Enterprise Redefinition Observed A Multiple-Case Analysis of Eighteen Enterprises in the Age of AI Naoki Kadowaki Founder & CEO, VURA Capital Innovation Holdings Inc. · Lecturer, Keio Business School ORCID 0009-0009-5295-1506 This version: August 2026 · Working Paper Version 1.2 Abstract This paper provides the first structured empirical application of the Enterprise Re‐ definition framework (Kadowaki, 2026b), which conceptualizes transformation in the age of AI as the continuous, coordinated redesign of five dimensions of the enterprise —Purpose, Business, Organization, Capital, and Leadership. Using a multiple-case design, eighteen enterprises—NVIDIA, Microsoft, Amazon, Apple, OpenAI, Tesla, Google, Meta, Netflix, Sony, IBM, TSMC, ASML, Arm, Costco, BYD, SpaceX, and An‐ thropic—are coded on a common instrument drawn from public information as of Au‐ gust 2026: which dimensions moved, at what maturity level, where in the Future Value Chain value originated, and through which redefinition patterns. Eight crosscase findings emerge. Core purpose was replaced in none of the eighteen cases; re‐ definition concentrates in Business (16/18) and Capital (15/18), which move together in fourteen cases; observed value originates upstream of Creation in seventeen cases and at Enterprise Value in none; patterns of redefinition never occur alone; cross-di‐ mensional coherence, not frequency of change, separates sustained redefiners; thirty percent of dimension-level assessments are unobservable from public information, concentrated in Organization and Leadership; every sustained redefinition rests on an institutionalized mechanism that secures time; and each enterprise’s strength and fragility derive from the same structure. To address survivorship bias, a contrast set of five failed or forfeited redefinitions—Kodak, Nokia, BlackBerry, Yahoo, and Sears— is coded on the same instrument from the public historical record: learning was fre‐ quently present, even pioneering, while the connection from learning through re‐ definition to capital allocation broke, purpose churned rather than anchored, and no time-securing mechanism held—an inverse signature consistent with the findings from the primary sample. In maturity terms the separation is numeric: no rated con‐ trast-set cell reaches Level 4 of the Enterprise Redefinition Maturity Model, and no rated primary-sample cell sits at Level 1. The findings motivate a three-layer refine‐ ment of purpose stability, an observation-asymmetry thesis with consequences for how capital markets price redefinition capability, and nine new research propositions 1 (P12–P20) extending the eleven advanced in the framework paper. Implications for executives, boards, and investors are developed. Keywords: Enterprise Redefinition; Future Value; Multiple-Case Study; Corporate Purpose; Capital Allocation; Organizational Transformation; Artificial Intelligence; Maturity Model JEL Classification: M10, M14, L21, O31, O33, G34, D23 This is a working paper presenting an empirical application of a conceptual framework in development. Comments are welcome. The two companion working papers are “Future Value Theory: A Management Framework for Enterprise, Capital, and Society in the Age of AI” (Kadowaki, 2026a), available at SSRN (https://ssrn.com/abstract=7120980) and Zenodo (https://doi.org/10.5281/zenodo.21255662), and “Enterprise Redefinition: Toward an Enterprise Evolution Theory for the Age of AI” (Kadowaki, 2026b), available at SSRN (https://ssrn.com/abstract=7210118) and Zenodo. The case evidence analyzed here is documented in Volumes IX and X of the AI Management Library (Kadowaki, 2026c, 2026d). 2 1 Introduction Two companion working papers have proposed a theory and a framework. Future Value Theory (Kadowaki, 2026a) argues that the primary objective of the enter‐ prise in the age of AI is not to maximize present enterprise value but to build the capacity to create value that does not yet exist, and models value creation as a chain—Purpose → Learning → Redefinition → Creation → Enterprise Value—in which enterprise value appears only as the final outcome. Enterprise Redefinition (Kadowaki, 2026b) develops the middle of that chain into a framework: transforma‐ tion in the age of AI is not an episodic project but a continuous capability, exer‐ cised across five interdependent dimensions—Purpose, Business, Organization, Capital, and Leadership—and maturing through five levels from the Reactive En‐ terprise to the Future Value Enterprise. The framework paper closes with a candid admission and an agenda: “the proposed constructs have not yet been empirically validated through quantitative or longitudinal research,” and comparative studies across firms and industries are named among the necessary next steps (Kadowaki, 2026b, §11.1, §9.9). This paper takes the first of those steps. It applies the Enterprise Redefinition framework, as published, to eighteen enterprises—ten treated in Volume IX of the AI Management Library (NVIDIA, Microsoft, Amazon, Apple, OpenAI, Tesla, Google, Meta, Netflix, Sony) and eight in Volume X (IBM, TSMC, ASML, Arm, Costco, BYD, SpaceX, Anthropic)—and asks what the framework reveals when held against evidence, and what the evidence reveals about the framework. The ques‐ tion is deliberately double-edged. A framework applied to cases it cannot fail is not being tested; it is being illustrated. The design therefore fixes the coding instru‐ ment in advance, applies it uniformly, records where public information does not permit a judgment, and reports the cells the framework cannot fill with the same prominence as the cells it can. Three features distinguish the study from conventional collections of success stor‐ ies. First, the sampling is theoretical rather than celebratory: the eighteen enter‐ prises differ in age (from an incumbent with more than a century of history to firms younger than fifteen years), in ownership (listed, newly listed, private, founder-con‐ trolled, trust-governed), in geography (the United States, Europe, and East Asia), and in industry position (component suppliers, platform operators, retailers, manu‐ facturers, research organizations)—variation chosen to expose the framework to conditions under which its claims could break. Second, the unit of analysis is the redefinition, not the company: for each case the coding asks what was rewritten, when, at which dimension, and at what price, rather than whether the company is admirable. Third, failure and reversal are inside the frame: the sample includes a declared redefinition that did not occur where declared (Meta), a decade in which a capable enterprise inverted the value chain and stalled (IBM, 2010–2017), a re‐ definition currently running ahead of its own creation capacity (Tesla), and a react‐ Enterprise Redefinition Observed VURA Working Paper Series 3 ive turn made only after members