Grid Dynamics (GDYN) 2026 年第三季法說會：營收創新高與 AI 推動利潤率擴張

重點摘要

• Grid Dynamics 公佈 2026 年第三季營收創下 1.143 億美元的新高，年增 9.7%，季增 5.7%。營收超出公司先前給出的 1.12 億至 1.14 億美元財測區間。
• 非 GAAP EBITDA 創下 2,290 萬美元的新高，占營收的 20.0%，而去年同期為 1,270 萬美元（占 12.2%）。管理層將此改善歸因於營收成長、成本舉措以及 AI 驅動的生產力提升。
• AI 業務營收連續第三個季度年增 45%，逼近總營收的 32%。管理層預期，隨著 AI 業務營收占比接近並突破 40%，整體的成長態勢將越來越顯著地反映 AI 業務的表現。
• 每位員工平均營收首次突破 10 萬美元。營收年增 9.7%，同時總員工人數下降 8.3%，兩者出現 18 個百分點的分歧。
• 公司預估 2026 年第四季營收為 1.135 億至 1.155 億美元，非 GAAP EBITDA 為 2,000 萬至 2,200 萬美元。
• 在既有客戶、更長期合約、AI 原生交付以及合作夥伴引導的商機支持下，管理層預計 Grid Dynamics 將在 2027 年重返強勁的雙位數成長。

核心財務數據

GAAP 與非 GAAP 毛利率均季增約 370 個基點。公司指出，這主要得益於資源利用率提升、非可計費資源與工作時間效率的改進。

Grid Dynamics 在本季度以 670 萬美元回購了約 90 萬股股票。自授權執行 5,000 萬美元回購計劃以來，該公司已累計以 3,750 萬美元回購了約 550 萬股股票。

業務與營運表現

科技、媒體與電信 (TMT) 仍是最大的垂直領域，占營收的 32.3%。該領域營收年增 29.3%，季增 7.6%，主要受惠於公司大型科技客戶的需求帶動。

零售業務占營收的 27.2%，年增 13.3%。金融服務占 22.5%，季增 3.9%，主要由全球資本市場客戶驅動。民生消費品與製造業貢獻了 10.9%，季增 5.8%。

AI 原生交付是公司生產力策略的核心。Grid Dynamics 表示，93% 的工程師已接受 AI 軟體開發生命週期實踐培訓，超越了截至 10 月底達到 90% 的目標。其軟體工廠模式利用 GAIN 平台和 AI 程式碼代理，支援以更小團隊執行固定價格與成果導向的專案。

管理層列舉了幾項可量化的客戶成果：某項代理型 (agentic) 解決方案將缺陷審查時間從 30 天縮短至 4 小時；某食品分銷生產解決方案所使用的工程師人數比原計劃減少了 40%；對於一家全球支付公司，現代化的金融犯罪平台將每筆交易的審查時間減少了將近 70%，合規誤報率降低了 90% 以上，每年為分析師節省超過 1,200 小時。

Grid Dynamics 還在實體 AI 與機器人領域擴展。專案包括為一家製藥公司部署內部物流人形機器人、為一家工業設備製造商開發自主系統，以及使用輝達 (NVIDIA) Isaac Sim 和 Google DeepMind MuJoCo 進行模擬工作。

客戶集中度有所上升。前 5 大和前 10 大客戶分別貢獻了第三季營收的 45.4% 與 63.3%，而去年同期分別為 40.1% 和 58.3%。

管理層財務預測

管理層預計 2027 年營收將實現強勁的雙位數成長，但未提供具體的年度目標區間。公司表示，利潤率的提升是結構性的，獲得更優質的 AI 專案組合、更高的費率、地點佈局策略、生產力工具以及嚴謹的營運支出控管所支持。

該公司還計劃在 2027 年建立規模更大的收購機會案源，目標鎖定能夠擴展現有業務或建立新業務的技術能力、領域專業知識與工程深度。

風險與觀察重點

• 第三季受益於工作天數效益以及 2025 年底啟動的營運費用優化舉措完全實現。管理層指出，受季節性因素影響，第一季的利潤率通常為全年最低。
• Grid Dynamics 計劃繼續投資機器人、實體 AI、網路安全、技術開發和員工培訓，這可能會抵消部分生產力提升帶來的效益。
• 固定價格專案涉及範疇的不確定性。公司的軟體工廠模式旨在降低變更帶來的成本與時間影響，但工程師仍須負責規格說明、專業判斷、驗證與控管。
• 大型客戶的營收集中度較去年同期上升，使這些主要客戶持續進行支出及其供應商定位的重要性日益提升。
• 管理層表達了在 2027 年進行更大規模併購的意願，但強調潛在交易在完成前仍存在不確定性。

分析師問答亮點

管理層表示，第三季利潤率的提升反映出更廣泛的結構性轉變，而非僅是單一季度的成本控制作為。關鍵因素包括更多固定價格的 AI 專案、更高的工程師邊際貢獻、更快速轉換為可計費任務，以及在招募、財務、人力資源和其他非工程部門實現自動化。

關於 2027 年的成長，管理層表示，無需依賴獲取大型新客戶，Grid Dynamics 即可實現雙位數擴展。公司認為在提升頂級客戶的錢包份額方面仍有龐大空間，同時超大規模雲端業者 (hyperscaler) 及科技合作夥伴關係也正為其開拓更多客戶群與垂直領域。

員工人數的減少被描述為主要集中在非工程部門。管理層預計隨著營收成長，工程人員數量將增加，但由於 AI 賦能交付和可重複使用的智慧財產權，增加幅度不會與營收呈直接正比。

關於諮詢服務，公司表示其服務著重於產品功能、實施優先順序以及商業目標與技術之間的連結，而非定位為傳統的高階策略諮詢。

電話會議完整逐字稿

---

完整財報電話會議逐字稿

管理層陳述

Operator

Good afternoon, everyone. Welcome to Grid Dynamics Third Quarter 2026 Earnings Conference Call. I'm Cary Savas, Director of Branding and Communications. [Operator Instructions] Joining us on the call today are CEO, Leonard Livschitz; CFO, Anil Doradla; and CTO, Eugene Steinberg. Following the prepared remarks, we will open the call to your questions. Please note that today's conference call is being recorded.

