Every few years, enterprise IT gets handed a new three-letter acronym that quietly redraws who gets paid for what. This time it’s FDE – Forward Deployed Engineer. In roughly six months, AWS, Microsoft, Google Cloud, OpenAI, and Anthropic have all stood up their own FDE organizations, backed by something like $5–7 billion in combined committed capital. For an industry that has spent thirty years selling the labor that makes platforms actually work inside messy enterprises, that’s not a footnote. It’s a direct move into the System Integrator’s core business.
So what actually is an FDE, why are hyperscalers and model labs suddenly building armies of them, and what should any of this mean to someone running or advising an IT services business? Below, I’ll walk through where this creates real opportunity for SIs, and where, if we’re honest, it quietly erodes the economics that have funded the industry for decades.
What is a Forward Deployed Engineer, really?
Palantir coined the term in the early 2010s. Its government and intelligence customers often couldn’t articulate requirements up front, the work was too sensitive, too undocumented, too tangled up in institutional tribal knowledge to spec out remotely. So instead of gathering requirements from a distance and shipping a document, Palantir embedded its own senior engineers physically inside the customer’s environment. They sat in on standups, wrote production code against the client’s actual data, and stayed until the thing worked. Palantir called them “Deltas,” and by 2016 the company employed more FDEs than conventional product engineers. The role was never a delivery afterthought, it was the primary way Palantir figured out what its product needed to become.
That distinction is worth holding onto, because it’s what separates an FDE from every services role that already existed. A Solutions Architect designs a system and hands off a document. A Consultant advises and plans, usually while a separate delivery team does the building. A traditional SI delivery engineer builds against a signed statement of work, typically offshore-leveraged, priced by effort, and structurally walled off from the vendor’s own product roadmap. An FDE, by contrast, writes and ships production code inside the customer’s own environment, gets measured on operational outcomes rather than billable hours, and critically feeds what they learn in the field back into the core product itself.
Then there’s the version of this role showing up inside IT services companies that aren’t themselves ISVs or platform owners, and it’s worth being precise about it, because SIs are already using the FDE label pretty loosely. An SI-side FDE is a senior, product-fluent engineer who embeds with a client to build and ship working software against someone else’s platform like Claude, Bedrock, Azure AI Foundry, Gemini Enterprise, an ERP, whatever the account happens to run on rather than the SI’s own IP. They don’t have a product roadmap to feed insights into, the way a Palantir or OpenAI FDE does. What they can offer instead is something the platform vendor’s own FDE structurally cannot: neutrality across platforms, account relationships built up over years, accumulated industry and regulatory context, and a willingness to stay accountable for a system long after the vendor’s engineers have rotated on to the next flagship logo. That’s the SI’s real moat. It’s just narrower than it used to be.
The land grab, with numbers
The pace here is what should get a services executive’s attention. In roughly a thirteen-week window in early-to-mid 2026, AWS committed $1 billion to a new Forward Deployed Engineering unit, which it says will grow to thousands of engineers working in pods of five or six per customer, alongside AI agents. Microsoft launched Microsoft Frontier Company, a $2.5 billion initiative it describes as going “beyond” the standard FDE model while explicitly saying it’ll lean on SI partners to help it scale. Google Cloud went on a public hiring push for dozens of forward-deployed roles, with CEO Thomas Kurian framing it as scaling “customer AI transformation,” and in the same breath promising to extend engineering support to its largest SI partners rather than simply competing with them. OpenAI launched the OpenAI Deployment Company, capitalized at more than $4 billion with TPG, Advent International, Bain Capital, and Brookfield as backers, and acquired a European deployment-engineering firm to get a roughly 150-person delivery team out of the gate. At the same time, OpenAI is running a “Frontier Alliances” program that certifies McKinsey, BCG, Accenture, and Capgemini as delivery partners, making those same four firms both OpenAI’s channel and, in the very same accounts, its competitor. And Anthropic formed a joint venture reportedly valued above $1.5 billion, with Blackstone, Hellman & Friedman, and Goldman Sachs, to embed Claude-focused engineering resources inside a standalone enterprise services company aimed at the mid-market.
Why now, specifically? Because the adoption gap has become impossible to ignore. By the end of 2025, close to nine in ten companies had deployed AI in at least one business function, according to McKinsey and yet 94% reported no significant, measurable financial benefit from that spend. Hyperscalers and labs that have collectively committed hundreds of billions of dollars to AI infrastructure need that spend to convert into visible enterprise value, and fast. Handing a customer a model API and a support ticket queue clearly isn’t converting fast enough. Sending in your own senior engineers to make it work does.
Where this genuinely helps Sis
It would be a mistake to read this purely as an attack. There are real openings here for services firms, and the more capable ones are already going after them.
Certified delivery capacity, for one, is turning into a kind of currency, and SIs are racing to build it up. Tavant rolled out W2W CoPilot to its entire workforce and has been training people across AI models more broadly. Accenture has put roughly 30,000 practitioners through a dedicated Anthropic business group focused on regulated industries. Deloitte has rolled Claude out across its full 470,000-person global workforce. Cognizant has done the same for 350,000 employees and is layering vertical agent platforms on top. Infosys has folded Claude into its Topaz platform and stood up a dedicated center of excellence. Add it up across the top firms, and well over a million practitioners are now nominally certified on at least one frontier model. That’s exactly the kind of trained, billable capacity hyperscalers and labs need if they want to reach mid-market and regulated accounts that their own FDE units simply aren’t resourced to touch.
