From EDS to Anduril: Why Services Firms Deliver Outsized Returns at Every Technology Inflection

Sixty years of evidence on the structural relationship between technological disruption, the firms that absorb it, and the investor returns that follow — with an examination of what the AI transition opens now.

Investors who funded Electronic Data Systems in 1968, IBM’s services pivot in 1993, and Accenture’s IPO in 2001 did not make three separate bets. They made the same bet, at three successive moments in time: that technological disruption always outpaces organizational absorption, that the gap between what a new technology can do and what an existing institution can deploy is durable and deep, and that the firm building a repeatable method for closing it will earn returns that no product company in the same wave can match.

This is the pattern. Every major technology transition in the past seventy years has produced it. The product companies capture the headlines and the imagination. The services companies, operating in quieter rooms, capturing the actual cash flows, compound the returns.

Every technology wave lands in the same organizational reality: the people who need to absorb it are structurally unprepared to do so. Mainframes in the 1960s, distributed computing in the 1990s, internet-scale software in the 2000s — each arrived before the institutions that needed it had built the internal capacity to deploy it. Absorption demands skills that take years to develop, political will that most bureaucracies suppress, and process redesign that existing operations resist. Enterprises cannot move fast enough on their own. The firm that moves for them — at scale, with proprietary method, under long-term contract — earns a structural position that no product company in the same wave can reach.

The product is the catalyst. The services firm is the reaction. The investor who confuses the two buys the spark instead of the fire.

Ross Perot Didn’t Build a Technology Company

In 1962, H. Ross Perot left IBM with $1,000 of personal capital and a single, precise observation: corporations across America had purchased mainframe computers they could not operate. IBM was in the business of selling hardware. Nobody was in the business of running it.

Mainframe computers in the early 1960s required operational sophistication that most corporate IT departments had never developed. The machines were expensive, temperamental, and architecturally alien to the business processes they were supposed to serve. Purchasing one did not confer capability — it conferred a new class of problem. Perot saw this gap and incorporated Electronic Data Systems to close it. EDS would not sell machines. It would absorb a client’s entire computing operation: staff it, run it, take the employees, sign long-term contracts at fixed prices, and deliver outcomes the client could not produce alone.^[1]

Perot grasped what IBM missed: the adoption gap between technology and organization was not a temporary condition that clients would resolve on their own. Enterprises would need help absorbing every successive generation of computing for decades. Five-year contracts at fixed prices embedded EDS so deeply in client operations that switching became structurally prohibitive. The client who handed EDS its data center in 1965 was not going to reclaim it in 1967. The integration ran too deep. The institutional knowledge had migrated to Perot’s organization and would not easily return.

By 1968, EDS had grown large enough and profitable enough for Perot to consider a public offering. He sold approximately seven percent of the company — 650,000 shares split evenly between himself and EDS — at $16.50 per share, raising $10 million total. The offering priced at roughly 118 times earnings, an extraordinary premium that reflected what investors understood even then: EDS was not a technology manufacturer subject to hardware cycles. It was a recurring revenue machine secured by long-term contracts, growing in lockstep with the adoption of computing across corporate America.^[1]

The market recognized this immediately. Within eighteen months of the IPO, EDS shares peaked at $160 — a 9.7× return for investors who bought the offering.^[2] The company then navigated a sharp correction as the broader market softened in 1973, but the business itself continued expanding: revenue doubled nearly every year between 1964 and 1970, and by 1979 EDS was generating $270 million annually with 8,000 employees and zero debt.

The culmination came in June 1984, when General Motors acquired EDS for $2.5 billion — at the time the largest transaction ever completed in the technology services industry.^[3] GM needed someone to coordinate and modernize its enormous, fragmented data processing infrastructure, which was costing the company $6 billion annually. Perot’s organization was the only one capable of absorbing that complexity at that scale. The $2.5 billion purchase price represented a multiplied return on the $1,000 Perot had invested twenty-two years earlier, and it positioned EDS for continued growth under GM’s ownership: revenue tripled in the first year after acquisition, headcount grew to 44,000, and the company expanded into telecommunications and factory automation.