Before we begin, I'd like to remind everyone that today's discussion will contain forward-looking statements. This includes our business and financial outlook and the answers to some of your questions. Such statements are subject to the risks and uncertainty as described in the company's earnings release and other filings with the SEC. During this call, we will discuss certain non-GAAP measures of our performance. GAAP to non-GAAP financial reconciliations and supplemental financial information are provided in the earnings press release and the 8-K filed with the SEC.

You can find all the information I just described in the Investor Relations section of our website. I now turn the call over to Leonard, our CEO.

Leonard Livschitz

Thank you, Cary. Good afternoon, everyone, and thank you for joining us today. Q3 2026 was a pivotal quarter with record financial results. Revenue of $114.3 million is the highest in our history and came in above the high end of our guidance range and well ahead of Wall Street expectations. Non-GAAP EBITDA was $22.9 million, also a record.

Our efforts are paying off. We're growing faster, more profitable and in the verticals and applications where AI enterprise spend is centered. We entered the fourth quarter with confidence and with a scalable client demand setting a positive backdrop for 2027. In the third quarter, our AI revenues grew by 45% on a year-over-year basis. This marks the third consecutive quarter where our AI revenue has grown over 45% on a year-over-year basis and is now approaching 32% of our total revenue.

Companies that recently initiated AI exploratory projects are now signing multi-quarter production programs with us. Last quarter, I set a goal, 90% of our engineers trained on AI SDLC by the end of October. I'm happy to report we are currently at 93% and continue to expand. Grid Dynamics has always been a technology-first engineering company and investing in our people ahead of client demand and simply how we operate. Going forward, I'm optimistic about our revenue growth trajectory. And that optimism is grounded in the structure, not sentiment. As we look in 2027, we expect to revert to healthy double-digit growth and sustain that momentum over the following years. Our AI revenue is compounding on an even larger base. As it approaches and exceeds 40% of our total revenue, the growth rate of the overall business will increasingly reflect the growth rate of the AI practice. Our AI-based GAIN solutions continue to grow. We generate more revenue per headcount at an accelerated pace. We expand our physical AI and capital markets offerings.

And our hyperscaler and NVIDIA partnerships create a cross-sell motion that complements our direct sales. Altogether, they are the reasons to be optimistic. We committed to 300 basis points of margin expansion in the first quarter of 2026 over the first quarter of 2025, and we exceeded our goal a quarter early. The structural drivers are a greater mix of AI projects, higher billing rates, optimal location strategy execution and disciplined OpEx management. These margins represent a new norm and in 2027, we'll continue to execute on the same operating framework.

But the most important trend in our business this year is the growing revenue per employee. In the third quarter, we exceeded $100,000 for the first time. Let me remind you that this is based on the total headcount and not just billable engineers. In the second quarter, revenue grew 7% year-over-year with headcount declined 3.5%, a divergence of more than 10 percentage points.

In the third quarter, the gap widened further. Revenue grew 9.7% year-over-year, while headcount declined 8.3%, a divergence of 18 percentage points. We believe this is one of the strongest headcount to revenue decoupling metrics in the IT services industry today. These divergence reflects systematic changes across how we run our operations and deliver to our clients. Internally, AI tools are transforming every nonengineering function, recruitment, RFP responses, knowledge management, finance and HR.

We are processing more candidates, responding to more proposals and managing more client relationship with the same or fewer people. Doing more with less is becoming the signature for every AI native program we're on. Beyond operations, the shift towards platform-based, fixed price and output-based engagements means we are monetizing intellectual property alongside the headcount. When clients engage with our GAIN platforms, they get a combination of talent, processes and IT.

As such component of our revenue grows, the link between headcount and revenue weakens by design. In a sector where headcount and revenue has historically moved in lockstep, Grid Dynamics is demonstrating that the 2 can should be decoupled. Now about AI native delivery. 12 months ago, AI native delivery was the direction we were heading towards. Today it is how we deliver. A few examples. At a leading global technology company, we secured a large AI native fixed price project. At one of the largest food distribution company in the world, we delivered a production solution and have the plan time with 40% fewer engineers. At a leading global technology platform, an agentic solution reduced defect review time from 30 days to 4 hours.

Our GAIN platforms are deployed across multiple active projects and our embedded engineers arrive with codified IP that clients are paying premium for. The commercial headline is clear. The AI native model is validating its economics. Our CTO will take you deeper into the technology progress in a moment. Our vertical diversification continues to reshape the composition of our business. Technology and financial services now dominate our top account relationships.

The retail exposure that once define our revenue mix is now smaller, and we are less dependent on them for growth. The growth engine of the business is firmly in the sector where AI adoption is accelerating, and this is precisely where we have concentrated our investments and our talent over the past few years. Our capital markets practice is growing rapidly with expanding teams across U.S., U.K. and India. In manufacturing and supply chain, a leading global energy infrastructure company selected us for AI native application modernization. A major automotive part retailer is expanding its modernization program and a leading global construction and industrial equipment manufacturer is advancing from a discovery to implementation of autonomous system program.

In physical AI, our first intralogistics humanoid deployment is on track for delivery at the global pharmaceutical company. The last couple of years have been a period of transition and execution as the company focused on growing revenue, enabling AI expertise and improving margins. During this time, our targeted approach to capability-driven acquisition in capital markets and robotics serve us well.

We will enter 2027 in a stronger position. The business is growing, margins are expanding and the strategic clarity we have built around AI native delivery, productized engineering and physical AI gives us a sharper lens for evaluating acquisition targets. We are acquiring capability, domain knowledge and engineering depth that either accelerates an existing practice or opens a new one. In 2027, our focus is on building a pipeline of larger and more strategically impactful opportunities. After 2 years of disciplined execution, we're ready to acquire larger companies in 2027.

During the quarter, Grid Dynamics delivered some notable projects. For a leading global technology company, Grid Dynamics built a data foundation for the launch of a new subscription package that brings together multiple online services as a bundle. The new system enables accurate tracking of the entire subscriber life cycle, including acquisition, retention, churn and transition between subscription offerings while ensuring reliable revenue attribution and engagement analytics across multiple services. Delivered in time for the global launch, the engagement involves modernizing interconnected production data pipelines, ensuring data accuracy at scale and establishing a scalable foundation for additional subscription models.

The solution provides the critical analytics and financial reporting capabilities needed to support the successful launch of the new bundle offering across more than 100 markets, serving millions of subscribers worldwide.

For a leading U.S.-based global bank, Grid Dynamics continued expanding the enterprise AI platform that powers AI-driven productivity and automation across the organization. This quarter, we delivered new capabilities that enable employees to launch long-running AI agents directly from their AI productivity workspace as well as a centralized service analytics platform that provides management with detailed visibility and control over AI adoption, access permissions and costs. These innovations enable employees to delegate complex time-consuming tasks to AI agents, freeing up their time to focus on higher-value work while giving management greater transparency and control over AI operations at scale.

These enhancements extend the capability of an AI workspace used daily by more than 180,000 employees across the bank. For a leading global payment company, Grid Dynamics modernized its core financial crime prevention platform, introducing machine learning fraud detection models, self-service anti-money laundering capabilities and agentic AI tools for sanctions, research and alert characterization. The modernized platform now screens approximately 14 million cross-border transactions annually through financial crime, which is 20% increase year-over-year while reducing screening time per transaction by nearly 70%. Compliance false positives dropped by more than 90%, significantly reduced unnecessary manual reviews.

AI-powered automation saves compliance analysts more than 1,200 hours annually, while self-service capability enabled teams to launch new ML rules in minutes instead of weeks. The platform also achieved 99.99% availability, supporting reliable, continuous payment process. For an advanced humanoid robotics company, we're leading simulation effort using NVIDIA Isaac Sim and Google DeepMind MuJoCo. We are focused on advanced physics-based simulation to refine complex articulated movements for industrial applications such as welding automation.

Leveraging our broader robotics capabilities, our technical scope quickly expanded into autonomy and manipulation architecture. Next, we plan to extend our engagement by helping the robots better perceive their surroundings and develop more dynamic movements. With that, let me turn the call over to our CTO, who will present our exciting technology offerings and AI-driven innovations. Eugene?

Eugene Steinberg

Thank you, Leonard. Good afternoon. Leonard described revenue decoupling from the headcount. Let me show you the engineering behind this process. The mechanism is transition to fixed-price projects and AI native delivery for billable work and doing more with less by using AI for our internal productivity. It all comes down to one idea, the software factory powered by our GAIN platforms.

On every client project, we start by building a software factory, which is a client-specific harness for general purpose AI coding agents. Since this factory does much of a heavy lifting, this allows a smaller team to deliver a large scope. Most importantly, it makes change cheaper. Such a factory approach makes projects shorter and more predictable. At the end of the day, this is what allows us to commit to a fixed price project in a face of scope uncertainty.

Every fixed price project starts half understood and half specified. But much of the detail only emerged during delivery, including problems nobody could predict. In traditional delivery, this risk is covered by pegging the time line and team size. The factory changes this situation. First and foremost, it makes change significantly cheaper and faster. When something unexpected comes up, the factory rebuilds effective components quickly and reliably so that the project becomes more flexible and easier to manage and control. While AI agent provides the execution capacity, engineers provide specs, exercise judgment and ensure verification and control. This is why we can bid with an aggressive time line and a smaller team while still delivering profitably.

Here is an example, an AI bid manager assistant for a major food distributor. Previously, every RFP bid required hundreds of hours from dozens of specialists. Now the AI assistant matches the RFP request with a distributor catalog and best supply offers. The first phase of this program is now in production with the goal to achieve faster, cheaper and more competitive bids.

We are building this program with a factory approach. The work runs in several work streams of 4 to 5 engineers each. These factory teams are measured to deliver twice as much progress as conventional AI system teams on the same program. When we build a software factory, our GAIN platforms like Rosetta or Allium, provides reusable engineering components and tools. After the project is done, we carry the learnings back to those platforms to continually compound knowledge.

The critical aspect of AI delivery is security. Clients want the speed of AI native delivery without adding new vulnerabilities and new agentic threats require an adequate agentic response. Traditional security tools check code after it is written. We go further by shifting security left. Our AI agents work inside the development process itself. So security checks happen while the code has been specified and developed.

Specifically in capital markets, banks want correctness they can prove in front of regulators. This is where we deploy Allium. It helps to clearly specify functionality and find and fix vulnerabilities before we might leak into the release.

We apply the same factory approach inside Grid Dynamics. It contributes to the productivity gains Leonard described outside client work. Our internal tools team can deliver more with less and growing numbers of agentic applications and workflows already run in the Grid Dynamics environment. We drink our own champagne.

Additionally, we continue to build up physical AI capabilities, pipeline and projects. Next-generation robots are trained on a mix of real-world examples and simulation. This is where Grid Dynamics fits perfectly. They bring deep robotics research, simulation experience and AI engineering that runs reliably at enterprise scale. We are working with our clients and partners to take AI controlled robots from labs to the production line. That is the work behind the 2 programs Leonard mentioned.

One is the autonomous vehicle program of the industry equipment maker. The other is the humanoid robot for a pharmaceutical company moving materials inside its facilities. Both programs are growing. Grid Dynamics continues to expand its robotics partnership network, adding 3 robotics partners in the third quarter. Together, we participate in top-tier industry events like GTC and Berlin, actively building our physical AI and robotics business. The Grid Dynamics team is proud of the progress we are making on AI transformation, and we are excited about the journey ahead. Anil, over to you.

Anil Doradla

Thanks, Eugene. Good afternoon, everyone. The third quarter came in at $114.3 million, above the high end of our guidance range of $112 million to $114 million. That represents 9.7% year-over-year growth and 5.7% sequential growth. Our top customers continue to execute well with our top 10 customers growing at 8.6% on a sequential basis.

Non-GAAP EBITDA was $22.9 million, our highest quarterly EBITDA on record, representing 20% of revenues and coming in well above our $16.5 million to $17.5 million guidance range. The strong sequential improvement in EBITDA was driven by a combination of strong quarter-over-quarter revenues growth, aided by our cost improvement initiatives and AI-driven productivity improvements that resulted in better-than-expected cost of sales and operating savings.

Looking at the performance of our verticals, TMT remained our largest vertical and accounted for 32.3% of total revenues for the quarter, with a growth of 7.6% sequentially and 29.3% on a year-over-year basis. The growth was primarily driven by higher demand from our largest technology customers.

Retail contributed 27.2% of total revenues in the third quarter of 2026. Revenues grew 8.5% sequentially and 13.3% on a year-over-year basis, primarily driven by increased demand from specialty and home improvement retailers. Finance accounted for 22.5% of total revenues in the quarter. Revenues grew 3.9% sequentially, largely driven by global capital market customers.

CPG and manufacturing represented 10.9% of quarterly revenues and grew 5.8% sequentially. The growth came from a leading wholesale food distributor and from our manufacturing customers.

Turning to our remaining verticals. Our other vertical contributed 5.3% of third quarter revenues, while health care and pharma accounted for 1.8% of our revenues for the quarter. We ended the third quarter with a total headcount of 4,559, down from 4,838 employees in the second quarter of 2026 and from 4,971 in the third quarter of 2025. Our total headcount trends reflects productivity gains, particularly from our non-engineering functions.

At the end of the third quarter of 2026, our total U.S. headcount was 377 or 8.3% of the company's total headcount. Our non-U.S. headcount located in Europe, Americas and India was 4,182 or 91.7%. In the third quarter, revenues from our top 5 and top 10 customers were 45.4% and 63.3%, respectively, versus 40.1% and 58.3% in the same period a year ago.

Moving on to the income statement. Our GAAP gross profit during the quarter was $46.1 million or 40.3% compared to $39.6 million or 36.6% in the second quarter of 2026 and $34.7 million or 33.3% in the quarter a year ago. On a non-GAAP basis, our gross profit was $46.5 million or 40.6% compared to $40 million or 36.9% in the second quarter of 2026 and $35.2 million or 33.8% in the year ago quarter.

On a sequential basis, gross margin expanded by approximately 370 basis points on both GAAP and non-GAAP basis. The increase in both GAAP and non-GAAP gross margin percentage on a sequential basis was primarily driven by a combination of improved resource utilization, efficiencies with nonbillable resources and working time.

Our non-GAAP EBITDA during the third quarter that excluded interest, taxes, depreciation and amortization and other income and expenses, fair value adjustments, stock-based compensation, transaction and transformation-related costs and restructuring costs as well as geographic reorganization expenses was $22.9 million or 20% of revenues versus $14.7 million or 13.6% of revenues in the second quarter of 2026 and was up from $12.7 million or 12.2% in the year ago quarter.

Our GAAP net income in the third quarter was $6.2 million or $0.08 per share based on a diluted share count of 81.9 million shares compared to the second quarter net income of $2.9 million or $0.03 per share based on a diluted share count of 83 million and a net income of $1.2 million or $0.01 per share based on 85.8 million diluted shares in the year ago quarter.

On a non-GAAP basis, in the third quarter, our non-GAAP net income was $14.7 million or $0.18 per share based on 81.9 million diluted shares compared to the second quarter non-GAAP net income of $9 million or $0.11 per share based on 83 million diluted shares and $8.2 million or $0.09 per share based on 85.8 million diluted shares in the year ago quarter.

On September 30, 2026, our cash and cash equivalents totaled $304.3 million, up from $298.4 million on June 30, 2026. During the third quarter, we repurchased approximately 0.9 million shares for a total consideration of $6.7 million. Cumulatively, since our Board authorized the $50 million share repurchase program, we have repurchased approximately 5.5 million shares for a total consideration of $37.5 million, reflecting our continued confidence in the long-term value of the business.

