管理層發言
Good afternoon, everyone, and thank you for joining us today. We delivered a solid second quarter. Consolidated revenue of $108.2 million, above the high end of our guidance range and ahead of Wall Street expectations, with non-GAAP earnings of $14.7 million, which also is beating consensus. As you may recall from my last quarter commentary, there were three areas I highlighted. First, improving revenue trends, especially with key accounts in the areas of technology and financial services. Second, our AI adoption and growth. Third, improving profitability trends. I'm happy to report that on all three fronts, our execution is solid and we're seeing the benefits. Growing top accounts relationships, continued AI momentum with expanded capabilities in robotic and physical AI, and solid progress toward our 300 basis point margin expansion commitment. For the second consecutive quarter, our top accounts are in technology and financial services. Technology and financial services now define our more strategic customer relationships and those are precisely the sectors where AI adoption is moving fastest and where our capabilities are the most differentiated. Our top accounts continue to drive our growth. Several delivered double-digit quarter-over-quarter growth with standout performances. These are not incremental gains. They reflect expanding programs, deeper program adoption, and the compounding effect of our capabilities that keep finding opportunity inside each client's organization. Several of these clients are now embedding our GAIN platform as core infrastructure in their own operations, not just a project tool, but as a sustained capability. This is a fundamentally different and more durable commercial relationship than what we've had two years ago. AI revenue reached 30.7% of the total company revenue in the second quarter, growing 54.6% year-over-year and crossing the 30% threshold for the first time. Two consecutive quarters of year-over-year growth over 50% tells us something important. This is not a spike. It's a sustained shift. The trajectory is clear, and we intend to build on it. Driving this strong performance is a combination of multiple factors. Our GAIN platforms are winning wider enterprise adoption. Our clients continue to transition enterprise AI workloads from pilots to production. Our engineers are more deeply embedded inside client organizations. Bottom line, we're winning entirely new programs that give us confidence in growth ahead. AI-first delivery is now the default, not the aspiration. Fixed price is a preferred approach on new RFP responses. The productivity and margin gains are real. We're executing well and delivering projects successfully. Our focus on executing larger AI platforms is aligned with significant progress we're making in upskilling our engineering talent. By the end of October, we plan to have 90% of our engineers trained on AI SDLC. Our GAIN platforms have expanded LLM partnerships meaningfully this quarter. We're now working with several of the world's leading AI companies, including the top four frontier providers with whom we're under commercial agreements. This approach ensures our GAIN platforms stay aligned with the leading AI platforms with broader reach across our enterprise client base. GAIN remains the backbone through which we bring AI capabilities to market. Its partner depth makes it stronger every quarter. Our client relationships are evolving, too. Clients who came to us for platform deployments now ask us to stay. They want us to be involved in advisory, execution, and ongoing operations. This meaningful shift is opening a growth vector that did not exist in our model two years ago. On the partnership front, partner influence revenue reached 19.1% of the company total revenue in the second quarter. That was driven primarily by our three core hyperscaler relationships with Google Cloud, AWS, and Microsoft Azure. A growing proportion of that revenue is coming from AI engagements. We are running agentic AI workshops across our Google Vertex AI search customer base, converting search engagement into broader agentic commerce programs. We extended our Google partnership in banking and financial services, closing our first joint win this quarter at a leading global bank. We're deepening our AWS relationship around application modernization and agentic AI in CPG, manufacturing, and financial services. Our NVIDIA partnership is gaining momentum across both agentic AI and physical AI. Our longer-term target remains 25%-30% partner influence revenue, and we're confident of achieving this target. Last quarter, I introduced our physical AI capabilities and our first commercial engagements in the space. Physical AI requires a deep understanding of multiple disciplines that include modeling real-world robotics movements, digital twins, verification in simulators, and integration with hardware systems. Our active programs span humanoid robotics for pharmaceutical intralogistics, autonomous driving stacks for construction equipment, and policy control platforms for manufacturing clients. We signed a strategic partnership with Doosan, a leading robotics manufacturer this quarter, elevating our NVIDIA relationship and opening an engineering office in Dresden, Germany, to support our European manufacturing clients. Grid Dynamics enhanced its robotics offering by welcoming Ekumen, a leading robotics engineering team that joined us in May. Their expertise resides in the Robot Operating System, a foundational open source standard that powers the vast majority of the world's industrial robots. Over the past decade, the company has built an invaluable list of some of the world's most respected robotics customers. Grid Dynamics brings advanced AI modeling, policy control, and enterprise-scale delivery capability. Ekumen brings deep knowledge of the foundational software layer that robot manufacturers depend on. Together, the combination is formidable, spanning the full stack from the foundational software layer through simulation, hardware integration, and enterprise-scale deployment. We believe no other service company in the market today matches this combined footprint and technical depth. Now, let me pass on to Vasily Sizov, Chief Revenue Officer, who will expand on key business aspects of Grid Dynamics' client engagements. Vasily?