were lost (Netflix, 2022). Beyond these interior negatives, the study adds an exterior one: a contrast set of five enterprises—Kodak, Nokia, BlackBerry, Yahoo, and Sears—whose redefinitions failed or were forfeited, coded on the same four axes (Section 6), so that the propositions rest on the difference between the two groups rather than on the successes alone. The analysis yields eight cross-case findings, developed in Section 5. Some corrob‐ orate the framework in a form that could have falsified it: across ninety dimensionlevel observations, no case locates the origin of value at Enterprise Value, as the Future Value Chain's ordering requires; no case replaced its core purpose, as the framework's stability thesis predicts. Others extend the framework: the two-layer account of purpose (Core Purpose versus Purpose Expression) proves too coarse for the data and is refined into three layers; the concentration of unobservable cells in Organization and Leadership motivates an observation-asymmetry thesis— the dimensions capital markets can price are not the dimensions that constrain re‐ definition capability—with consequences for investors and for disclosure design. Still others discipline the framework's practical use: simultaneous movement of all five dimensions, far from indicating superior capability, appears in only two cases and reads in both as a risk state; and every sustained redefinition in the sample rests on an identifiable institutional mechanism that secured time, supporting the framework's claim that redefinition capability without a time horizon produces nothing. The paper proceeds as follows. Section 2 summarizes the theoretical background and the analytical instruments. Section 3 sets out the research design, the coding protocol, and its observability rule. Section 4 introduces the eighteen cases in brief. Section 5 develops the eight cross-case findings. Section 6 codes the contrast set of five failed or forfeited redefinitions and derives the failure signature. Section 7 states nine research propositions, numbered P12–P20 in continuation of the framework paper's P1–P11. Section 8 discusses implications for theory, for execut‐ ives and boards, and for investors. Section 9 records limitations, and Section 10 concludes. 2 Theoretical Background 2.1 Future Value Theory and the Future Value Chain Future Value Theory (Kadowaki, 2026a) reframes profit, share price, and enter‐ prise value as outcomes of a prior capacity: the capacity to create value that does not yet exist. The theory's process claim is the Future Value Chain, Purpose → Learning → Redefinition → Creation → Enterprise Value in which purpose identifies the challenge worth addressing, learning converts an unformed territory into organizational knowledge, redefinition rewrites what the Enterprise Redefinition Observed VURA Working Paper Series 4 enterprise is in light of that knowledge, creation brings new markets and offerings into existence, and enterprise value arrives last, as the market's recognition of the sequence. The chain is a causal ordering, and orderings of this kind are empirically exposed: if enterprises routinely generated durable value by beginning from enter‐ prise-value targets, or if observed value creation clustered at the end of the chain rather than upstream, the ordering would fail. Section 5.3 confronts the chain with the eighteen cases on exactly these terms. 2.2 The Enterprise Redefinition Framework The framework paper (Kadowaki, 2026b) defines Enterprise Redefinition as “the continuous process through which organizations re-examine and, when necessary, redesign their Purpose, Business, Organization, Capital, and Leadership in order to create sustainable Future Value.” Three of its constructs are used directly in this study. The five dimensions. Purpose is the reason the enterprise exists and the future it intends to create (Drucker, 1954; Collins & Porras, 1994; Gartenberg, Prat, & Sera‐ feim, 2019; George, Haas, McGahan, Schillebeeckx, & Tracey, 2023); the frame‐ work distinguishes a relatively enduring Core Purpose from its evolving Purpose Expression. Business is the design of how value is created, delivered, and cap‐ tured. Organization is the enterprise understood as an adaptive human–AI valuecreation system, extending beyond the legal boundary of the firm. Capital extends beyond finance to knowledge, data, trust, relationships, and managerial attention, with allocation oriented to future value rather than historical assets (Ocasio, 1997; Bower, 1970). Leadership is the integrating capability that connects the other four, shifting from directing execution to designing the future. The maturity model. The Enterprise Redefinition Maturity Model (ERMM) distin‐ guishes five levels: Level 1, the Reactive Enterprise, which changes only after per‐ formance deteriorates; Level 2, the Improvement Enterprise, which pursues opera‐ tional excellence without questioning its business model; Level 3, the Transforma‐ tion Enterprise, which redesigns itself periodically, as a project; Level 4, the Con‐ tinuous Redefinition Enterprise, in which redesign is embedded in normal manage‐ ment processes and begins before external disruption requires it; and Level 5, the Future Value Enterprise, which actively shapes the industries and assumptions within which it will compete. The framework insists that levels are assessed per di‐ mension, that progression is non-linear, that the appropriate level differs by in‐ dustry, and that treating Level 5 as a universal target is a misuse of the model. This study inherits all four qualifications and adds a fifth of its own: a dimension for which public information provides no basis for judgment is coded as unobservable, not estimated. The capability model. Enterprise Redefinition Capability (ERC) comprises six com‐ ponents—Strategic Intelligence, Learning Capability, Design Capability, Capital Reallocation Capability, Leadership Capability, and AI Collaboration Capability—po‐ sitioned as a meta-capability above operational and dynamic capabilities (Teece, Enterprise Redefinition Observed VURA Working Paper Series 5 Pisano, & Shuen, 1997; Teece, 2007; Eisenhardt & Martin, 2000; Winter, 2003). The capability construct is not directly measurable from public information; where this study refers to it, the reference is to observable behavior consistent or inconsistent with it, a distinction maintained throughout. 