Coming to the fourth quarter guidance, we expect revenues to be in the range of $113.5 million to $115.5 million. We expect our fourth quarter non-GAAP EBITDA to be in the range of $20 million to $22 million. For the fourth quarter 2026, we expect our basic share count to be in the range of 81 million to 82 million and our diluted share count to be in the range of 82 million to 83 million shares. That concludes my prepared remarks. We're ready to take your questions.

Operator

[Operator Instructions] The first question comes from Maggie Nolan of William Blair.

分析師問答

Margaret Nolan

Leonard, so much went right this quarter across growth and margins and productivity. So if you had to distill it down to what really happened here, maybe what changed most meaningfully in the business and how sustainable is that over the next several years?

Leonard Livschitz

It's always good to start the conversation in such a nice moment of time. It's been long awaited for. So I think instead of making a long speech about all these details, I think that fundamentally, we have approached the structural change of the company, and we're making the change implemented throughout various organizations. So there are several parts of it. On the client side, you heard the prepared remarks, you will hear from the guys. We're doing more fixed bid projects. We're having a better positioning with the growth of AI implementation solutions.

So we're doing the right things, so it actually helps us with the margin profile. On the cost side, there are several elements. And one of the very key one is that we've been talking for quarters in terms of that we are the client zero for many of the solutions we roll out to the clients. And that actually helps us with ability to create a more efficient back-end organizations, tooling, recruiting, hiring operations.

You've noticed there is probably most significant divergence from the perspective of revenue versus headcount. And that's a reason for that. And the reason being is not only we're automating more of the solutions, but we are impacting virtually the ability of us to kind of consolidate many of the functions.

Remember, in 2022, we kind of grew to so many countries, and we had to contain the ability to actually manage our organization through all these different countries and offices. I think we come to the point we are fully in control. So that's one thing on the cost side. The other one is, again, more efficient deployment of engineers. I think we are having more predictive tools. We understand where technology leads, so we can have enough time to prepare the engineers and proper rightsize the teams where they're properly deployed.

Margaret Nolan

Very helpful. And then, Eugene, as CTO in the middle of this transformative technology cycle, I'm curious where you're spending most of your time and how that's maybe shifted over the last year?

Eugene Steinberg

Yes. I'm spending my time. I'm spending my time across the 3 horizons I've been always talking about on these calls. I'm scaling things which work, things like AI-first modernizations of applications and data, GAIN platforms, which are taken off the ground and scaling.

On the second horizon, I'm working to nurture and help to take over the ground the things which are emerging like AI harness engineering, physical AI projects. And of course, I'm always looking ahead and what happens next, things like AI native cybersecurity, sovereign AI and many other interesting emerging themes, which we consider as the next horizon for us.

Operator

The next question comes from Puneet Jain of JPMorgan.

Puneet Jain

So you recently announced winning like the Partner of the Year award at Cisco, given that client also works for -- works with other large vendors that have much larger accounts. Could you share like how you were able to outperform other vendors? How you were able to punch above your weight, if you will, at that account? And is that replicable at other accounts?

Leonard Livschitz

Sure. Well, the hero of the day is not here today. We will have our earnings call early next year, and we will have Vasily, who runs our sales operations. He will tell you all the details, but I was actually putting very intimately involved with the account as well. So what happened is many things start with understanding of the customer tasks. And usually for Grid Dynamics, you need to rely on the previous background. Even though it's not our traditional retail, but as many things, logistics related to B2B instead of B2C, we developed and governed many of the capabilities and accelerators were in that.

So the people who we engage with, they actually realized that our expertise would give us the good chance to start the relationship. What happened then -- it was not a typical land and expand. Of course, the common gravity was the partnership with one of the major hyperscaler. But then we had a runaway train because we quickly not only uncovered the challenges customers need to solve rather rapidly, but scale the solution and prove the real-time return on investment fairly quickly.

To make story short, we're well positioned in multiple regions. The customer utilizes us across the globe. Right now, part of our team is actually in one of the offshore location, helping them to catch up because we are very, very friendly with the clients. We don't want to cannibalize everything from themselves to take over. Our business is to be partnering with them. So the business is growing, and I was there at the presentation. And what really humbled me is how necessary and how timely our engagement has happened. So capabilities, unrelentless support, knowledge and factual performance.

Puneet Jain

No, that's great to hear. And also nice to see this new factory approach. But given adoption of this new model will require thousands of employees across the globe embracing and execute against this delivery model. So talk to us how you -- like your approach to achieve that, both from a change management perspective as you delink revenue from headcount, headcount was down this quarter as well as from a technical skill perspective for your employees?

Leonard Livschitz

Well, there are multiple facets of this question. I think, Puneet, you have to show the secret sauce of Grid Dynamics. And there are many spices. So maybe it's good to wait for a celebration to give you some more runway. But in short, there are a few key elements. First of all, it starts with a really tiresome or I would say, not tiresome shoulders of our CTO and his organization. Many of the practices when it comes to the implementations need to be trained and run internally. We need to evaluate multiple models. We need to read out solutions, which perhaps are a bit too optimistic. And more important, being a thrifty company as we are, we need to make sure that there is no overreach and overspend. There are companies who are reporting in the season and some of them are very much overextended. We don't want that to happen to our clients.

So I think what happens in first and foremost, we build a mousetrap for the client to deliver the capabilities within reasonable time and control cost. Of course, the solution change rapidly. So we have to build more and more internal know-how, which we do. But the pressure comes on actual implementation teams. And that's where -- and you will see that a little bit as we come forward that we have more on-prem onshore resources, our talented enterprise architects and our leading specialists who work very closely with the customer trying to crack down exactly what they need. And this is ultimately what fixed projects are.