Thank you, Leonard. Let me begin with three demand trends we observed during the quarter. First, clients are prioritizing AI investments that deliver clear, measurable business outcomes. Second, as clients move from isolated use cases to enterprise-scale initiatives, they realize that the underlying technology layers must be modernized to support AI adoption. Third, clients increasingly recognize that successful AI transformation requires more than technology alone, driving interest in AI process consulting, performance benchmarking, and change management. These trends align closely with our strategy and the capabilities we are building. Let me discuss each of them in more detail. First, the demand environment remains constructive, with clients directing AI investments toward practical application with tangible business impact. We are seeing particular interest in AI-enabled automation that improves operating efficiency, scalability, and speed. Importantly, these investments are increasingly moving beyond experimentation, with clients deploying AI capabilities into production to automate complex manual processes, improve customer service, reduce operating costs, and create new sources of revenue. Second, as clients move from isolated AI use cases toward enterprise-scale transformation, they are finding that their data, application, and core platforms must be modernized and made AI-ready. As a result, AI adoption is creating broader demand across the underlying technology landscape. This trend aligns closely with our core expertise in data engineering, application modernization, cloud and platform engineering and reinforces the relevance of these capabilities in the era of AI. Third, we are seeing growing demand for AI process consulting, performance benchmarking, and change management as clients focus on converting AI investments into measurable business value. They need to identify the business processes where AI re-engineering can create the greatest value, establish clear performance baselines, redesign those processes, build the technical enablers, and drive enterprise-wide adoption. We have been deliberately strengthening these capabilities to help clients realize measurable value from AI across the enterprise. These trends are reflected in our client work. Let me highlight a few engagements from the quarter that demonstrate how these capabilities are being applied in practice. For a leading food service distribution company, we built and deployed an AI-powered product credit claims platform that automatically validates customer claims against photographic evidence. The platform cross-checks product, manufacturer label, and shipping label images against the claim's reason code in real time, replacing a fully manual salesperson-mediated review process. In performance testing, the system processed approximately 400 claims supported by 1,000 images end to end in under 15 seconds per claim. The capability is now live in production. The client has approved a long-term roadmap to further enhance the system and extend automated decision-making into more advanced credit adjudication scenarios. For a leading home improvement retailer, Grid Dynamics enabled next-day delivery by designing and deploying a high-load service that modernized the retailer's logistics operations. The solution includes an AI-powered routing capability that assigns fragile items to the appropriate vehicle types, eliminating hundreds of delivery errors each week. As a result, the solution cut average delivery time by more than half from three and a half days, and is expected to support up to half a billion dollars in incremental annual revenue for the client. For a global technology company, we modernized large-scale data processing infrastructure, migrating more than 1,000 data pipelines to a serverless execution model. This reduced idle compute capacity, reduced infrastructure costs, and improved scalability. Our proprietary AI-powered automation accelerated the migration and established a reusable delivery approach that is now being applied across broader initiatives at this client. Now let me turn the call to Yury Gryzlov, our Chief Operating Officer.
Thank you, Vasily. Let me build on the physical AI and robotics work Leonard introduced. Physical AI needs a full technology stack, and we operate across everything between the robot and the enterprise. The devices themselves come from our hardware partners. At the foundation is the Robot Operating System, ROS and ROS2, the open source layer the majority of the world's modern robots are built on, connecting the hardware to everything above it. Through Ekumen, we're not just users of it, we are among its maintainers and the founding member of the alliance that governs it. On the top of that sits the intelligence, the AI models that let a robot perceive its surroundings, generate its own motion, and handle real-world variability. We design and validate that in simulation before it ever runs on a real robot. Our own platform, Incarnum, our GAIN platform for Physical AI, is where enterprises bring it all together, building manipulation and inspection workflows, deploying those models, and monitoring robotic lines with digital twins. What unifies it is our focus on the enterprise, expanding this capability to the companies that have robots deployed at scale. Here are a few examples that illustrate our work across the stack. For a leading manufacturer of construction and mining equipment, we are building a next-generation stack for autonomous driving, loading, and excavation. We're helping them design the platform, onboard the first use cases, and add capabilities like policy-based control. What began as our first commercial physical AI engagement is now a multi-year program across several regions. With Ekumen, we've proven two-arm manipulation, grasping, and assembly, trained entirely in simulation and then run reliably on a real robot. Bridging that gap from simulation to the physical robot is one of the hardest problems in the field. Humanoids are the next step. A leading life sciences company is piloting humanoid robots for intralogistics, moving and repacking containers of chemicals, work that was out of reach only a couple of years ago and is now possible thanks to new AI models that generate motion. We provide the platform those robots run on, working with Wandelbots and on NVIDIA's stack. The customer calls it a lighthouse project for their industry, and it's the opening step in a much wider program. We're also building the channels to scale. This quarter, we announced a strategic partnership with Doosan Robotics, a global leader in collaborative robots deployed across 45 countries. It's a full stack collaboration. Our platform, plus the foundational AI components, integration services, and engineering around it, paired with Doosan's cobots and our combined global reach. Together, we can provide what traditional robotic software can't: dual arm assembly, inspection of complex geometry parts, and packing of deformable items. It sits alongside our elevated NVIDIA partnership, and we are in active talks with several more hardware and software vendors. Considering the economics of software services in this space and our positioning, we are confident that we have a material market advantage. Reliable performance in the physical world takes engineers who understand simulation, control, and hardware variability, working through problems that have no templated solution, and so can't be easily automated. This combination is hard to assemble. Ekumen's decade of foundational robotics depth, together with our strength in AI modeling, simulation, and enterprise delivery. We don't believe another services company matches it today. Closing that gap isn't a matter of hiring a team. It's years of hard-won experience, which we are now putting to work for our customers. In summary, robotics and physical AI is a growing market, measured in the trillions over the coming decade. Our expanded capability is helping us capitalize on the early traction we saw last year, reflected in a rapidly growing pipeline from both existing customers and new logos. Another important part of my update is tied to our capital markets focus, where a similar pattern is playing out in software rather than robots. As our banking clients push agentic AI deep into their engineering, the hard part is no longer producing code, it's doing it safely with quality, security, and control they can provide to a regulator. This quarter, that showed up most sharply around security. Banks want the speed of frontier models and AI-generated code without introducing new vulnerabilities. Our answer is spec-driven agentic engineering led by Allium, part of our GAIN platform for AI SDLC, and it's exhibiting real traction across our banking clients. The clearest example is at one of the world's largest banks, where Allium is being used to build new tools as part of a bank-wide initiative to modernize business operations. Working across London, New York, and India, we're bringing specification-driven development to both new and existing systems, starting with tools for AI-assisted productivity and extending to agents that automate operational work. Taken together, physical AI reaching the enterprise and the AI-native engineering scaling inside the world's largest banks, this is the frontier work that keeps Grid Dynamics differentiated. Over to you, Eugene.
Thank you, Yury. Good afternoon. Last year, I described our AI strategy through three horizons. This quarter, I'll describe them by maturity, what has reached scale and what is beginning to scale. Horizon One: scaled — AI-first modernization and the agentic platform. Modernization remains the foundation of our business; AI is changing how the work gets done. Agents can now accelerate work across most of the modernization life cycle, particularly code generation and testing. The remaining work — business acceptance, production scaling, and complex coordination — still depends on human judgment and accountability. An agent can write code. A person still makes a call and stands behind it. We have invested in a set of GAIN tools that support this life cycle. Rosetta governs how agents operate. Allium analyzes legacy systems to create reliable specifications for their replacements. SpecFlow, our latest open source contribution, uses those specifications to support autonomous feature implementation. Rosetta has progressed from its first lighthouse clients to larger engagements across retail, financial services, and manufacturing. At a Fortune 30 U.S. home improvement retailer, approximately 550 of the client's engineers are working with the platform. In one program, seven COBOL services were moved to a modern technology stack with approximately 90% of the code generated by agents. All seven services entered production this quarter. The client already had capable engineers and access to many of the same AI tools we use. What it needed from us was domain knowledge, governance, and control, the capabilities that turn powerful agents into dependable enterprise systems. This productivity is helping us expand client relationships. It is also creating opportunities to use more fixed price and outcome-based commercial models when the scope and accountability are clearly defined. Allium also reached an important milestone this quarter. It is being piloted across five major banks and has begun moving into its first commercial banking engagements. Allium analyzes legacy code to help establish reliable functional specifications for replacement systems. It also supports controlled migration and rollback, reducing the operational risk of moving critical applications onto modern platforms. At one major North American bank, a one-hour GAIN demonstration in February led to a signed contract in April. The bank was managing 150 applications with limited test coverage and a growing security backlog. We translated identified issues into failing tests inside the bank's own tooling, allowing its engineers to independently reproduce and assess each finding. At another Tier 1 bank, this same approach is supporting a security modernization program spanning more than 100,000 systems. This part of the modernization work is co-funded by the client's cloud provider. Our differentiation is not limited to code generation. Our agents can also incorporate context such as security advisories, dependencies, and upstream changes. That broader context helps identify problems that code-only tools can miss and provides the traceability and evidence regulated enterprises expect. We deliberately make selected GAIN platforms open source. The immediate objective is adoption and technical credibility, not software license revenue. Open code allows engineering leaders to evaluate our capabilities directly and strengthens our position when client needs help deploying those capabilities at enterprise scale. The same pattern applies to data. Enterprise AI cannot deliver reliable results without accessible, well-governed data. That is increasing demand for data platform modernization. Our new AI data migration accelerator, released this quarter, is already being deployed in a data lake modernization program for a global consumer products manufacturer. The second scaled component of Horizon One is GAIN Agentic Runtime. Enterprise agents need access to trusted data, evidence that their behavior is controlled, and governance over operating costs. For a global payment client, we brought these capabilities together as shared services, with the retrieval layer now supporting 25 enterprise consumers. We also converted the client's dispute architecture, including fraud, chargebacks, and KYC, to configuration-driven workloads. A common foundation now supports four use cases. By automating much of this configuration, the program rebuilt a decade of business logic in just six months and reduced integration and release cycle times by 96%. At our largest banking client, an internal platform built with our support now centralizes the registration, governance, and operation of AI agents across the organization. The client reports regular adoption by more than 80% of its employees across more than 80 markets. As adoption