Hyperscalers, for their part, are explicitly positioning FDEs as complementary rather than a replacement, at least for now. Both Microsoft’s Frontier Company and Google Cloud’s FDE expansion name SI partners as a way to scale, not as casualties. AWS has framed its FDE pods as leaving behind “self-sufficient” customer teams rather than staying on as permanent embedded staff which implies someone still needs to run, extend, and govern what the FDE built after they rotate out. That governance and change-management layer, the multi-year operational reality of it, is squarely SI territory, and not something a hyperscaler particularly wants to staff at scale.
And the overall pie is still growing. Gartner puts total 2026 IT services spending into application and infrastructure implementation, managed services, and IaaS – combined budget of above $1.8 trillion, growing at double-digit rates, with AI-related services spend up roughly 40% year over year. A shrinking share of a fast-growing market can still add up to growing absolute revenue, for firms that reposition quickly enough. The risk is concentrated among the firms that don’t.
Where it genuinely hurts
The more honest reading is that the FDE wave squeezes the SI business model from both directions at once and the squeeze lands hardest on exactly the work that has historically funded the industry: custom implementation, integration, and modernization billed by the hour.
From the top, the highest-margin, most strategic engagements are now directly reachable by the vendor itself. An FDE working for the OpenAI Deployment Company, or an AWS FDE pod, can route a customer’s requirement straight into the underlying product roadmap, something no third-party SI engineer can ever do. Model labs and hyperscalers are deliberately seeding these units with marquee accounts and private-equity portfolio companies, which happen to be historically the most profitable logo wins for the top-tier global consultancies.
From the bottom, hyperscaler-built automation is eating into the labor-arbitrage layer that funded offshore delivery models for two decades. Tools like AWS Transform now handle code analysis, dependency mapping, refactoring, and test generation. This is the exact modernization work Indian IT majors have billed by the hour for years and offer it at little to no incremental charge, because the hyperscaler makes its money on the destination (cloud consumption) rather than the journey (the labor to get there). The effect already shows up in the numbers: the five largest Indian IT services firms cut a combined roughly 7,000 jobs in FY26, reversing a net addition of nearly 13,000 the year before.
And increasingly, customers can skip the services layer altogether. Global Capability Centers, where enterprises’ own captive engineering hubs, of which more than 1,700 now operate in India alone and are starting to function less like back-office cost centers and more like internal AI labs, deploying Claude or a hyperscaler’s models directly through cloud marketplaces and building their own integrations without ever engaging a third-party integrator. That’s demand disappearing before it reaches an SI’s pipeline at all, not demand being won by a competitor.
Where the overlap actually plays out
The clearest way to think about where FDEs and SIs will collide, and where they mostly won’t, is as a market sorting into three rough tiers rather than one single battlefield.
At the top sit the premium and marquee accounts, where lab-owned and hyperscaler-owned FDE units have direct model or infrastructure access and real roadmap influence. This is where global SIs are most exposed, particularly on the high-margin, greenfield AI transformation work at flagship logos.
Below that is the global consultancy tier: multi-platform certified firms bringing change management, multi-vendor neutrality, and regulatory depth that no single-platform FDE org can replicate. This tier is best positioned to absorb hyperscaler-funded delivery support and turn it into scaled engagements, provided they actually build proprietary IP rather than just reselling access.
And then there’s the offshore-led, cost- and scale-competitive delivery tier, which faces the most direct structural pressure of the three, as hyperscaler-automated modernization tools erode the labor-hour economics this tier has always depended on. That’s pushing firms toward governance, domain-specific agents, and outcome-based pricing as pretty much the only durable differentiator left.
In practice, most large transformation programs already pull from more than one tier at once; a model lab or hyperscaler FDE for the core AI build, a global SI for governance and change management, often a specialist boutique for a narrow domain gap. It’s not a clean handoff between these players; it’s concurrent, sometimes uncomfortable co-existence in the same account, with the vendor now sitting at the table as both partner and competitor.
Hyperscaler and lab-owned FDE organizations aren’t going to replace the SI industry, the scale, coverage, and account depth needed to serve the global mid-market simply isn’t something AWS, Microsoft, Google, OpenAI, or Anthropic can or wants to staff directly. But the FDE wave does permanently narrow what SIs can charge a premium for. The old annuity of billing by the hour to make someone else’s platform work is the exact function platform owners are now internalizing, because they have both the capital and the incentive to do it themselves. What’s left for services firms to own is everything the FDE model structurally can’t deliver: multi-year accountability after the vendor’s engineers rotate out, governance across a whole portfolio of AI systems rather than one deployment, regulatory and domain context that spans platforms, and the trust relationship that survives whichever model happens to be state-of-the-art this quarter. Firms that build real proprietary IP and outcome-based commercial models around that layer will grow through this shift. Firms still selling generic implementation hours are competing directly against their own largest partners’ balance sheets and that’s a fight the balance sheets are built to win.
None of that is a two-year problem. It’s a next-two-quarters problem. The SIs that come out ahead won’t be the ones that wait for their hyperscaler or lab partner to define the FDE relationship for them — they’ll be the ones that move first on three things: naming, right now, which two or three accounts justify an embedded, outcome-priced delivery model instead of a time-and-materials one; converting certified headcount into a packaged, multi-platform governance offering before a hyperscaler’s own FDE unit gets there first; and deciding, deliberately, which client relationships are worth defending with proprietary IP versus which ones are already lost to the vendor’s balance sheet. The firms that make those three calls this year will still be setting the terms of the conversation when the next acronym shows up. The ones that don’t will be negotiating from inside someone else’s pod.