You have to be able to understand to a much higher level of granularity, the expected outcomes. So we're learning it. We're training that. It's on the R&D side, on the delivery side, and it also really comes from the ability for customers to build their stepping stone in the solution. We're not offering just one solution fits all. And that's why we kind of enjoyed in place between 2 very large campuses.

So we're not heavy on OpEx. We're not spending a huge amount of money on the back-end hardware capability. And we're not married to one particular enterprise software product. So we're kind of on our groove with the AI deployment solution as we were years ago in digital transformation and cloud migrations.

Operator

The next set of questions comes from Bryan Bergin of TD Securities.

Bryan Bergin

I want to start first on encouraging commentary here on 2027. So the growth profile that you see the recovery there, getting back to double digits. But not trying to get too far ahead of ourselves. But just based on what you're seeing in the conversation today, do you expect the growth to kind of be more broad-based growth attribution across industries and across clients? Or is it more so you have visibility to some of these really large strong relationships and that they're going to really carry the load or continue to carry the load?

Leonard Livschitz

It's a little bit of both. I think, Bryan, the right answer is the clients where we have a very strong long-lasting relationship, we do a lot of experimentation. We see a little bit better. The budget versus returns going more than just a quarter ahead and sometimes more than a year ahead. We start signing much longer deals than we had before. So this is factual. And that gives us a little bit more productivity from a planning perspective.

You've been with us for many years, and you've been -- the biggest question is how is the engineering company can leave out of every quarter resigning new SOW, right? So that has changed dramatically. We expanded it beyond the year more horizon. We're not there yet on a BPO type of style, 5 years, 7 years renewables, but we are getting very, very well controlled. So that's on existing clients.

The partnerships brought us to the clients who need solutions the most right now, right? And those relationships start accelerating way more faster than they were before. Typically, in the past world, we would start with a proof of concept, and it will take 3 months to 6 months just to analyze the situation. Now it's a fail fast.

In other words, we quickly determine what works, what doesn't work. Something always works in a new world, right? So we just need to be experts on a consultancy, technology consultancy, which we have become and growing. So when we looked at forecasting the business, it's too soon to guide exactly where we are, and you guys love the numbers. And for me, I just look at now and I look back where we were during the good years, and I see so many proven similarities. So that's why we look for the next 2 years, 3 years with a very high level of confidence, plus the mousetrap is widening, and there are more and more partners creating the relations with us in new verticals. And those verticals may not give us revenue in the next 3 months to 6 months as much as the existing business, but they're rapidly getting deployed.

Anil Doradla

And adding one point to that, Bryan, if you look at our top customers, we talked about this, right, vendor consolidation. So as vendors are getting consolidated, we get more wallet share, too. So we're seeing that also.

Bryan Bergin

Okay. Very good. My follow-up is on the announcement you had this week on the business consulting launch. So as it relates to a couple here really. How much of this opportunity would you say is incremental kind of consulting-led spend versus simply pulling forward AI engineering work you were already pursuing? Kind of what do you see as being the biggest challenges as you try and blend this model with the engineering prowess that you have? And just how do you -- how would you say it's different than traditional management and advisory consultancies?

Leonard Livschitz

Okay. Well, I think we'll start the question with Eugene, do you want to start?

Eugene Steinberg

Yes. So the business consultancy, which we are talking about is not about strategy. It's about functionality and the product. What is happening right now is with AI agents, which are doing the heavy lifting of the coding itself, the value shifts very quickly towards what to do and why. And this is where we see a lot of value which we can bring advising our clients not only on how to do things, but also what to do, why to do that and bringing our best specialists, architects and product managers to the forward deployed roles and help them to get their head together what is possible right now with AI, where to invest and how to connect all these to technology. This is the essence of this process right now, not talking about brand strategy with the customers.

Leonard Livschitz

Well -- and also to add to that, I think it's important to understand that the definition of Eugene alluded to the business consultants has changed. Some of those companies who are very strategic business consultants come to us. And they want us together with them to create this kind of gap filler between this high-level business needs and realistic implementations. We usually work, as you know, historically with the technology teams with the CTOs, CIOs, VP of engineering, in some cases, marketing people. Now they're going instrumental across all the facets of the companies because they want to see their spend to be projected and understood for years to come.

And of course, with the token factories with all the spend, many of the top decision-makers and boards are a bit worried. So we're giving them a bit more, I would say, concise and more confident road map of how the spend will convert into some of the solutions we are offering as an engineering organization.

Operator

The next set of questions come from Mayank Tandon of Needham & Company.

Mayank Tandon

Congrats on the quarter. So I wanted to just start with -- Anil, you made a comment about growth potential within your top 5, top 10 accounts. So how much capacity do you have left to grow with them? And then as you try to grow low double digits or whatever, I think Leonard mentioned double digits, I shouldn't say low double digits. Do you need a certain level of new logo activity to be able to drive that level of growth? Or can you grow within your existing base and growth from new logos would be maybe more grading on top of that?

Anil Doradla

So look, I mean, if I look at the wallet spend of my top 10 customers, I think we're not even a rounding error at this point. I mean they are very small portions. So Mayank, if the question is, based on the spend of our top 5, top 10, top 20 clients, can we continue growing? The answer is yes. And you've seen it, right? Our top 5, top 10 continue to grow quarter after quarter, year after year. I think a couple of things are changing. Grid Dynamics is approaching $0.5 billion.