When HP ultimately acquired EDS in 2008 for $13.9 billion,^[4] the full arc of Perot’s original insight had compounded across four decades. His personal fortune reached $4.1 billion.^[5] The return on the 1968 IPO, measured from entry to the GM acquisition price equivalents, dwarfed anything available in mainframe hardware during the same period. IBM, which sold Perot the observation that became his business, generated nothing comparable for its shareholders.

EDS proves the thesis at its origin. Perot was a salesman who had never written a line of code. What he possessed was a precise reading of organizational incapacity — and the discipline to build a company around closing it. Sixty years later, the intellectual descendants of that reading are running the same play.

The Most Important Decision Lou Gerstner Made Wasn’t Saving IBM

By 1993, IBM had lost $8 billion in two years. Its stock had collapsed from $43 in 1987 to $13 — a 69% drawdown that left the company technically solvent but strategically unmoored.^[6] The popular diagnosis was that IBM had missed the personal computer era, built its strategy around proprietary architectures that the market had abandoned, and needed to be broken apart before the damage spread. John Akers, IBM’s outgoing CEO, had already begun drafting breakup plans. Bill Gates gave IBM seven years.

Lou Gerstner arrived from RJR Nabisco in April 1993 with no technology background, no existing thesis about IBM’s industry, and no mandate beyond preventing collapse. His first decision — the one that rewrote the industry playbook — was to refuse the breakup. His second, which followed from it, was to reorganize IBM entirely around selling outcomes rather than products. Gerstner’s conviction, shaped by years of selling to large organizations at American Express and RJR Nabisco, was that IBM’s customers needed someone to make their technology work end-to-end — and that IBM, uniquely, could deliver that. The complexity of the client-server era had grown past what most IT departments could manage: integration across mainframes, distributed systems, networking infrastructure, and emerging internet protocols demanded a depth of knowledge that no internal team could sustain at acceptable cost. IBM, with forty years of accumulated expertise across every layer of enterprise computing, was the only firm on earth capable of absorbing that complexity on a client’s behalf.

IBM Global Services had been formalized as a business unit in 1991, but under Gerstner it became the company’s existential center of gravity. IBM stopped selling hardware as a standalone proposition and began selling integrated solutions — hardware, software, and people configured to produce the outcome the client needed. The company dismantled internal silos, aligned its product divisions to support services delivery, and in 2002 acquired the consulting arm of PricewaterhouseCoopers for $3.5 billion, adding management consulting depth to its technology delivery capability.^[7]

The financial results vindicated Gerstner’s bet across two decades. IBM Global Services grew from roughly $4 billion in annual revenue in the early 1990s to $50 billion by the mid-2000s. By 2007, IBM reported $98.8 billion in total revenue, with services accounting for approximately 55% of that figure. Hardware had contracted from IBM’s dominant segment to under 10% of revenue. The 2005 sale of IBM’s PC business to Lenovo — surrendering the very product category that had once seemed IBM’s future — completed the transformation.^[8]

IBM’s trajectory proved what EDS had suggested but never demonstrated at this scale: a services business operating within a technology company sustains higher gross margins, more predictable revenue, and stronger competitive positioning than the product business it replaces. Services generated structural lock-in. Once an enterprise outsourced its critical systems to IBM, extracting them meant rebuilding, over years, the institutional knowledge that had transferred to IBM’s organization. The enterprise almost never does it. IBM’s stock, which had bottomed near $13 during the crisis of 1993, climbed toward $215 by March 2013 as the services-led decade compounded.^[9] A twenty-year hold from the crisis low produced approximately 16.5×.

Gerstner’s most important decision, examined from sufficient distance, was choosing the model. By reorganizing IBM around outcome delivery rather than product sale, he demonstrated to an entire generation of technology companies that the firm absorbing client complexity — rather than selling instruments for managing it — captures a structurally superior share of value.

One Wave. Four Models of Irreplaceability.

The Internet and ERP era of the mid-1990s through the 2000s produced a more complex version of the same phenomenon. Where the mainframe era generated EDS and the client-server era generated IBM Global Services, the internet era generated four distinct services archetypes simultaneously — each capturing a different kind of irreplaceability, each compounding it into a different investor thesis.

