Abstract
For almost two decades, digital leaders grew as asset-light firms, scaling on software, data, and talent. The arrival of generative AI changed that logic. Compute and energy are now production inputs, not background utilities. They demand chips, cooling, sites, grids, and permits, and therefore massive capital and operating capabilities. Yet ownership of those assets is not a one-size-fits-all solution. Firms can choose from a continuum: on-demand cloud, committed cloud capacity, dedicated servers, leased data center space, hybrid core-plus-cloud, or full ownership. In this new landscape, how should AI-driven firms, from mid-sized adopters to hyperscalers, secure capacity and what must leaders do next?
Keywords
Key Recommendations
Treat compute as a strategic input, not routine IT overhead.
Align product roadmap and innovation speed with compute procurement and capacity.
Make cost to serve and capital discipline nonnegotiable.
Mid-sized firms should choose carefully where they belong on the rent-to-own continuum.
Large firms that build infrastructure must rethink planning, governance, talent, and metrics.
Add relevant ESG metrics as AI operations become more energy- and water-intensive
For the first 25 years of this century, the corporate story was dominated by a small set of technology giants. The Magnificent Seven (Nvidia, Microsoft, Apple, Amazon, Alphabet, Meta, and Tesla) came to define modern capitalism. Except Tesla and Amazon, these firms also rewrote the playbook for growth. Their success looked different from twentieth-century industrial giants like Ford, GE, and Walmart. 1 Their advantage came from innovation, patents, brands, data, algorithms, organizational know-how, and talent. These assets could be deployed repeatedly around the world with relatively little additional spending. They grew exponentially with rapid scaling of digital products, often with powerful network effects and with fewer physical bottlenecks than traditional industrial businesses. They scaled with little need for physical infrastructure and raw materials, a shift aptly captured in the phrase “capitalism without capital.” 2 Cloud computing reinforced this logic by making infrastructure feel readily available and more pay-as-you-go than fixed and upfront investment. That is why the word hyperscaler gained traction. It described a firm that could grow output much faster than its owned physical footprint. Many came to believe that the best companies were those that avoided heavy assets altogether.
That model is now under pressure. The launch of ChatGPT and the arrival of generative AI have set off an arms race that signals the end of seemingly costless digital scaling. 3 The launch also revealed a constraint in a resource that many leaders had underestimated: compute. Although compute feels like a service, it depends on physical assets. The emerging world of AI requires advanced chips, fast networks, large electricity needs, grid and site upgrades, cooling systems, land, backup systems, and specialized construction and operations teams. The new model differs not only in operations but in the economics of scaling and risk. