EPAM Systems built its reputation on the unfashionable end of tech services: hard engineering. While rivals rode waves of outsourcing and app maintenance, EPAM won accounts by designing and building digital products—commerce platforms, data backbones, and cloud-native systems that had to work at scale. That focus left the company both exposed and advantaged as generative AI barrels through the consulting industry. The question now is whether EPAM’s engineering-first culture becomes an AI-native edge—or whether AI automates away too much of what clients used to pay it handsomely to do.

The timing of this strategic turn is noteworthy. After digesting the shock of Russia’s invasion of Ukraine and the company’s subsequent exit from Russia, EPAM has re-accelerated. In the June quarter, revenue rose 18% year over year to $1.35 billion, with non-GAAP operating margin at 15% and guidance nudged higher for the full year—signals that budget caution in 2024 is giving way to new spend, much of it tied to data and AI programs. This rebound coincides with a leadership handoff: founder Arkadiy Dobkin moved to executive chairman on Sept. 1, 2025, as long-time operator Balázs Fejes—an engineer by training and former CRO—took over as CEO. The transition was telegraphed in May and underscores a bet on continuity with a sharper AI push.

EPAM’s moat has never been about rate cards; it has been about credibility in building complicated things. That moat has three layers. First, product-engineering DNA: the firm grew up inside software and high-tech clients, where success is measured by shipped features and reliability rather than slides and service-level agreements. It is a founding member of the MACH Alliance, the evangelists of microservices, API-first architectures and headless commerce—architectures that are now the default target state for many enterprises modernizing for AI. Second, global delivery depth: before 2022, EPAM’s engine room sat across Ukraine, Belarus and Russia. It exited Russia and launched a $100 million Ukraine assistance fund, then sprinted to diversify, building out India into a five-figure talent hub and expanding nearshore capacity in Latin America, recently reinforced via the NEORIS deal. The result is a denser mesh of time-zone-aligned teams without abandoning Eastern Europe’s senior engineering core. Third, account stickiness: EPAM isn’t the cheapest bidder; it tends to win multi-year programs where it co-designs data models, platform interfaces and integration patterns—work that is difficult to rip and replace mid-stream. That’s reflected in utilization and steady organic growth returning in 2025.

AI complicates moats across IT services, but it also rewards firms that can turn models into working systems. EPAM has been busy productizing its know-how with DIAL, an open-source, agentic orchestration platform designed to help enterprises stitch together models, tools, and data stores without locking into a single vendor. This year the company released DIAL 3.0 and listed it in AWS’s new AI Agents & Tools marketplace—a pragmatic move that meets clients where their cloud budgets already live and taps Amazon Bedrock’s model catalog (including Anthropic’s Claude) for rapid experimentation. The surrounding services story—advisory, data plumbing, retrieval pipelines, governance—matters as much as the platform. EPAM has been formalizing that stack, from “AI-native engineering” methods across the software life cycle to reference solutions that show rather than tell, like a travel-planning Copilot built on Bedrock and Kendra or a GenAI productivity program for an industrial client. Done well, this becomes repeatable IP that shortens delivery and improves margins.

The demand backdrop is supportive. In its 2025 survey of more than 7,300 executives and engineers, EPAM found companies expect to lift AI spending by roughly 14% year over year, even as many still struggle to scale pilots into production. That “last mile” is exactly where engineering-heavy integrators earn their keep. The company’s June-quarter commentary echoed the pattern: organic growth is back, and the strongest calls are in industries that need data modernization to unlock AI—financial services, healthcare, software, and consumer goods.

But AI is also the sharpest knife pointed at EPAM’s cost structure and pricing. Generative tools compress the labor hours needed for code generation, test creation and documentation—activities that once filled benches. That threatens the “effort times rate” math across legacy work, especially for simpler systems. At the same time, hyperscalers are racing to sell prebuilt copilots and industry “solutions” that nibble at services scopes. The mitigant is to move up the value chain: design the data layer, build agentic workflows that orchestrate multiple systems, wrap in governance and observability, and own the reference architectures. That is effectively the DIAL thesis. If EPAM can make its platform the default way clients industrialize AI on AWS and beyond, it preserves margin while remaining vendor-agnostic enough to avoid a single-cloud trap.

Execution risk remains real. Geopolitics rewired EPAM’s delivery map in months, not years; while the company did the right thing exiting Russia and supporting Ukrainian staff, shocks of that scale create cultural strain and account friction. The pivot to India and Latin America—accelerated by the NEORIS acquisition—adds resilience, but also management complexity. Blending utilization discipline with a culture that prizes craftsmanship is tricky when projects get chunked into smaller AI sprints. Competition is no gentler: Accenture is arming thousands of consultants with GenAI accelerators; Infosys, TCS and Cognizant have cost advantages at scale; Globant and Endava nip at EPAM’s heels in digital-product niches. The antidote is to keep the bar high on engineering talent and to own more of the client’s platform roadmap rather than a single application lane.

Investors should also watch for how the new CEO frames the balance between platform and people. Fejes has already signaled an AI-native posture, and the operating metrics hint at a company running tighter—78% utilization in Q2, up year on year, and three straight quarters of positive organic growth. The danger would be letting platform talk outrun the reality that services firms live and die by account expansion and delivery quality. If DIAL becomes a doorway to larger data and modernization programs, it’s a meaningful asset. If it’s merely a demo shelf, it risks distracting from the core craft.

The base case is more optimistic than not. EPAM’s core customers aren’t chasing AI for novelty; they’re trying to make sprawling estates searchable, compliant, and automatable. Those are engineering problems first, procurement problems second. A company that has spent three decades knitting together complex systems should be able to translate that muscle into AI era patterns—RAG pipelines with real provenance, agents with guardrails, copilots wired to back-office logic—so long as it keeps owning the messy middle where models meet enterprise data. The market seems to be voting with budgets: June-quarter growth, a raised outlook, and a leadership team that speaks in architectures rather than adjectives. None of that guarantees outsized returns. But in a services landscape where hype is cheap and hard problems pay, EPAM remains one of the few shops that clients call when they need a platform built, not just a slide deck written.

If there’s a one-line answer to whether AI is an opportunity or a threat, it’s this: for EPAM, AI is both a margin squeeze on yesterday’s work and a catalyst for tomorrow’s. The outcome depends less on prompt tricks than on whether the company can keep turning engineering into enterprise outcomes at scale, while making its own tools the shortest path from pilot to production. The founder’s hand is still on the tiller as executive chairman; the new CEO is an operator steeped in the culture. That gives EPAM a shot to let AI widen, not wash out, its moat. At an EV/EBITDA of 8x, a lot of bad news is priced in. That's why EPAM is our latest high conviction idea.