2.3 Analytical Instruments from the Practitioner Series Two further instruments originate in the practitioner volumes of the series rather than in the framework paper, and are adopted here because the case evidence was compiled with them. The first is a typology of six redefinition patterns (Kadowaki, 2026d, drawing on the case analyses of the series): the Layer Shift pattern, in which the enterprise moves the layer of the value structure at which it operates (component to platform, product to infrastructure); the Capability Redeployment pattern, in which a capability built for one purpose is redeployed as a business in itself; the Ownership-to-Usage pattern, in which the unit of sale moves from owner‐ ship of an object to continuing use of a relationship; the Customer Redefinition pat‐ tern, in which the enterprise changes who counts as its customer; the Societal Challenge pattern, in which a societal problem is converted into the entry point of a market; and the Medium pattern, in which the form through which value is de‐ livered is itself rewritten. The second instrument is the observability convention of Volumes IX and X: company analyses rest on primary sources—financial disclos‐ ures, official announcements, regulatory filings—confirmed as of August 2026; fig‐ ures that could not be confirmed are not stated; and matters of internal decisionmaking are marked as outside what public information can establish (Kadowaki, 2026c, 2026d). 2.4 The Empirical Gap The framework joins a family of theories—dynamic capabilities (Teece et al., 1997; Helfat & Peteraf, 2015), organizational ambidexterity (O’Reilly & Tushman, 2008, 2013; March, 1991), business model innovation (Foss & Saebi, 2017; Zott & Amit, 2010), digital transformation (Vial, 2019; Hanelt, Bohnsack, Marz, & Antunes Marante, 2021), and the emerging literature on AI and competitive advantage (Iansiti & Lakhani, 2020; Krakowski, Luger, & Raisch, 2023; Raisch & Krakowski, 2021), all standing on the strategy and evolutionary canon (Porter, 1980; Penrose, 1959; Chandler, 1962; Nelson & Winter, 1982; Freeman, 1984; Christensen, 1997) and increasingly informed by experimental evidence on AI and knowledge work (Brynjolfsson, Li, & Raymond, 2025; Dell’Acqua et al., 2023; Noy & Zhang, 2023)— whose common weakness is well documented: constructs multiply faster than evid‐ ence. The framework paper is explicit that its contribution is theory development, not hypothesis testing (Kadowaki, 2026b, §11.1). What a conceptual framework at this stage needs is not yet large-sample validation, for which no measurement scales exist, but a structured confrontation with cases selected to strain it—the role that multiple-case designs serve in theory elaboration (Eisenhardt, 1989; Eis‐ enhardt & Graebner, 2007; Siggelkow, 2007; Yin, 2018). That is the task taken up here. Enterprise Redefinition Observed VURA Working Paper Series 6 3 Research Design 3.1 Method and Case Selection The study uses a multiple-case design in the theory-elaboration tradition (Eisen‐ hardt, 1989; Yin, 2018): the framework supplies constructs and predicted relation‐ ships; the cases supply variation against which those constructs are coded; and the analysis reports both replication (findings recurring across cases under different conditions) and refinement (points at which the framework's categories prove too coarse for the data). Eighteen cases exceed the four-to-ten range conventional for inductive theory building (Eisenhardt, 1989), deliberately: the purpose here is not to induce a new framework from rich narratives but to apply an existing instru‐ ment uniformly and count what it finds, a design in which breadth converts into falsification opportunities—ninety dimension-level cells, any distribution of which could have contradicted the framework's claims. The sample comprises the eighteen enterprises analyzed in Volumes IX and X of the AI Management Library (Kadowaki, 2026c, 2026d): NVIDIA, Microsoft, Amazon, Apple, OpenAI, Tesla, Google (Alphabet), Meta, Netflix, and Sony (Vol. IX); IBM, TSMC, ASML, Arm, Costco, BYD, SpaceX, and Anthropic (Vol. X). The selec‐ tion is theoretical sampling under three criteria. First, variation on the frame‐ work's own constructs: the sample includes candidate cases for every maturity level from 2 to 5, all six redefinition patterns, and every position on the Future Value Chain. Second, variation on rival explanations: if redefinition capability were merely a synonym for the platform economics of large American technology firms, a warehouse retailer (Costco), a lithography supplier (ASML), a contract manufac‐ turer (TSMC), a Chinese battery-to-vehicle integrator (BYD), and a Japanese con‐ glomerate (Sony) should code differently in kind, not merely in degree. Third, in‐ clusion of negative and ambiguous space: declared-but-undelivered redefinition (Meta), inverted-chain stall (IBM's 2010s), reactive redefinition (Netflix's 2022 turn), and redefinition whose verification is structurally impossible from outside (SpaceX and Anthropic as recently private or private companies). 3.2 Data The evidentiary base is the case documentation of Volumes IX and X, which was compiled under a stated protocol: every company analysis confirms primary sources—financial disclosures, official announcements, and filings with securities regulators—as of August 2026; figures, shares, and rankings that could not be con‐ firmed are not stated; and internal decision-making is not asserted (Kadowaki, 2026c, 2026d). Facts cited in this paper carry their source and as-of point in the form used there (for example, “earnings release of February 25, 2026”). Using an existing case corpus has a cost and a benefit that should be stated plainly. The cost is that author and analyst coincide: the corpus and this paper share an author, and Enterprise Redefinition Observed VURA Working Paper Series 7 the coding cannot claim inter-rater independence (Section 9). The benefit is sym‐ metry: all eighteen cases were documented on the same axes, over the same win‐ dow, under the same evidentiary discipline, which is precisely what uniform coding requires and what post-hoc assemblages of heterogeneous case studies lack. 3.3 Coding Protocol Each case is coded on four axes, fixed before analysis and identical across cases (Figure 1). Axis 1 — Dimensions moved. For each of the five dimensions: did the case evidence show the dimension being redefined (moved), moving only in expres‐ sion, scope, or field of application while its core held (expression-only), not mov‐ ing, or providing no basis for judgment (unobservable)? For Purpose, a further three-layer sub-code is applied, developed in Section 5.1: replaced, re-articu‐ lated (the core newly put into language or formally re-described, with year), or re-expressed (only expression, reach, or application changed). Axis 2 — Maturity. The ERMM level per dimension, following the assessments documented in the case volumes (Kadowaki, 2026d, Appendix C), including their withheld judgments. A withheld cell is recorded as unobservable and counted; it is never interpolated. Axis 3 — Position on the Future Value Chain. The stage, or boundary between stages, at which the case evidence locates the origination of the value later recognized as enterprise value. Axis 4 — Patterns. Which of the six redefinition patterns the case