grows, we are also developing the operational tooling needed to govern and support the platform at that scale. Across these engagements, the pattern is consistent. AI accelerates production, but enterprise value comes from the domain knowledge, governance, and accountability required to put it all into production responsibly. Horizon Two: scaling — harness engineering and physical AI. Horizon Two covers capabilities that are moving from research and internal validation towards repeatable client deployment. The first is agentic harness engineering. Traditional agentic workflows are most effective when the task and sequence of steps are already known. Harnesses are designed for more dynamic work, situations in which an agent must select tools, adjust its approach, and respond to new information while remaining within defined controls. The harness provides those controls. It records what the agent did, tests its output, manages exceptions, and introduces human review where accountability requires it. This allows enterprises to apply agents to more complex work without giving up oversight. During the second quarter, our India engineering center developed nine harness-based solutions. Following our client-zero approach, we are testing them first with our own operations. The objective is to establish evidence of reliability, define the necessary controls, and improve the solutions before introducing them into client environments. The second area is physical AI and robotics. We are investing here because the engineering challenge is fundamentally different from conventional software development. A coding agent can generate software and test it in a digital environment. A physical system must also operate safely and reliably in the real world. It must account for geometry, motion, changing conditions, and the behavior of physical environments. Validation, therefore, has to take place both in simulation and on hardware. A language model alone cannot close this loop. Our research is focused on bringing physics, geometry, simulation, and continuous validation into the agent's operating environment. That is also the strategic rationale for the robotics engineering team we acquired in May. Members of this team have long contributed to core infrastructure in the Robot Operating System ecosystem, with particular expertise in simulation and validation. Their capabilities are now contributing to GAIN for Physical AI, our platform built on Incarnum. During the quarter, we released three new components: tools for composing robotic policies, a continuous improvement loop, and a sandbox environment for control testing. We are beginning to validate the platform through early client and partner deployments. A leading lifecycle company is piloting humanoid robots in its warehouse operations using our platform. Separately, a robotics partner has incorporated the platform into its own offering, creating a distribution channel for our physical AI technology. Horizon Two is not yet the same maturity as our modernization and agentic platform business. Our focus now is to demonstrate repeatability, convert technical validation into production deployments, and establish scalable commercial models. The opportunity is to build differentiated intellectual property in areas where success requires not only generating software but providing how that software behaves in the physical world. Across both horizons, the pattern is clear. The cost of producing software is falling, while the value of governing it, validating it, and taking responsibility for it in production is increasing. This quarter, more components of GAIN moved from tools and pilots into broader enterprise adoption. At the same time, our investments in agentic harnesses and physical AI progressed from research towards controlled client deployments. As these capabilities mature, they allow us to reuse more of our engineering, deploy solutions faster, and take greater responsibility for measurable outcomes. Our advantage is not simply that our agents can generate code. It is that we combine those agents with domain knowledge, operational controls, and the engineering discipline required to make them dependable at enterprise scale. That is where we believe durable value will be created in the agentic era, and where Grid Dynamics is positioned to lead. Anil, over to you.
Thanks, Eugene. Good afternoon, everyone. Second quarter came in at $108.2 million, slightly above the higher end of our guidance range of $106 million to $108 million. That represents 7% year-over-year growth, including de minimis contributions from Ekumen. Non-GAAP EBITDA was $14.7 million or 13.6% of revenues and was closer to the high end of our $14 million to $15 million guidance range. Looking at the performance of our verticals, TMT remained our largest vertical and accounted for 31.8% of total revenues for the quarter, with a growth of 11.7% sequentially and 36.4% on a year-over-year basis. The growth was primarily driven by our largest technology customers. We continue to benefit from vendor consolidation at these customers, which has driven increased wallet share across new and existing programs. Retail contributed 26.5% of total revenues in the second quarter of 2026. The vertical was flat in absolute dollars on a year-over-year basis and grew 3.1% sequentially. The sequential growth was supported by demand from key accounts, including a major specialty retailer. Our finance vertical accounted for 22.9% of total revenues in the quarter and grew 1.2% on a sequential basis. Within this vertical, we witnessed solid demand from our fintech service engagements, including increased contributions from a major payments network, which helped offset the successful completion of engagements with insurance and data analytics and consumer credit reporting clients in North America. Looking ahead to the remainder of 2026, we remain bullish on our growth outlook within this vertical. CPG and manufacturing represented 10.9% of quarterly revenues and grew 2.1% on a sequential basis and 4.2% on a year-over-year basis. Within this vertical, we are witnessing robust demand from a leading wholesale food distributor, along with growth from some of our manufacturing customers. Turning to our remaining verticals, our other vertical contributed 6% of our second quarter revenues, while healthcare and pharma contributed 1.9% of our revenues for the quarter. We ended the second quarter with a total headcount of 4,838, down from 4,964 employees in the first quarter of 2026 and from 5,013 in the second quarter of 2025. We continue to rationalize our overall headcount as well as align our skill sets and geographic mix. At the end of the second quarter of 2026, our total U.S. headcount was 379, or 7.8% of our company's total headcount versus 7.2% in the year-ago quarter. Our non-U.S. headcount, located in Europe, the Americas, and India, was 4,459, or 92.2%. In the second quarter, revenues from our top five and top 10 customers were 43.5% and 61.5% respectively, versus 37.5% and 57.3% in the same period a year ago, respectively. Moving to the income statement, our