We're now, I call a global company, truly global country company where we are having our engineers in the locations where most of our global clients are. So the way they're treating us is slightly on the higher scale, so to speak. And I made a comment with -- I think it was Puneet or Bryan, vendor consolidation is a very important theme over the last 2 years. See, at these very, very large logos. They've got dozens and dozens of partners. And as this technological revolution is taking place, they want to focus. They want to consolidate.

Grid Dynamics is always an outlier. Our DNA, our structure and another interesting thing in the past, scale was considered a positive thing. Now lack of scale or smaller size is considered a very positive thing. So the number of conversations we're seeing with C-level Fortune 1000, Fortune 100 customers, I love your size, Grid Dynamics. You're in the 4,000 to 5,000. When we kick off this AI project, 50 people, 100 people, you bring the granularity, you bring the focus, you bring a high quality. When we look at the well-established vendors, they're in hundreds of thousands, and they're still thinking in terms of headcount. So I gave you a little bit long answer, but in a nutshell, we will continue on.

Now Leonard pointed out something on the partnerships. That's very exciting. And through partnerships, we find we get different types of clients, clients that we never were able to get access to. And in some cases, these clients have grown very fast. So the pipeline of new clients through our partnership is a multiplier effect. And within our existing book of business, we're just becoming more meaningful to our partners.

Leonard Livschitz

So just to be very clear, Mayank, we can grow double digits without major new fines where we would need to go out and search new business with a new client. As Anil said, there is a strong position with existing clients, top 10, 10, 20. The share of the wallet goes up. At the same time, you know very well, there were years, and it may happen in the future as well, there could be some of the fluctuation with these clients. So now we have the faster-growing acquisition of the new clients.

We talk about Sysco Foods, it's been a matter of 2 years where we got to that level and faster. But even more on the technology side, on the capital market side, on the manufacturing industrial, on robotics, on solutions for pharmaceutical companies, U.S., Europe, Southeast Asia. So just the sheer momentum is exactly where we want to be, to be confident on the growth. And of course, being where the clients are. I think they need to have fewer with extremely relevant people. And we may not be staying at this 4,000 to 5,000 people. I think we're going to grow, but it's not going to be incremental growth to meet the client demand proportionately to the headcount. We crossed 100,000 per employee. We grow significantly faster than we expected. We're growing, of course, by doable engineers and all this good stuff. So just keep watching us.

Mayank Tandon

Very helpful. That's super encouraging to hear. And then maybe turning to margins. Obviously, you're going to end up with your target goal of 300 basis points year-on-year. So just thinking about '27, again, not looking for specific guidance, but just directionally, as you approach $0.5 billion in revenue potentially with double-digit growth, how should we think about growth and EBITDA margins into '27?

Anil Doradla

I knew this was going to come. Look, I mean, you're absolutely right. We delivered 1 quarter before. But remember, Mayan, the work started in Q4 of last year, right? This is not something we've started working on 1 quarter or 2 quarters. It's been in the works for some time. The timing happens such that sometimes things align sooner, we're executing better. I think in Leonard's comments, a very important thing he said, which is structural.

Look -- and now I know we are very excited that our margins is getting up, but look back way back when, we're approaching our historical margins, right? So it's not like we've never done this. We used to do it, but obviously, over the last 3 years, you know what happened. So we're approaching that. All the things that we're doing are not a 1-quarter thing, right, whether it's on the COGS side, whether it's on the OpEx side. And we've listed almost half a dozen of these.

So all I can say is that some of these things that we're doing are not 1 quarter. Obviously, you're going to have seasonality in the course of the year, so that goes without saying. And what we are trying to get to is a company as we get into 2027, where we have a certain growth vector on the revenue and a certain structure on the margins. So I'll just leave it at that. And look, we'll come back and give more color.

Leonard Livschitz

And Mayank, very important from me as a CEO perspective. We're not trying to squeeze the water out of the stone. There are a couple of areas of investment will only continue to grow. there's unrelentless commitment to technology investments. We have to do it. This is the foundation and training and developing proper engineers. We talk about 90-plus percent SDLC training, we're leveraging our GAIN model, 93-plus percent. Those are not just numbers. This is a huge work behind the scene. And to me, we will properly position ourselves from margin investment perspective to make sure you guys will see our growth continue for a long time.

Operator

The next set of questions comes from Surinder Thind of Jefferies.

Surinder Thind

I'd like to revisit the margin debate or question here. Obviously, there's a big move margins between where you were in 2Q versus what you produced in 3Q and the guide. How should we actually think about the run rate at that point, right? And we're talking about completely different levels of margin versus a blended year-end rate, right? Is 4Q the right starting point for seasonality? Or how should we think about that? Or is there some onetime adjustments that we need to think about here?

Anil Doradla

Yes. So let's look at what happened in terms of the optics of it, right? Revenue grows up by $6.1 million. The COGS is flattish. The OpEx goes down a little bit. So effectively, the revenue pickup happens because you have more billable headcount -- the other very interesting thing is look at the average revenue per person. So the margin -- the contribution margins are also picking up at an individual engineer level, right? And there are reasons for that.

On the COGS side, we made comments. There's more efficiencies due to incorporation of these tools, AI included. There is more upskilling in there. And also, there's more utilization. Of course, in Q3, we had working time. There was a benefit from there. And as you go down on the OpEx, we started our efforts late last year, and we fully realized the benefits in Q3. So there was a little bit of a compounding effect by the fact that it was Q3.