exhibits, in‐ cluding multiplicity; for declared-versus-actual divergence (Meta), declared and actual patterns are coded separately. Three conventions govern the coding. The instrument rule: Axes 2, 3, and 4 follow the consolidated cross-case instrument of the source corpus (Kadowaki, 2026d, Ap‐ pendix C), which was compiled across all eighteen cases on uniform definitions; in‐ dividual chapter narratives occasionally name a subset of patterns, or place the ori‐ gin of value with wider bounds, than the consolidated instrument records, and where the two differ the instrument governs, with material differences noted in the profiles of Section 4. The observability rule: public information systematically overdocuments some dimensions (Business, Capital) relative to others (Organization, Leadership); rather than letting the framework's categories manufacture false completeness, every cell that cannot be grounded in a primary source is left open, and the distribution of open cells is itself treated as a finding (Section 5.6). The counting rule: every numerical claim in Sections 5 through 7 is computed mechan‐ ically from the coding table reproduced in Appendix A, so that any reader can re‐ count. One cell-level discrepancy against the source corpus is noted: Volume X, Ap‐ pendix C states 26 withheld cells in its maturity table; recounting under this pa‐ per's protocol yields 27, the difference lying in one Business-dimension cell (ASML, where the chapter text declines a judgment this paper's recount preserves as with‐ • • • • Enterprise Redefinition Observed VURA Working Paper Series 8 held). The discrepancy is retained rather than reconciled, and does not affect any finding. Figure 1. The four analytical axes applied uniformly to the eighteen cases. 3.4 The Contrast Set A study coded only on enterprises that redefined and survived cannot distinguish the correlates of redefinition from the correlates of survival. To address this, the design adds a contrast set of five enterprises whose attempted redefinitions failed or were forfeited: Kodak, Nokia (devices), BlackBerry (devices), Yahoo, and Sears. Selection follows the same logic as the primary sample—variation on the framework's own constructs, here on how failure arrived: failure with pioneering learning present (Kodak), failure by reactive timing from a burning platform (Nokia's and BlackBerry's device businesses), failure by purpose churn (Yahoo), and failure by chain inversion (Sears). Two of the five also contain partial suc‐ cesses—Nokia's and BlackBerry's surviving entities redefined around a narrower core—which the coding preserves rather than simplifies. Three caveats bound what the contrast set can support, and each is reflected in how it is used. First, it is a contrast set, not a matched control group: the five differ from the eighteen in era, industry mix, and terminal state, and no matching pro‐ cedure is claimed. Second, the evidentiary basis is asymmetric: no equivalent of the Volume IX–X corpus exists for these enterprises, so the coding rests on the pub‐ lic historical record—bankruptcy filings, contemporaneous company announce‐ ments, and first-tier press coverage, each cited with its date—and is correspond‐ ingly coarser: Section 6 codes fewer cells, and its maturity codings carry two addi‐ tional constraints, being assigned only where documented behavior supports them and time-anchored to each case's decisive window rather than to the enterprise's whole history. Third, retrospective coding of famous failures invites hindsight bias; the discipline applied is to code only documented events and dated decisions, nev‐ er imputed states of mind, and to state the failure signature as a pattern the pro‐ positions predict, not as an explanation the deaths prove. Within these limits, the contrast set does the one thing the primary sample cannot: it shows what the coding instrument returns when redefinition did not sustain. Axis 1 · Dimensions moved Purpose / Business / Organization / Capital / Leadership — moved, expression-only, unobservable Axis 2 · Maturity (ERMM) Level 1–5 per dimension; withheld cells recorded as unobservable, never interpolated Axis 3 · Position on the FVC Purpose → Learning → Redefinition → Creation → Enterprise Value: where value originated Axis 4 · Redefinition patterns Six patterns; multiplicity coded; declared vs. actual paths coded separately Enterprise Redefinition Observed VURA Working Paper Series 9 4 The Eighteen Cases Table 1 lists the cases with the redefinition each one's evidence centers on. The profiles that follow are deliberately compressed—each case receives book-chapterlength treatment in the source volumes—and are limited to what the coding uses: what was rewritten, when, on what evidence, and where the structure is exposed. Table 1. The eighteen cases Case Source What the case evidence shows being redefined NVIDIA IX, 081 The unit in which computing is delivered: component → development environment (CUDA, 2006) → system → infrastructure Microsoft IX, 082 The criterion of the objective: from protecting a product (Windows) to customer outcomes (2015 mission); all five dimensions follow Amazon IX, 083 The boundary between inside and outside: internal capabilities re-read as external markets (AWS, seller services, advertising) Apple IX, 084 The unit of value from single sale to continuing relationship; the bound‐ ary of in-house design (silicon in, foundation models out) OpenAI IX, 085 The vessel of the mission: nonprofit → capped-profit (2019) → PBC under foundation governance (2025), purpose sentence held Tesla IX, 086 Industry assumptions (EV viability) converted into design problems; four units of sale side by side; simultaneous redefinition under way Google IX, 087 The form of the route to information (search → generated answers), run on top of the largest revenue source, plus a second unit (compute) Meta IX, 088 Declared: the metaverse. Actual: the content of the advertising business, from placement inventory to outcome prediction Netflix IX, 089 The skeleton of the business, rewritten three times (DVD → streaming → originals → advertising)—twice begun while healthy, once cornered (2022) Sony IX, 090 The unit of value (“emotion,” 2019) that binds unlike businesses and licenses the ending of others (finance, EV, discs) IBM X, 091 The correspondence between one unchanging role (holding up enter‐ prise cores) and its era-technology; both chain orders in one history TSMC X, 092 The industry's division of labor (foundry, 1987), enabled by the founding constraint of never competing with customers ASML X, 093 The boundary of the enterprise as a unit: the relationship structure (customers, ZEISS, institutes) designed before the EUV machine existed Arm X, 094 How an industry shares a computing foundation: the decision not to manufacture; value creation and value capture visibly separated Costco X, 095 Where profit comes from (product margin → membership fees, 1983), aligning the enterprise's interests with the member's—once, deeply Enterprise Redefinition Observed VURA Working Paper Series 10 BYD X, 096 The boundary between component and finished product, climbed in re‐ verse (battery → vehicle, 2003); capability placed ahead of demand SpaceX X, 097 The price of reaching orbit (≈85% below historical average), which brought a market into existence; 24 years of private time, listed 2026 Anthropic X, 098 The location of purpose: from statement to legal form (PBC, trust with board-majority power, written stopping conditions) Source volumes: Kadowaki (2026c) for IX, Kadowaki (2026d) for X. Chapter numbers run across the series (081–098). 