GAAP gross profit during the quarter was $39.6 million, or 36.6%, compared to $36.2 million, or 34.8%, in the first quarter of 2026 and $34.5 million or 34.1% in the year-ago quarter. On a non-GAAP basis, our gross profit was $40 million, or 36.9%, compared to $36.7 million, or 35.3%, in the first quarter of 2026 and $35.1 million or 34.7% in the year-ago quarter. On a year-over-year basis, the increase in the gross margin percentage was primarily driven by revenue growth outpacing delivery cost. On a sequential basis, the increase in gross margin percentage was due to a combination of working time and improved resource utilization. Non-GAAP EBITDA during the second quarter, which excluded interest income, expenses, provisions for income taxes, depreciation and amortization, stock-based compensation, restructuring, expenses related to geographic reorganization, and transaction and other related costs, was $14.7 million, or 13.6% of revenues, versus $12.5 million or 12% of revenues in the first quarter of 2026 and was up from $12.7 million or 12.6% in the year-ago quarter. The sequential and year-over-year growth in EBITDA was largely due to a combination of higher revenues and strong operating leverage across our non-engineering overhead. Our GAAP net income in the second quarter was $2.9 million, or $0.03 per share, based on a diluted share count of 83 million shares, compared to the first quarter net loss of $1.5 million or a loss of $0.02 per share based on a diluted share count of 84.7 million and net income of $5.3 million or $0.06 per share based on 86.4 million diluted shares in the year-ago quarter. On a non-GAAP basis, in the second quarter, our non-GAAP net income was $9 million or $0.11 per share based on 83 million diluted shares compared to the first quarter non-GAAP net income of $7.5 million or $0.09 per share based on 85.9 million diluted shares, and $8.3 million or $0.10 per share based on 86.4 million diluted shares in the year-ago quarter. On June 30, 2026, our cash and cash equivalents totaled $298.4 million, down from $327.5 million on March 31, 2026. Since our first quarter earnings call, we repurchased approximately 2.6 million shares for a total consideration of $17.3 million. Cumulatively, since our board authorized the 50 million share repurchase program, we have repurchased approximately 4.4 million shares for a total of $30.8 million, reflecting our continued confidence in the long-term value of the business. Coming to the third quarter guidance, we expect revenues to be in the range of $112 million to $114 million. We expect our third quarter non-GAAP EBITDA to be in the range of $16.5 million to $17.5 million. For the third quarter, we expect our basic share count to be in the range of 81 million to 82 million shares and our diluted share count to be in the range of 83 million to 84 million shares. For 2026, we're maintaining our full year revenue outlook of $435 million to $465 million. That concludes my prepared remarks. We are now ready to take questions. Carrie?
分析師問答
Great. Thank you. Congrats on the quarter, Leonard and Anil. There was a lot of detail around AI. Just to step back, Leonard, could you maybe talk about the AI efforts and the implications for both growth and profitability over the next, say, 12 to 24 months? Maybe you can help reassure investors that AI will actually be a net positive for you, because there's still a lot of skeptics out there that think it's going to be a net negative over time.
Thank you, Mayank.
Right. Thank you, Mayank. It's a pretty comprehensive question. If I answer all of the parts, there'll probably be nothing left for the other end. I'll try to be concise in terms of the key elements. Then we can talk a little bit more in detail. First of all, we are reaching many aspects of AI implementations. We talked about it in the past. We're adding those features now. We're talking directly or indirectly about forward-deployed engineers. We make announcements. We train a substantial number of the people in the workforce, and these people are basically driving a new way of implementing our solutions because, as we tend to get more focused on fixed-bid and fixed-budget projects, it helps us to identify not only the execution of the various modernization projects, but also create technology consulting. That's with respect to the people and why it's accretive to us. When it comes to agentic AI, as part of the implementation of the suite of our solutions, we are driving our customers to adopt our GAIN platform model. All the elements of the model are driven by internal tests and developments, but also tailored to our customer needs. They will need to adapt the solution where they see the best fit for themselves, but we also guide them through the process to create the best ROI for that. Before I talk about physical AI, I want to address your point in terms of net positive versus net negative. If you look at the increased growth in just these two areas, that substantially exceeds some of the aged businesses which would eventually drop out. Because the gloom and doom from many facets were about that engineering and consultancy being less relevant. Moreover, people would say it's easier to train FTEs. We embrace FTEs. We embrace our clients. At the same time, as many of the leaders in the industry are saying, we can do more work, we can do more engagements, which we prove with all the listed examples. I'm not going to go through all of them because we have a lot of people who can give you more details on that. As a consolidated effort, as we go today through further discussions, we will demonstrate on specific examples where this accretiveness works. I want to emphasize forward-deployed engineering and agentic AI. The third part, which is also super critical for us and actually drives the adoption and partnership enhancement of our relationships to the next level, is our preparation for physical AI work. We not just made a small acquisition. We not just announced opening additional robotics labs. We've been working with our clients for a long enough time to understand what it means for their own platform, what it means for their application and solutions across industrial, modern machinery, logistics companies, and the industrialization of various new solutions. The material side, the remuneration for physical AI is still to come, but now we have evidence of substantial players looking at Grid Dynamics, again, in a leadership role by expanding our capabilities to the practical world of their usage. This is pretty much a summary, and then, of course, we'll go into more detail. That's very helpful, and sorry if you can't see me. I'm having an issue with my video. I'll try to get that fixed eventually. Just a very quick follow-up, Anil, for you. In terms of the guide, I just want to get a sense of the visibility that you have today versus last quarter. What I mean by that is: is the pipeline now converting faster? Have you seen evidence of that? Does that maybe give you more confidence in the sustainability of growth acceleration once we get beyond fiscal 2026 into fiscal 2027?