Accenture — The Systems House

Accenture began its independent life in 1989 following a bitter split from Arthur Andersen, its parent since the 1950s when it had installed General Electric’s UNIVAC computing system.^[10] The separation forced a clarifying question: what, exactly, does this firm sell? The answer Accenture arrived at — and has refined across four platform transitions — is that it sells the capacity to absorb whatever technology wave arrives next. ERP rollouts in the 1990s, internet infrastructure builds in the early 2000s, cloud migrations in the 2010s, and now generative AI implementations in the 2020s: Accenture treats each wave not as a disruption to its model but as evidence that the model is correct.

By 2025, Accenture employed 779,000 people and generated $69.7 billion in annual revenue, sustained by acquisitions — Avanade, Droga5, Faculty, Ookla — that continually refreshed its capabilities at the frontier of each new platform shift.^[11] For investors who purchased the 2001 IPO at $14.50 per share, the stock reached an all-time high of $389 in February 2025 — a 26.8× return over twenty-four years. No hardware company that competed in the same era produced a comparable figure.

Booz Allen Hamilton — The Mission House

Booz Allen’s transformation from management consulting firm to what Bloomberg once called “the world’s most profitable spy organization” is a case study in vertical depth over horizontal breadth. Founded in 1914, Booz Allen built its early reputation serving corporations. After World War II, Washington became its primary client, and by the 2000s it had embedded itself so thoroughly in the national security apparatus — running cyber defense programs, developing AI-enabled intelligence tradecraft, staffing cleared operations that the government could not man internally — that disentangling Booz Allen from its clients had become operationally impossible.

When Booz Allen listed publicly in November 2010 at $17 per share, it offered investors access to a business whose competitive position was not replicable. You cannot build a cleared workforce of 33,000 professionals overnight. You cannot manufacture the institutional relationships with intelligence agencies that Booz Allen had spent sixty years developing. Its all-time high of approximately $180 in 2024 represented a 10.6× return from IPO investors.^[12] The position was not a services business in any ordinary sense — it was a structural monopoly on a category of human expertise that the government needed and could not produce internally.

Genpact — The Process House

GE Capital International Services launched in Gurgaon, India in 1997 as an internal back-office operation, initially designed to handle Y2K contingency work at scale.^[13] When it spun out as Genpact in 2005, with 17,000 employees and a Six Sigma playbook covering everything from credit card processing to insurance underwriting, it had already proved a different version of the services thesis: that codifying operational method — rather than staffing operational capacity — is the durable competitive position. Genpact’s advantage was never the labor cost arbitrage that characterized India’s BPO industry broadly. It was the method itself: the ability to take a client’s messiest, most document-intensive processes, reduce them to precise operational procedures, and then apply those procedures with machine-like consistency at scale.

Genpact and Cognizant, the fourth archetype — an empathy-driven services firm that won in regulated industries by building proximity and trust into its operating model — are noted here for completeness, not for the investor returns they delivered. Their multiples, while positive, do not carry the weight of the inflection-point thesis. The EDS, IBM, and Accenture returns do. The distinction matters when sizing the opportunity that AI represents.

Four Data Points. Sixty Years. One Structural Claim.

These returns share a structure. The adoption gap between what a new technology can theoretically accomplish and what the median enterprise can actually deploy persists for years — sometimes decades — before internal capability catches up. The services firm that closes that gap earns fees proportional to the gap’s size and lock-in proportional to the depth of integration. The deeper the integration, the less the client can undo it without rebuilding, over years, the institutional knowledge that has migrated to the services firm’s organization. EDS embedded its people into the client’s data centers. IBM absorbed the client’s IT architecture entirely. Accenture rewired the operating systems that client businesses ran on. The returns in each case reflected structural position — the same position a toll booth holds on the only road into a city.

The recurring revenue this creates compounds in a way that product revenue cannot. Hardware sales spike with platform introductions and crater with the next product cycle. Services contracts renew year after year while the underlying technology evolves, generating new requirements and new fees. IBM’s services backlog insulated the company from hardware downturns for a decade precisely because the revenue was structural rather than transactional. Investors who understood this held through cycles that shook out the traders.