4.1 Volume IX: The Platform Era Redefiners NVIDIA. Revenue for fiscal 2026 (ended January 25, 2026) was $215.9 billion, up 65 percent, with GAAP gross margin of 71.1 percent (earnings release of February 25, 2026)—but the coding treats these figures as outcome, not object. The redefinition dates to 2006, when CUDA committed die area and development resources to an application with no market; GTC 2026 marked the platform's twentieth an‐ niversary. The unit of sale then moved repeatedly—component, development envir‐ onment, system, and, in the words Jensen Huang used at Computex on June 2, 2026, “not just a GPU company, not just a systems company, but an infrastructure company.” Business, Organization, and Capital moved; Purpose moved only in ex‐ pression (graphics acceleration to computation itself). Leadership operates by pub‐ lishing an architecture schedule to which an industry synchronizes—a mechanism the case codes as its strongest term and, because a single broken promise desyn‐ chronizes the industry, also its most exposed. Fragility concentrates in buyer con‐ centration and in reported (unconfirmed) involvement in customer-side financing (Axios, July 27, 2026). Microsoft. The only case besides Tesla in which all five dimensions moved—and the sequence is documented. The 2015 mission revision (“to empower every person and every organization on the planet to achieve more”) replaced an achieved product-naming objective; the same year, a $7.6 billion phone-hardware impair‐ ment with up to 7,800 job reductions (July 8, 2015) converted the declaration into capital allocation. Business followed (packaged software to cloud to agents; Azure past $100 billion annual revenue, July 29, 2026), Organization followed (Office for iPad, GitHub's independence after the June 4, 2018 acquisition), Leadership re‐ designed the layer structure rather than a product. The chain position follows the consolidated instrument—the rewriting of Purpose into Learning—while the chapter narrative locates realization at the capital reallocation that begins with a recognized loss; on either reading, the 2015 criterion change is what unlocked the rest. Fragility codes at Trust—the European Commission made commitments on Teams tying legally binding on September 12, 2025—and at partnership asym‐ metry. Amazon. The redefinition is the boundary line itself: internal inconveniences re-read as external markets. Systems opened to outside developers from 2002; storage and Enterprise Redefinition Observed VURA Working Paper Series 11 compute services launched in 2006; first separate AWS disclosure in 2015. By the year ended December 2025, AWS carried $45.6 billion of $79.9 billion consolidated operating income on under a fifth of revenue. The 1997 shareholder letter (“It's All About the Long Term”) is coded as the investor-facing time mechanism: the declar‐ ation preceded, by years, the allocations it would later justify. Business, Organiza‐ tion, and Capital moved; Purpose only extended its reach (who counts as a custom‐ er widened four times). Maturity is deliberately unbalanced in the coding—Busi‐ ness and Capital at Level 4 against Organization at Level 2—registering the case's own evidence that governance maturity has not kept pace (FTC monopoly suit filed September 2023; $2.5 billion Prime settlement, September 2025). Apple. The unit of value moved from the single device sale to the continuing rela‐ tionship: Services gross margin of 75.4 percent against 36.8 percent for Products in fiscal 2025, paid subscriptions past 1.5 billion (call of July 30, 2026), an installed base past 2.5 billion devices. Organization moved as the boundary of in-house design moved—inward for silicon (transition announced June 2020) and modems (C1, February 2025), outward at the newest layer, where Gemini was adopted for Siri (joint statement, January 12, 2026). The case codes the cost of slowness dimen‐ sion by dimension: slowness that builds quality accumulated trust; slowness from late learning at the AI layer accumulated dependence. Purpose is coded expres‐ sion-only; Leadership is coded not-moved on the evidence of designed continuity (the April 20, 2026 succession from the hardware-engineering lineage), a reading the source text explicitly flags as testable only over years. OpenAI. The purpose sentence—ensuring that AGI benefits all of humanity—held from 2018 through August 2026 while its vessel was rebuilt three times: nonprofit (2015), capped-profit LP (March 11, 2019), and public benefit corporation under foundation governance (October 28, 2025), with a confidential S-1 disclosed June 8, 2026. Each rebuilding tracks the same trigger: required capital exceeding the conditions under which the purpose could be pursued—billions in the 2019 explan‐ ation, a single $250 billion cloud commitment by 2025, $122 billion of funding com‐ mitments at an $852 billion post-money valuation by March 31, 2026. The case codes at Creation on the chain—the unit cost of intellectual work was rewritten for an industry—and is the sample's clearest instance of the paradox that the larger the purpose, the more its means can threaten it. Fragility codes at Trust and at governance still being carried out. Tesla. The founding redefinition converted an industry belief—EVs are unviable— into a design problem, creating the market over twenty years. The current phase is the sample's test of simultaneity: robotaxi (launched June 2025), Cybercab produc‐ tion, the Optimus line installed where Model S/X lines stood, storage, semiconduct‐ ors—while the funding core thinned to a 1.4 percent operating margin in Q2 2026 (announced July 22, 2026) with free cash flow of negative $1.09 billion. All five di‐ mensions moved; Purpose codes as re-articulated (Master Plan Part 4's “sustain‐ able abundance,” September 2025) with the source text's caution that extension and replacement cannot be distinguished from outside. The case supplies Finding Enterprise Redefinition Observed VURA Working Paper Series 12 2's central lesson: simultaneous redefinition without a ceiling is dispersion, not transformation. Google (Alphabet). Learning ran far ahead: the Transformer was published openly in 2017; the TPU had been under development for ten years (call of February 2026). The redefinition—deploying the frontier model into the largest revenue source on announcement day (Gemini 3 into AI Mode, November 18, 2025)—came years after the learning, and the case codes the gap itself: the chain position is Learning, with the connection into Redefinition late. A second unit of value (compute) grew along‐ side: Cloud backlog of $514.0 billion at end-June 2026, more than double six months earlier, and TPU capacity sold even to a competitor (Anthropic agreement, April 6, 2026). Capital converted accumulation into infrastructure at a scale that turned Q2 2026 free cash flow negative ($5.9 billion outflow) under capex guidance raised to $195–205 billion. Purpose codes expression-only—the mission's two verbs, organize and generate, now pull against each other. Meta. The sample's declared-versus-actual case. The declaration (October 28, 2021) renamed the company for a bet about means; nearly five years on, Reality Labs rev‐ enue was $431 million—about 0.7 percent of the Q2 2026 total—against an operat‐ ing loss of $4.619 billion. The actual redefinition ran undeclared inside advertising: the unit sold moved from placement inventory to outcome prediction (user-under‐ standing models lifting clicks 8.3 percent and conversions 15.7 percent; Advant‐ age+ at a $75 billion run rate; call of July 29, 2026). Only Capital is coded as moved on the declared path—and it moved to AI compute (2026 capex guidance raised to $130–145 billion), not to the declared territory. The case grounds Finding 5's incoherence reading (Business at Level 2 against Capital at Level 4) and Sec‐ tion 8.2's instruction: measure purpose on the capital allocation table, not in de‐ clarations. Netflix. Three rewritings of the business skeleton in under two decades, two of them begun while the existing business was healthy: streaming launched in 2007 in the same quarter the mail business grew 40 percent (earnings release of April 18, 2007); originals began in 2013 capped at a single-digit percentage of content spend (shareholder letter, April 2013). The third turn was reactive—advertising was adopted in October 2022 only after roughly 1.2 million members were lost across two quarters—and the case codes it as a warning, not a model. The chapter's distinctive leadership evidence is error reversal without loss of authority (the 2011 DVD-separation withdrawal: “I messed up”). Fragility codes at content as fixed cost (2026 amortization up about 10 percent), declining external verifiability (quarterly subscriber disclosure ended from 2025), and an untested post-founder culture (Hastings left the board in June 2026). Sony. The 2019 Purpose (“Fill the world with emotion”) is coded not as branding but as the unit of value that licensed subtraction: the financial spin-off (executed October 1, 2025, turning the FY2025 bottom line into a ¥326.9 billion loss beside record operating income of ¥1,447.5 billion), the EV cancellation (March 25, 2026), Enterprise Redefinition Observed VURA Working Paper Series 13 and the end of disc production for new titles from January 2028 (announced July 1, 2026). Purpose, Business, Organization, and Capital moved; entertainment rose from about 30 percent of revenue in FY2012 to about 67 percent in FY2025. The case supplies the sample's clearest instance of deliberate short-term deterioration as evidence of chain order—and its fragility is the mirror image: with the smooth‐ ing device released, IP volatility passes into consolidated results unbuffered (the ¥120.1 billion Bungie impairment as the concrete instance). 4.2 Volume X: The Industry Makers and the Incumbent IBM. The only case whose own history contains both chain orders. In the first half of the 2010s an enterprise-value target stood first—Roadmap 2015's non-GAAP EPS of at least $20, abandoned October 20, 2014—capital worked backward from it, and revenue fell below the prior year for twenty-two consecutive quarters, a run ending in Q4 2017. From 2019 the order re-inverted: Red Hat in (closed July 9, 2019, about $34 billion), the managed-operations business out (Kyndryl spin-off, November 3, 2021), models opened (Granite under Apache 2.0, May 21, 2024), and a published, verifiable quantum timetable (Starling, 2029; June 10, 2025). Purpose codes expression-only across a century—the role, holding up the core operations of enterprises, has not moved. IBM carries the sample's only explicit period codings: Capital at Level 2 in the early 2010s and Level 4 in character since; Business at Level 3 with Level 4 in character from 2019. The case is the analysis's fixed point for Finding 3 and for the investor sequence test of Section 8.3. TSMC. The founding decision of 1987—a dedicated foundry that would never com‐ pete with its customers—redefined the industry's division of labor and created the fabless category on the other side of the boundary. Purpose codes as re-articulated at founding and unmoved since; the constraint (no products of its own) is coded as the asset that makes the purpose believable. The chain position is between Pur‐ pose and Redefinition: trust was set first, and financial capital moved after—cus‐ tomers commit to fabs before they exist and co-own them (JASM: Sony ≈6.0 per‐ cent, Denso ≈5.5 percent, Toyota ≈2.0 percent; announcement of February 6, 2024). Q2 2026: revenue of US$40.2 billion, up 36 percent, gross margin 67.7 per‐ cent (release of July 16, 2026); an additional $100 billion Arizona commitment brought cumulative U.S. investment to $265 billion. Fragility is the strength's shad‐ ow: geographic concentration, demand skew (HPC about two-thirds of revenue), and neutrality tested not in wording but in capacity allocation. ASML. The redefinition preceded the product: before EUV worked, the company de‐ signed the relationship structure that could make it work—the 2012 customer coinvestment program (Intel, TSMC, and Samsung contributing €1.38 billion of R&D funding and €3.85 billion for 23 percent of shares; announcement of August 27, 2012) and the 2016 ZEISS SMT stake (24.9 percent for €1 billion plus about €760 million of support). From the 1995 origin of the optics program to 2013, the tech‐ nology generated no sales; more than €6 billion went into EUV R&D over seven‐ teen years. Business, Organization, Capital, and Leadership moved; the chain posi‐ Enterprise Redefinition Observed VURA Working Paper Series 14 tion is between Learning and Redefinition. Leadership hands an industry its sched‐ ule (2027 capacity: low-NA EUV and DUV immersion each up 30 percent; July 15, 2026). The fragility is the net's fewness of knots: sole-source dependencies above and below, countable customers, export-control exposure—each a consequence of the same design. Arm. The purest observable divergence of value creation