You're talking about next year. Let's talk about this year, and then we'll get to next year. As you go into the second half, Mayank, you see, if you look at our visibility and our second half, there are a couple of factors. Number one, remember the 85/10/5? Most of our revenue comes from customers who've been with us for two years and beyond. That formula more or less stays well intact. That you're seeing in the top 5, top 10 customers, right? Because most of the absolute dollar and year-over-year growth is coming there. That stays intact. As you go into the second half, there are three layers. First is working time. The second half is higher than the first half. Second thing is billable headcount. We're seeing new programs kicking in. Maybe without addressing your pipeline question directly, indirectly: yes, we're seeing an increased billable headcount as we go into the second half. The third thing is that we are planning some acquisitions. All these three add up to layers. When you look into 2027, I think I'll let the business guys chime in here, but from my point of view, I see two things that are very interesting. Number one, the relationships that we're having with our technology customers, our financial customers, our top 10 and 20 customers, are going deeper and deeper. Things that we've not done, we're doing. Application modernization programs, which we've not done, we're addressing. The addressable market that we're going after is larger. I overhear these conversations week after week. Which leads me to believe as you go into 2027, if we continue winning at the rate that we're winning, it should play out incrementally past it. I don't know, Vasily or Yury, whether you want to add anything to that.
Yes, let me chime in. I would say that our position with most of our largest clients has been strengthening over the last few years through vendor consolidation. What we see is that we should benefit in the coming years from this consolidation, which means bigger programs would come our way, as customers cut loose the long tail of vendors which are no longer relevant. Given our strong technology positioning in agentic AI, which is a very hot topic for most of our customers, we are well-positioned to benefit from that.
Terrific. Thank you so much. Congrats.
Thank you.
Thank you.
Thank you, Mayank. The next questions come from Bryan Bergin, Janney Securities. Go ahead, Bryan.
Hey, y'all. Good afternoon. Thanks. Maybe just to start, a follow-up on that last question as it relates to the second half; more of a near-term question. As it relates to, you give us 3Q guide, implied 4Q is still a decent ramp. Are you seeing a broadening of momentum in other sectors? You're obviously doing quite well in technology. Are you seeing a broadening of momentum elsewhere that gives you that confidence? As it relates to potentially some M&A requirements, any way you can share with us how you're thinking about maybe the organic contribution remaining versus any needed M&A that you have to go get?
Right, Bryan. Let me point out that as you know, there's a certain seasonality in our business. As we go into Q3 and Q4, that's well established. As I said, there are three levels at which we're operating. Number one is just the working times of the second half of the year, and you guys know it's better. Second thing is that the billable headcount and the trends are positive, and all our prepared commentary should lead you to include that. The third thing is that there is a certain amount of acquisition, and we do have a pipeline. It varies. I always joke: an acquisition is not done till the money is transferred to their bank. We've seen acquisitions that we thought are not going to happen, happen. We've seen acquisitions that were locked and loaded, and we are just not able to close. If you look at that second half, I don't want to comment too much upon Q4 other than saying that we have a seasonal pattern for the year. As we go from the low end of our full-year guide to the high end of the guide, the first component of working time stays intact. The second component of billable headcount, we have variable calculations. The third component perhaps picks up a little bit more is the acquisitions.
Okay. Understood. My follow-up is on margin and a tie-in with the delivery model question. You reiterate the confidence in the 300 basis point expansion, that's good to hear. I'm just curious how much of this margin improvement is coming from structural changes, automation, and efficiencies in the delivery versus traditional cost control cutting measures. I know it's notable you had 7% revenue growth while headcount was down. I know you're saying you're going to add billable headcount, but is there a lasting change in this delivery model? Just maybe talk about that AI-driven efficiency and delivery that you're seeing.
There are three, four parts to this question. Let me take the first part, and then when it comes to some of the AI trends, I'll pass it on. When you look at what we set out to do, we said that on a year-over-year basis, we're going to deliver 300 basis points margin improvement on a Q4 by Q4 basis. Part of that effort is efficiency. It's just the way we're organized. As you know, we've ramped from a handful of countries to 19 countries. We've got many incorporated entities. There's a little bit of efficiency that we brought in, and some of those are one-time, but we operate at a certain level. The second part that we are seeing here is we're embracing a little bit more change in the way we're doing business, whether it's AI, whether it's fixed price, whether it's embracing more tools. That is creating a certain level of, I would say, not yet visible now, but over time you'll see a non-linearity perhaps that is in. The movement that you've seen on the headcount right now was largely driven by efficiency improvements on non-engineering headcount. People should not worry. It's not that we let go of billable headcount. No, it's just non-engineering, non-billable headcount. We cleaned it up. From this point onwards, beyond the 300 basis points that you'll have from Q4 to Q4 as you go into 2027, there is a plan for us to leverage more of these tools. There is a plan of bringing a certain level of non-linearity. We have the plans. The clients have to accept it, and we have to proceed with that. Go ahead, Leonard.