Now the question is, look at where we are on Q4, $20 million to $22 million EBITDA, right? I mean you do the math. Obviously, there is a setup point. Now historically, as we went from, call it, Q3 to Q4, somewhere in that range, maybe a step below, flattish, we've been there. But the point is we enter 2027 with a fundamental different structure. And it's a structure that involves headcount. It's a structure that involves tools and is a structure that, to some extent, involves clients that are paying us differently as you get more and more AM projects. Now as we go into Q1, you have seasonal, that's the lowest margins every quarter, right, every -- in the whole year. But I like to believe that we are entering 2027 structurally different than what we have had over the last couple of years.

Now I have to balance that with investments, as Leonard said, there's a lot of work going on robotics. There's a lot of work going on physical AI, a lot of work going on cybersecurity. So there will be some investments. But in general, the average revenue per engineer is going up, the contribution margin of engineers are going up, and we are benefiting from a little bit of a decoupling of spending money. So I think these things are all playing out. Now I thought it would show up in Q4. It just happened to show up in Q3.

Surinder Thind

Fair enough. And then where does that actually put you guys at from a utilization perspective? So are you at the target that you want to be at? Or is there more that you can squeeze here to Leonard's point about not wanting to squeeze too much?

Anil Doradla

So look, I mean, we've always had an 8 handle on utilization. We've always had that. Now if you look at some of the headcount metrics that you saw here may be a little misleading because it's not like there was a big change in our engineers, total number of engineers. What ended up happening is that we've obviously had greater conversion to billability. So in terms of our utilization, we have some movements every quarter, but we're not expecting a very fundamental change.

You will see an increased headcount as our revenue picks up because we benefited in 2026 with some of the nonengineering resource optimization. So you're seeing some of the benefits there. But in terms of engineering headcount, the utilization levels are in the same range, and we expect that to continue on.

Surinder Thind

But then I guess as a point of clarification here, the reduction in headcount that occurred, how much was billable and how much was nonbillable?

Leonard Livschitz

So we don't guide on a specific breakdown, but pretty much it was all non-engineering. So instead of talking about billable, nonbillable part, we'll talk -- I mentioned you about utilization. Majority of reduction, significant majority was purely from the non-engineer headcount. When it comes to -- I understand where you're coming from, I think the rate of speed, how we deploy people, the time to billability becomes shorter.

The onboarding become shorter. The training of people in advance because, again, the technology structure solutions become available sooner. So we're able to determine what kind of skills are most relevant to the client become faster. So it's efficiency gain, not reduction of engineers.

Surinder Thind

Fair enough. And then as a follow-up, just commentary on -- it sounds like you're gearing up to use the balance sheet at this point for M&A. It sounds like you're working on some maybe a bigger deal in the pipeline or bigger deals. How should we think about your capital allocation at this point? What's the minimum you want to keep your willingness to lever up or where you want to end up from a capital structure perspective? Or how big a deal can you do at this point? And maybe what you're interested in from a technology perspective?

Anil Doradla

So I think the way I'll answer is pretty much the way Leonard commented, but it's very important. The last 24 months has been changes in the environment, right? We don't have to go through it. And we've done a lot of work. We enter 2027 with incremental confidence. We feel really confident about where we are, how we are settling in, both from the number of countries where we're located, the skill sets, the upskilling towards SDLC. So we've gotten our house in order.

Now the mandate from the Board and Leonard to me is we need to now scale. So look, I'm not going to say what's in the pipeline or not, but the appetite right now is for larger acquisitions, which we hope to do in 2027. Now as you know, Surinder, it's not done until it's done, right? We've had many close calls, but the intent is there. And we are entertaining -- and the areas that we want to entertain are areas where we want to scale some of our skill sets, entering into some of the new ones. And we believe we are ready in 2027.

Surinder Thind

And then in terms of the skill set, is there any appetite for -- when you start to think about the evolution of some of the consulting businesses towards a more product-oriented strategy, how do you guys think about that versus adding more I'll just say, technical skills or services-oriented skills versus maybe coming in with more of a product on perhaps a particular vertical that you're interested in?

Leonard Livschitz

Yes. I think we went a little bit too far, Surinder, in terms of philosophical discussion. Obviously, not this quarter. The overall discussion in terms of what solutions and offerings can become a full-fledged product is internal discussion between the services and consultancy. And everybody said, okay, if you have some solutions, you can kind of lean upon your business better. And that's what we do. When we talk about product versus services, we're a service company, and we have a very clear path where the customers, where are the partners looking for us.

Our job today is to assist our partners, our infrastructure with building accelerators on top of what they offer. But in terms of the pure product side, you may have seen some of our, I would say, explorations. But we're very, very clear, we're not competing with our partners. We are complementing adding. Sometimes you build a solution. It's pretty much a product, but then you work with both vendors and clients to make it work that ecosystem is very strategically aligned.

So I think more discussion about the topic would be great for our Investor Day. So when we announce, I mean, we're happy to have much longer conversation. But I appreciate you thinking about it because, obviously, everybody wants to have even higher revenue per person. And from that perspective, obviously, we consider many avenues.

Operator

Ladies and gentlemen, this concludes the Q&A session for today. I will now pass it over to Leonard for closing comments.

Leonard Livschitz

Q3 2026 was a quarter of Grid Dynamics success. The results validated our vision, strategy and execution. Record revenue, record EBITDA and AI native platforms in production at some of the world's largest enterprises. Margin improvements that are structural and durable and a business increasingly anchored in verticals where AI is moving fastest. I'm confident in the fourth quarter and the momentum we are carrying into 2027. Thank you.