from value capture: designs in more than 350 billion shipped chips and a developer base above 22 mil‐ lion, against revenue of $4.92 billion in fiscal 2026 (announced May 6, 2026). The redefinition was the decision not to manufacture—distributing the design layer made the industry's sharing of a computing foundation possible, and set the ceiling on capture at the same stroke. Purpose codes expression-only across four decades (computation per unit of power, from batteries to data centers); Business moved re‐ peatedly in the unit supplied (license, royalty, compute subsystems, and from May 2026 own-design silicon with Meta as principal co-development partner). Owner‐ ship structure is the case's time mechanism: the 2016 SoftBank take-private (about £24 billion) and 2023 relisting changed the length of time management could use and the party owed an explanation. The 2026 step into own products re-tests the neutrality on which the position rests. Costco. The counter-case that anchors the frequency–coherence distinction. One re‐ definition, at the founding era (form in place since 1983): the enterprise buys on behalf of its members rather than selling to them, and profit comes from member‐ ship fees rather than product margin—fiscal 2025 membership fee revenue of $5.3 billion against operating income of $9.8 billion, gross margin held at 11.12 per‐ cent, renewal rates of 92.3 percent (U.S./Canada) and 89.8 percent (worldwide). Purpose and Business moved once, deeply; Business and Organization code at Level 2 by design; Leadership codes as the discipline of not changing—each peri‐ od's decision not to raise margin, not to widen assortment, not to cut the labor that produces trust (starting wage raised to above $20.00 in March 2025). Prepaid fees are the time mechanism: trust arrives as working capital. The stated fragility is internal—discipline loosening—ahead of any external threat. BYD. A battery maker that climbed the value chain in reverse: founded 1995; a 77 percent stake in a Xi'an automaker acquired in 2003 (about HK$269 million); first mass-produced car in 2005; engine-only production ended March 2022; 4.6 million vehicles in 2025. The core—electrochemical energy storage converted into motive power—codes as unmoved; the application widened. Business, Organization, and Capital moved, with the organizational boundary drawn deliberately inward (bat‐ teries, semiconductors, carriers in-house) and capability placed ahead of demand (968,900 employees at end-2024; R&D personnel on the order of 100,000). The 2026 deceleration is coded as the structure's test, not an anomaly: H1 2026 units of 1.81 million against 2.15 million a year earlier, Q1 net profit down 55.4 percent —vertical integration converts falling utilization into per-unit fixed-cost burden. Strength and fragility from one structure, in its most literal form in the sample. Enterprise Redefinition Observed VURA Working Paper Series 15 SpaceX. The redefinition moved a price: early Falcon 9 brought launch cost to roughly $2,700 per kilogram, about 85 percent below the historical average (pro‐ spectus of June 12, 2026, citing NASA material). Cheap transport brought a con‐ stellation market into existence, and the company became its own largest custom‐ er: Starlink at about 10.3 million subscribers (March 31, 2026), communications operating income of $4.423 billion in 2025 against a space-segment loss. Business, Organization, Capital, and Leadership moved; Purpose—the multi-planetary sen‐ tence—codes as stable for 24 years while the path evolved. The time mechanism is the sample's starkest: 24 years of private capital before audited figures ever be‐ came public, and a listing (June 2026, at $135 per share) designed so capital ar‐ rives from the market while the decision time-axis does not—approximately 82.4 percent of voting power retained. The same device codes as the fragility: a structure that protects the time horizon also resists correction. Anthropic. The redefinition relocates purpose itself: from statement to legal struc‐ ture. The corporate purpose sits in a Delaware public benefit corporation; a trust of five independent trustees holds a separate share class whose rights expand to electing a board majority; stopping conditions are written into a published policy (Responsible Scaling Policy, September 19, 2023); and the protocol standard the company created was donated to a neutral foundation co-founded with a competit‐ or (MCP to the Linux Foundation's Agentic AI Foundation, December 9, 2025). Pur‐ pose, Organization, and Leadership moved—the last by dilution of control rather than concentration. Scale followed the constraints: run-rate revenue of $14 billion and a $30 billion Series G at a $380 billion post-money valuation (February 12, 2026), all self-reported and unaudited, a caveat the coding preserves. The case's open question is stated in its source: constraints are easy to place while small; the machinery has no record of operating under conflict. 5 Cross-Case Analysis This section develops eight findings. Each is stated first as a count that could have come out otherwise, then interpreted. All counts derive mechanically from the coding table in Appendix A. 5.1 Finding 1: Core Purpose Was Replaced Nowhere—and Two Layers Are Not Enough Across eighteen enterprises whose businesses, boundaries, and capital structures were rewritten—sometimes repeatedly, sometimes radically—the number that re‐ placed its core purpose is zero. This is the strongest single replication in the study, and it was falsifiable: nothing prevented the sample from containing enterprises that pivoted purpose as readily as product. NVIDIA's core held from graphics to in‐ frastructure; Amazon's held while who counts as a customer widened four times; OpenAI's one-sentence mission survived three rebuildings of its own corporate ves‐ sel; Arm's computation-per-watt premise governed four decades; IBM's role—hold‐ Enterprise Redefinition Observed VURA Working Paper Series 16 ing up the core operations of enterprises—has outlived every technology that expressed it. But the coding also strained the framework's two-layer account of purpose (Core Purpose versus Purpose Expression). The cases divide into three groups, not two (Table 2). In twelve cases, only expression, reach, or field of application changed— re-expression, the framework's second layer. In six cases something stronger and rarer occurred: the core was newly put into language, or formally re-described, as a deliberate management act with a date—Microsoft's 2015 mission revision, Sony's 2019 Purpose, TSMC's founding constraint of 1987, Costco's founding-era inversion of 1983, Tesla's 2025 restatement, and Anthropic's embedding of purpose into legal form. These are not replacements—in each case the underlying reason for existence is continuous—but neither are they mere expression drift: they are ar‐ ticulation events, and in four of the six the case evidence treats the articulation it‐ self as the hinge on which subsequent redefinition turned. A two-layer scheme forces these events into one of two ill-fitting boxes. The data therefore motivate a three-layer coding—replacement (absent in this sample), re-articulation (six cases, each dated), and re-expression (twelve cases)—proposed in Section 8.1 as a refinement to the framework. Table 2. Purpose: three layers, eighteen cases Layer Cases Enterprises Replacement of core 0 — Re-articulation (dated) 6 Costco (1983), TSMC (1987), Microsoft (2015), Sony (2019), Tesla (2025), Anthropic (embedded in legal form; machinery completed at the Series C round) Re-expression only 12 NVIDIA, Amazon, Apple, OpenAI, Google, Meta, Netflix, IBM, ASML, Arm, BYD, SpaceX 5.2 Finding 2: Redefinition Concentrates in Business and Capital—and Full Simultaneity Reads as Risk Counting the dimensions coded as moved (Figure 2): Business moved in 16 of 18 cases, Capital in 15, Organization in 13, Leadership in 7, Purpose in 6. Business and Capital moved together in fourteen cases—the empirical signature of the framework's claim that redefinition which does not reach the capital allocation table is declaration, not redefinition. The two cases where Business moved without Capital are instructive rather than exceptional: TSMC and Costco each redefined once, at a deep structural layer, whereafter capital allocation executes the structure rather than redefining it. Enterprise Redefinition Observed VURA Working Paper Series 17 Figure 2. Movement by dimension across the eighteen cases. At the other end of the distribution, all five dimensions moved together in exactly two cases: Microsoft and Tesla. The pair is a controlled contrast. At Microsoft the movements were sequenced over a decade—the purpose criterion rewritten first (2015), a recognized loss freeing capital the same year, and business, organization, capital, and leadership then moving from that single point—and the case codes as the sample's most complete sequenced redefinition. At Tesla the movements are simultaneous and ongoing—robotaxi, Cybercab, Optimus, storage, semiconductors —while the funding core thinned to a 1.4 percent operating margin; the source analysis states the mechanism plainly: raising the redefinition term by multiplying its objects thins the learning and capital-allocation terms, and in a multiplicative structure that can lower the product. Simultaneity across all five dimensions is therefore not the framework's ideal state; in this sample it is either the visible residue of a completed sequence, or a risk state in progress. Redefinition capability shows itself in the ordering of movement, not in its breadth at an instant. 5.3 Finding 3: Value Originates Upstream—the Chain's Ordering Survives a Test It Could Have Failed Axis 3 locates, for each case, the stage of the Future Value Chain at which the value later recognized as enterprise value originated (Table 3, Figure 3). The distri‐ bution is one-sided in a way the chain predicts and rival orderings do not: seven‐ teen of eighteen cases locate value origination at or upstream of Redefinition—be‐ fore Creation—and zero cases locate it at Enterprise Value. The single case at Cre‐ ation (OpenAI) is the exception that measures the rule: what was created there was not a product ahead of its market but a rewriting of an industry's cost assump‐ tions—creation functioning, in effect, at the boundary where redefinition operates on the industry rather than the firm. Table 3. Where value originated on the Future Value Chain Position Cases Enterprises Purpose → Learning 2 Microsoft, Anthropic 2 Sony, TSMC Business 16 / 18 Capital 15 / 18 Organization 13 / 18 Leadership 7 / 18 Purpose 6 / 18 Cases in which the dimension is coded as moved (re-articulation counted for Purpose; expression-only movement not counted) Enterprise Redefinition Observed VURA Working Paper Series 18 Purpose → Redefinition (order set at purpose) Learning 1 Google (connection into Redefinition came late) Learning → Redefinition 7 NVIDIA, Amazon, Meta (actual path), Netflix, IBM, ASML, BYD Redefinition 4 Tesla, Arm, Costco, SpaceX Redefinition → Creation 1 Apple Creation 1 OpenAI Enterprise Value 0 — Figure 3. The Future Value Chain with the coded origin of value for the eighteen cases. The modal position—seven cases—is the boundary between Learning and Redefini‐ tion: an enterprise learns in a territory with no market, and the learning rewrites what the enterprise is. CUDA's twenty years from commitment to financial con‐ sequence; Amazon's engineers losing time to infrastructure and management rereading the inconvenience as a market; ASML's seventeen years and more than €6 billion of EUV R&D before volume sales; BYD's eight years of battery mass produc‐ tion before the 2003 vehicle entry. The chain's practical content is a prohibition on reading these sequences backward—treating the 2026 financial statements as the cause rather than the residue. And the sample contains the prohibition's violation as a documented interior case: IBM's Roadmap 2015 placed the chain's final term first, worked capital allocation backward from it, and produced twenty-two consec‐ utive quarters of revenue decline. The chain's ordering is thus supported not only by distribution but by the one episode in the sample where management demonstrably ran it in reverse. 5.4 Finding 4: Patterns Never Occur Alone Coding the six redefinition patterns yields two regularities (Table 4). First, multipli‐ city is universal: all eighteen cases exhibit at least two patterns; no case in the sample redefined through a single pattern. Second, the Layer Shift pattern is dom‐ inant but never sufficient: it appears in thirteen cases—more than twice the fre‐ quency of any pattern except Ownership-to-Usage (7)—and in every appearance it is accompanied by at least one other pattern. Moving up or down the value structure, in other words, is how redefinition most often travels, but what makes Purpose Learning Redefinition Creation Enterprise Value 0 +4 1 +7 4 +1 1 0 at stage spanning at stage spanning at stage spanning at stage at stage Counts of the eighteen cases by coded origin of value. Spanning counts sit between the stages they connect. The +4 after Purpose comprises Purpose→Learning (2: Microsoft, Anthropic) and Purpose→Redefinition (2: Sony, TSMC); Table 3 gives exact positions. No case codes at Enterprise ValueEnterprise Redefinition Observed VURA Working Paper Series 19 the movement a redefinition rather than an extension is the second pattern riding with it: the capability being redeployed (NVIDIA, Amazon, IBM, BYD), the customer being redefined (TSMC, OpenAI), the unit of sale converting from ownership to us‐ age (Apple, ASML, Arm, Tesla), or the societal challenge being converted into a market (OpenAI, Tesla, SpaceX).