Yeah. Let me share a couple of things. First of all, to complete the answer on the first question about diversification across verticals: I think it's very critical to understand that this is not overnight we suddenly diversified verticals. First and foremost, we've been in payments, we've been in financial services, and we've been in industrial modernization. Second, you can actually see from the previous comments about us, what Vasily said, replacing some incumbent vendors is because we're operating at a higher level. In the past, there were always a couple top vendors and a couple mid-level vendors. Now we only compete with the top vendors. The reason being is, I think AI adoption and technology implementation equalize the field a bit. We've always been prepared for the big tasks and large programs. We also gained a reputation for the consultancy part. As we get more access to the bigger projects, inevitably what happens is better visibility, better projection, and better positioning. We're improving with tools, adding more capabilities, and removing redundancies from the past. We are continuing investment in India and LatAm. As we do more, we create a global platform internally to optimize efficiency. It's a cost structure improvement, it's performance-based, it's tooling, and it's removing redundancies from the past.
Hey, thanks for taking my question. How are your AI and robotics partnerships different from your traditional hyperscaler relationships like with Google, AWS, and Microsoft Azure that generate much of your 19% of partnership revenue? The partnerships you have with NVIDIA and model companies — do they differ or do they offer a different revenue trajectory potential or client ownership structure than your other partnerships?
Thank you so much for the question, Puneet. We definitely value our relationship with NVIDIA and believe that's a great partnership to build a pipeline of future opportunities on. As you understand, the manufacturing industry is going through a massive transformation and new tools like agentic AI and physical AI bring new technology to more traditional manufacturing. We see this as a great opportunity to build a new pipeline of opportunities and a new type of engagement which would help us transform those manufacturers on a bigger scale. For example, we have an active engagement with one of the world's largest industrial equipment manufacturers on building an agentic AI platform which allows them to manage a fleet of autonomous vehicles and deploy physical AI capabilities on edge devices. We see more and more interest in such opportunities. It's definitely one of the top priorities for us to grow.
Got it. I'd like to follow up on the prior question, specifically around headcount. I noticed your non-U.S. headcount was down despite Ekumen, which probably contributed employees in Argentina. The U.S. headcount, by comparison, was up on a sequential basis. Should we expect this remix to continue as you do more AI-based services? Will that require more on-site headcount or U.S. headcount compared to in the past? If that's true, what does that mean for margin and change management within your employee base?
Very good. Puneet, what you said is music to Eugene's ear because he's been the one who is architecting the acceleration of some of the U.S.-based presence, both from the technology office perspective and from technology consultancy with the clients. I'm not saying there's more shift toward onshoring as a broad trend. If you look back pre-COVID days, our onshore presence between onshore technology people and the offshoring engineers who would come on long-term projects reached almost close to 20%. It's never been so low. When the onshoring presence pulled back due to remote work, a lot of work has been going offshore. We're not saying that work is no longer relevant, but there is more demand for presence on-site with clients to work together on complex cases because rapid transformation sometimes creates client uncertainty. There are two approaches clients take: one is status quo, and the other is rapid acceleration with concern about spend such as tokens and other costs. That's where our onshore headcount increases. The recent reduction in headcount was largely in non-engineering and non-forward-looking specialties. From the budget perspective, these people were not necessarily extremely expensive, but the infrastructure for them is no longer needed to serve markets better. To answer your question: we do see some additional growth of onshoring. The ability for us to prove margin expansion will continue is driven by how many fixed-bid, fixed-budget projects we can adapt, how many of our internal tools are accepted by clients, and how much nonlinear value we're bringing. We are executing on a clear plan to continue margin improvement.
Got it. Thank you.
Thank you, Puneet.
Great. Thanks, guys. Congrats on the results. I wanted to see if you could double-click on this new consultancy practice that you're talking about. Can you discuss more of how you see this business developing? I know you talked about activities like change management, but what sort of opportunities are you seeing in the pipeline built there? Who are you going up against in these bake-offs, and how is the competitive environment different from your traditional work, maybe?
I will start briefly, and then Vasily will expand on it. Matt, there are two parts. First, consultancy has always been part of our DNA. Nothing is earth-shattering because our clients consider us technology consultants, and that's why we're able to compete against the big firms. What has changed is the distribution of that offering. People we hire are extremely technical but also customer-oriented. We're expanding consultancy from pure technology consulting to AI infrastructure consulting, hardware selection consulting, tool selection consulting, and to some extent getting more into business consulting.
