tasks. They’re also much cheaper than deploying next-gen GPUs when they can work for the specific use case. AMD’s new AI data center CPUs “extend leadership in performance per watt and dollar.” I

NPU: Neural Processing Unit: Used for AI-enabled personal computers (PCs).

TOPs: Tera Operations Per Second. This measures NPU performance, with more TOPs being better. TOPs superiority is imperative for running Copilots and GenAI apps on PCs with optimal latency, hallucination rates and performance.

Q2 revenue guidance beat estimates by 2.2%. This includes a $700M revenue hit from new Chinese export restrictions. For the full year, that impact will be $1.5B, but it did reiterate expectations for 10%+ annual revenue growth despite the new headwind. Analysts currently expect 21% Y/Y growth for the year.

The China restrictions will lower gross margin by a full 11 points in Q2. 54% non-GAAP GPM guidance, which excludes this impact, beat by 2 points. EBIT guidance beat estimates by 3.9%. $0.90 EPS guidance also beat estimates by $0.02. 

“Despite the dynamic macro and regulatory environment, our first quarter results and second quarter outlook highlight the strength of our differentiated product portfolio and consistent execution positioning us well for strong growth in 2025.”

CEO Lisa Su

AMD is expected to compound earnings at a 34% clip for the next two years. At 22x EPS, that’s a highly compelling growth multiple. But? That growth multiple is only accurate if estimates stop falling. This quarter should help stabilize estimate trends.

Data Center – General Compute & CPUs:

While everyone loves to focus on GPU demand (for very good reason), that does not mean CPU demand will vanish. There are still plenty of use cases that simply require static, instruction-based, step-by-step compute. These use cases can save a lot of money by using a CPU over a GPU. 

During the quarter, AMD again took CPU market share as its 5th generation EPYC processors (“Turin”) are enjoying fantastic demand. 30 new 4th and 5th generation EPYC cloud instances launched during the quarter, and it expects that pace to pick up throughout 2025 as Turin ramps. This quarter, Amazon launched new EPYC-powered instances with Field-Programmable Gate Arrays (FPGAs), which are similarly malleable to custom Application-Specific Integrated Circuit (ASICs), but with post-manufacturing malleability. It plans to use these for “compute intensive workloads” within areas like network security, streaming and genomics. Generally speaking, Forbes 2,000 EPYC cloud instances are up 2x Y/Y, with key wins from world-class firms such as CrowdStrike; CrowdStrike is enjoying significant performance gains with EPYC. It has 10/10 of the largest telecom, aerospace and semiconductor component companies as EPYC clients as it continues to absolutely dominate the CPU space.

“Every major cloud provider is deep in development on Turin programs with a steady stream of public instances and internal deployments expected to ramp into production over the coming quarters.”

CEO Lisa Su

“Looking forward, we see a clear path to continued share gains as customers ramp their fifth-gen EPYC offerings that deliver unmatched performance, efficiency and TCO across every major cloud and enterprise data center workload.”

CEO Lisa Su

Data Center – Accelerated Compute & GPUs:

This is where the GenAI infrastructure boom shows up in AMD’s results and where Nvidia is currently enjoying virtually all of its success. AMD is hard at work on closing the performance gap with NVDA, with a move to an annual platform release cadence to match Nvidia’s innovation rhythm. It’s hard to catch the leader when they’re also racing, but AMD has the best chance out of anyone else. M1325X shipments are ramping on schedule, as virtually all hyperscalers are moving to offer these accelerators in more cloud instances. It also landed a deal with “one of the largest frontier model developers” for this version of its Instinct GPUs. Model cost deflation is merely amplifying this momentum by making AMD’s products more accessible for the masses.

MI350 will come next and is already sampling for some customers. It comes with a new manufacturing architecture (called “cDNA”), which will boost memory and bandwidth by 50% vs. MI325X and improve throughput and performance by 35 times vs. MI300X. MI350 customer interest was called “very strong” and they plan to begin scaled shipments late this year. MI400 will come after that.

AI Software:

Slick and powerfully native software offerings on Nvidia’s platform are an extremely durable part of their moat. One can argue they create unnecessary vendor lock there, but who can blame them? If they have the best GPUs, they have every right to create the stickiest relationships they can. They’d be stupid not to.

AMD needs to overcome this and they’re working hard to do that. It now releases updates for its software platform (called ROCM) every 2 weeks instead of every 3 months. It’s also passionately working on extending its roster of open-source partnerships and integrations to create a more collaborative and interoperable software/hardware environment. AMD probably can’t win all by itself… but it can potentially win with everyone else at its side. As part of this, it’s enabling 3rd-party developers to ship ROCM updates every night and offering 2,000,000 models that can now “run out of the box on AMD.” This includes day zero access to Llama 4, Gemma 3 and DeepSeek R1. Out of the box is important. Nvidia is elite at handling end-to-end AI infrastructure usage and application deployments. This is AMD trying to get there. It continues to use software to drive unique model and hardware performance gains – thus extending its value proposition. For example, it used ROCM to optimize DeepSeek’s R1 model and deliver best-in-class model performance with its own chips.

ZT Systems M&A:

The ZT Systems deal has closed. AMD is excited to combine “world-class system design expertise” with its “chip and ROCM leadership.” It will unlock the ability for AMD to offer “ready to deploy AI solutions” and, in doing so, will “reduce deployment time for hyperscalers.” AMD sees significant interest in the manufacturing arm of ZT that it plans to sell.

“The ZT Systems team is fully engaged and already co-designing with key customers on rack-level designs. These are optimized for our upcoming MI400 series. It’s also working with customers and partners to accelerate time to market for our MI350 series.”

CEO Lisa Su

Brief Notes on Non-Data Center Revenue:

More Notes:

Good quarter. The company is executing and delivering solid growth in highly exciting areas. The issue is that its direct competitor (Nvidia) is delivering historic levels of growth and then some in the same field. That’s leaving many to wonder “why don’t your results look like theirs?” Especially considering the forward multiples for the companies are fewer than 3 turns apart. Nvidia’s growth multiple is actually a bit lower. Industry pundits who I have access to are noticeably excited about MI350 and MI400 as potential launches to finally close at least part of the gap with Nvidia. And? It doesn’t need to get all the way there. Through networking equipment partnerships and not needing a 70% GPM, it can approach Nvidia’s performance and still enjoy great demand at lower price points. That needs to happen and leadership is quite confident that it will. The GenAI chip boom will not last forever. Demand runways will be exhausted and growth will slow. The clock is ticking for AMD to approach the type of product parity needed to capture this massive revenue opportunity while it lasts.

Lisa Su is a great CEO and AMD’s work in the CPU space is nothing short of admirable. They supplanted Intel. Whether or not they can match that success in the more exciting GPU space remains to be seen. It’s possible that Nvidia will just not be caught. But? AMD is more capable of figuring this out than arguably anyone else.