Late acquirers learned the hard way that these positions cannot be purchased at closing. HP paid $13.9 billion for EDS in 2008 and eventually spun the business into DXC Technology. Dell paid $3.9 billion for Perot Systems in 2009 and sold it to NTT Data in 2016 for $3.05 billion — less than the acquisition price. The institutional knowledge, client relationships, and operational culture that constituted EDS’s and Perot Systems’ real value had taken decades to develop and could not be made to transfer by corporate transaction. The investors who backed those builds at year three earned returns the year-twenty acquirer could not reconstruct.

The highest multiples accrued to specialists, not generalists. EDS owned data center operations. Booz Allen owned national security tradecraft. Accenture owned platform integration at enterprise scale. Depth precedes irreplaceability, and irreplaceability sets pricing that commodity staffing cannot approach.

These returns do not reflect superior technology. EDS owned no proprietary computing architecture. IBM did not build better servers than Sun Microsystems. Accenture wrote no software that its competitors could not eventually replicate. What each firm possessed was a proprietary method for absorbing enterprise complexity — and a structural position that deepened with every engagement, because each engagement added institutional knowledge that the next engagement could draw upon. The returns are not a reward for innovation. They are a reward for irreplaceability.

The Model That Won the Last Era Is Ending in This One

In April 2025, the U.S. Department of Defense announced $5.1 billion in contract cancellations targeting Accenture’s federal subsidiary, Booz Allen Hamilton, and Deloitte. Defense Secretary Pete Hegseth’s language was unambiguous: the contracts represented work that the government’s civilian workforce could perform internally, delivered by firms charging fees the DoD described as reflecting “$500-an-hour business process consultants.”^[14] The cancellations reached into the Defense Health Agency, the Air Force, the Navy, and DARPA — covering everything from IT helpdesk services to cloud infrastructure reselling to broad management consulting engagements.

The event was significant not for its scale — $5.1 billion is a fraction of the federal consulting market — but for what it revealed about the structural vulnerability of the time-and-materials consulting model. Booz Allen’s stock fell sharply on the news. Accenture’s federal division went undefended. The firms that had built their positions on relationship depth, incumbency, and RFP execution discovered that none of those advantages protect against a client who decides the work should be done differently — or not at all.

AI is shrinking instruction sets at a rate that threatens every firm whose competitive position rests on the complexity of the work rather than the irreplaceability of the method. The firms that weather this cycle own the method, not the labor supply.

What Palmer Luckey Understood About Lattice

Anduril Industries was founded in 2017 by Palmer Luckey — the creator of Oculus VR — alongside Trae Stephens of Founders Fund, Brian Schimpf, Matt Grimm, and Joe Chen. Its founding thesis was that the defense Primes’ terms — cost-plus contracts, requirements-driven development, government-funded R&D — had made them structurally slow, and that a software-native firm building capabilities before the contracts existed could displace them on performance rather than price.^[15]

The traditional defense procurement model works as follows: the Department of Defense identifies a requirement, issues a Request for Proposal, and selects a contractor to develop the capability using government funding on a cost-plus basis. The contractor has limited incentive to build efficiently, because cost overruns are recoverable. The government has limited leverage to demand performance, because replacing the contractor mid-program would cost more than completing it. The result is a model optimized for contract capture rather than capability delivery — which is why a program like IVAS (the Army’s Integrated Visual Augmentation System) accumulated years of schedule delays and performance shortfalls under Microsoft’s stewardship before being novated to Anduril in April 2025.

Anduril built its capabilities with private capital first, demonstrated operational performance second, and won contracts on the strength of working systems rather than promising slides. At the center of this strategy sits Lattice OS — an AI-driven operating platform that fuses sensor data from autonomous systems across air, land, sea, and space into a unified command-and-control interface. Lattice was built before any specific contract required it, because Anduril’s founders understood that autonomous systems operating across multiple domains would eventually need exactly this integration layer — and that the firm owning that layer would hold a structurally irreplaceable position in every future program that deployed it.