Think about business consulting as a natural extension of our technology enabler build-out capabilities. Essentially, the focus of customers is shifting from just creation of a system to creation of a system to change business processes. Therefore, they would like to analyze first which business processes are the best candidates to improve, which value is hidden there, then to build a technical enabler to reveal this value, and then adopt that technical enabler on an enterprise-wide scale. That's exactly where the focus of our consultancy is — not only to create the technical enabler, but also to help capture the value at enterprise scale. Right now we have several active engagements specifically on consulting and change management, which go along with technical enablers, and we see this opportunity ahead for strong growth in the future.
I can add that many of our customers observe performance and productivity gains from our delivery teams using our GAIN platforms. They become interested in having those platforms and methodologies inside their own software factories, and we are helping them to establish the tools, methodology, and change management required to capture similar productivity across the broader organization.
That's a good segue, Eugene, for my follow-up on GAIN adoption and just the S-curve that implies. As you accelerate GAIN rollout, how should we think about that adoption curve and pure AI revenue? Is it likely to scale linearly, or are you talking about wallet share gains from AI, such that we could see some exponential growth? How could you drive a sharper inflection in AI penetration with GAIN?
What is interesting about our GAIN strategy is that we are consolidating all our IP from multiple accounts and practices under the same umbrella, and AI helps us do that very rapidly. Our embedded, forward-deployed engineers are all tasked to bring back learnings and what works to the GAIN platform. Part of GAIN is also open source, which helps drive insights from the broader community and exposes GAIN to engineering leaders. At this point in time, we observe growth in direct revenue from our GAIN platform, but more importantly we observe growth in connection and expansion of our relationships inside accounts and new accounts driven by these platforms. We see inbound interest and conversations that result in new leads, opportunities, and converted business stemming from GAIN.
Good.
Thank you, guys. I'd like to start with a question around moving from proof-of-concept to actual implementation projects. That commentary seems to be more universal. From your perspective, can you talk about what the revenue journey for that looks like? For example, if a proof-of-concept project is a few hundred thousand dollars, does that become a $2 million project? What range of outcomes can we expect as more of those POCs convert and how will that impact growth rate?
Surinder, I'll give a high-level answer and then Vasily will add color. There are different types of proof-of-concept. The definition can be quite stretched both in intent and dollars associated with it. When we look at POCs resulting from partnerships, some become very meaningful programs where the customer embraces both the partner solution and our offering. For our GAIN productivity suite, many POCs are substantial projects because the measure of POCs is often time to implement rather than dollars. The short-term engagements that used to precede or follow a POC are now often the POC itself and then become major rollouts. This conceptually changes the definition of POC and the revenue associated with it. Vasily will add more details.
Many customers start with implementing smaller pieces of business cases to demonstrate results to their boards and management, and then use that as a precedent to request more investment. This typically converts into platform build-out and broader programs. For platform work, this work is much more sticky and longer-term in nature than POCs. Having built the platform, there is a growing appetite to add more business cases on top of it, and that is reflected in our pipeline.
I wanted to add that it also depends on the industry. We mentioned physical AI and robotics — there are many POCs in those areas, but at the same time, it depends on how deep you are in the customer relationship and other programs. Those POCs can be significant; sometimes a few weeks' engagement, sometimes six months or more. The GAIN model and our platforms aim to condense knowledge and speed up implementation, which influences the ratio of POC revenue versus longer-term implementation revenue.
To summarize for Surinder: POCs associated with forward-deployed engineering, consultancy, and GAIN modernization are substantial from the start. Physical AI POCs are more traditional in that they are innovative and their revenue will follow the typical scale-up pattern you expect as those technologies move from pilot to production.
Cool. Then thinking about the data and AI practice and the high growth rate that we're seeing there — AI revenue at 30% of total. Can you help me understand what's going on in the other 70%? When I do the math, it looks like roughly a 10% decline in that other 70% of revenues. How much of that is just cannibalization by the data and AI component, since new work probably falls into that bucket? Are there components in the legacy bucket like pricing compression or other factors to think about?
Very good question. I'll make it simple. From time to time we might see a significant customer — a top 30 or top 40 customer — have changes. Maybe there's a change in strategy, a big project completes, or something changes, and we can have some volatility. If you look over the past few quarters, those volatilities explain some of the variation. Second, regarding cannibalization: what we see is when we get into our clients, especially top 20 clients, we're going deeper and doing incremental AI work. Third, on pricing: we are not seeing pricing pressure. In fact, in the AI world there is a premium. When you look at what we're doing over the past couple of years, you don't see pricing declines across geographies. The reduction in cost per project or per functionality is happening because of higher productivity and shortened timelines, which leads to more work and more projects, rather than a reduction in revenue. So the apparent change in the composition is largely definitional and the result of a few volatile customer outcomes rather than systematic cannibalization or pricing erosion.
It's a question of semantics. The cost per project or cost per piece of scope is getting reduced because of higher productivity and shortened timelines. That leads to more work and more projects. That's an important point.
Our top accounts are expanding. Our AI programs are moving consistently from pilot to enterprise-scale deployment, and our platform portfolio is deepening both organically and through the capabilities we have added in robotics and physical AI. I'm confident in the second half of 2026: the strategy is working and the momentum is building.