Lattice’s strategic logic mirrors EDS’s original insight. When a military service integrates Lattice for command-and-control, adding additional Anduril hardware becomes plug-and-play. Every sensor, drone, or autonomous vehicle that connects to the platform deepens the integration and raises the switching cost. The software lock-in Anduril pursued with Lattice is the twenty-first century equivalent of Perot’s five-year data center contracts — except that software lock-in scales without the headcount required to run physical infrastructure.

The investor returns validate the model with unusual speed. Anduril’s Series A implied a valuation of approximately $1 billion in 2018. By June 2025, the Series G — led by Founders Fund’s $1 billion check, the largest single investment in the fund’s history, in a round that was eight to ten times oversubscribed — valued the company at $30.5 billion. Secondary market transactions as of December 2025 implied a valuation of $84.1 billion.^[16] Revenue doubled to approximately $1 billion in 2024, with Sacra projecting $4.3 billion by 2026 as Arsenal-1, Anduril’s hyperscale manufacturing facility in Ohio, reaches production scale.^[17]

The IVAS program crystallizes why the model compounds. Microsoft built the system for years under a $22 billion program, struggling with software release cycles of approximately 180 days. After Anduril took over and integrated Lattice into the program’s architecture, release cycles contracted to 18 hours.^[19] The architectural explanation is straightforward: a software-native firm with a platform built for continuous iteration versus a hardware and cloud company managing a program built for a procurement cycle. Anduril won IVAS by having already built the thing the program needed, not by underbidding the incumbent.

That sequencing — build first, win contracts second — is the structural inversion that separates Anduril from every legacy defense prime. Lockheed Martin, Boeing, Northrop Grumman, and Raytheon are products of the cost-plus procurement era: their organizations, incentive structures, and financial models presuppose contracts awarded before capabilities are built. Anduril makes existing capability so compelling that contracts follow the technology rather than fund it. The prime cannot adopt this model without dismantling the incentive structures that sustain its margins. Anduril has no such constraint.

An AI-Native Operating Partner That Arrives With a Thesis

A 2024 Boston Consulting Group study found that 74% of enterprises report significant barriers to scaling artificial intelligence — barriers attributed primarily to internal skill gaps and the absence of clear implementation frameworks.^[20] The figure understates the problem. Frontier models improve faster than enterprises can hire, train, and govern the people needed to deploy them. Every quarter that passes without a deployment framework widens the gap between what an organization’s competitors might do with AI and what that organization can actually execute. The gap is not closing on its own.

Enterprises in 1962 had purchased mainframes they could not operate. Enterprises in 2026 hold access to frontier model capabilities they cannot govern, integrate into production workflows, or deploy at the speed that competitive pressure demands. The operational problem is the same. The technology that created it is different. And the services opportunity that follows from it is larger, because AI touches every enterprise function simultaneously — underwriting, claims, sales, legal review, product development, customer service — where prior waves touched one department at a time.

Caerus Alpha was founded to close that gap as an AI-native operating partner — a firm that arrives with a thesis and builds alongside the enterprise until the capability is embedded and compounding. A firm that delivers a strategy deck and bills for the engagement has created no durable asset. A firm that installs proprietary IP into the client’s operational stack, trains the client’s teams on its use, and governs the capability through a framework that deepens with every deployment has created something the client cannot easily replace or in-source.

The proprietary framework Caerus Alpha brings is Teleological Machines — an AI orchestration architecture that governs how enterprise agent fleets pursue goals, manage decisions, and escalate when conditions exceed their authority.^[21] At its core, TM addresses the problem that makes enterprise AI deployment genuinely hard: not model capability, but governance under uncertainty. An agent fleet operating across financial services workflows, healthcare operations, or industrial processes must know what it is trying to accomplish at every level of the decision stack — which actions it can take autonomously, which require human review, and under what conditions its confidence in its own reasoning should trigger escalation. TM encodes goal-seeking behavior into the orchestration layer itself, so that agents operating across the enterprise pursue defined outcomes within governed boundaries rather than executing disconnected tasks. That architecture — goal-pursuit under governance rather than instruction execution — is what separates deployable enterprise AI from the experimental variety.

The functional analog to Lattice in enterprise AI is Teleological Machines. Lattice OS fuses sensor data from autonomous defense systems into a unified command-and-control interface. TM fuses decision signals from enterprise agent fleets into a governed operational framework that knows what it is trying to accomplish and how to escalate when it encounters conditions outside its authority. The parallel runs deeper than architecture: Lattice was built before Anduril had contracts requiring it; TM was built before the engagements that deploy it. Each deployment sharpens the framework’s domain specificity, reducing the cost of the next deployment and raising the switching cost for any client that has integrated it into production. The IP compounds across engagements. Billable hours do not.

The Enterprise Innovation Group — Caerus Alpha’s deployment arm — operates on the same sequencing Anduril proved in defense. EIG selects engagements where the TM framework can deploy at sufficient depth to generate proprietary domain intelligence and where the client’s operational context will advance the framework’s capabilities in ways that benefit subsequent deployments. Each engagement sharpens the framework. Each deployment makes the next one cheaper to execute and harder for incumbents to displace.

The consulting prime’s business model requires the client to know what to ask for. Caerus Alpha arrives before the question is fully formed — with a thesis about what the organization needs and a framework already built to deliver it.

The defense Primes lost ground to Anduril because their entire organizational model presupposes a world where the government defines requirements and the contractor executes them. Caerus Alpha’s counterpart in enterprise AI is the management consulting firm whose model presupposes a world where clients define the problem and the firm delivers the slide deck. When the Pentagon cut $5.1 billion in consulting contracts in April 2025, it signaled that the advice-and-depart model had reached the end of its useful life in a budget environment that demands operational outcomes.

Enterprise buyers are reaching the same conclusion. The firms that survive the AI transition are the ones that embedded operational AI capability into production workflows, governed it through frameworks that deepened with use, and built positions their clients could not replicate without years of equivalent investment. Strategy documents are evidence of intent. Embedded frameworks are evidence of position.

The Sand Does Not Stop

In 1983, a sandstorm rolled through a factory in Delhi and settled on the first indigenous microcomputer India had produced at scale. The company in that factory — Hindustan Computers Limited — chose to reinvent rather than defend. Its R&D arm spun out as HCL Technologies, eventually designing India’s first electronic stock exchange and building the software capability that would employ hundreds of thousands of engineers worldwide. The sandstorm that could have buried the company instead stripped away the contingent parts and left visible what was irreplaceable.

Commoditization does this. Every wave of technological disruption erodes the position that the previous wave created and exposes the underlying capability that survives it. Perot’s irreplaceable capability was operational discipline — the capacity to run a client’s infrastructure with more rigor and continuity than the client could sustain. Gerstner’s was end-to-end integration knowledge accumulated across forty years of enterprise computing. Anduril’s is a software platform that makes autonomous systems interoperable across every domain and locks in every military service that deploys it. Caerus Alpha’s is the orchestration framework that governs how enterprise agent fleets pursue goals — the governed intelligence layer that turns experimental AI into operational AI. Each of these positions took years to build. None can be rapidly replicated.

EDS IPO investors who bought at $16.50 in September 1968 and held through the early 1970 peak earned 9.7× in eighteen months — while Wall Street was still trying to determine what a data processing center was worth. IBM investors who bought at $13 in 1993, when the company was openly expected to collapse, earned 16.5× over twenty years as the services pivot compounded through three successive technology waves. Accenture IPO investors who paid $14.50 in 2001 have watched the stock compound to 26.8× as platform complexity has persistently outrun enterprise capacity to manage it. Each of these investors held the same structural conviction at the same structural moment: the gap between technological capability and organizational absorption was wide, durable, and addressable by a firm with the right method.

That moment is here again. The gap between frontier AI capability and enterprise deployment capacity is measurably larger than any prior technology adoption gap on record. It widens each quarter. The services response to it — the firms that will close it at scale, with proprietary method, at the speed the market demands — are being assembled now. The window at each prior inflection point closed faster than the investors who arrived late could act. The returns that accrued to early entry cannot be reconstructed at later entry prices.

The companies worth backing are the ones already building new tools while the storm is still raging.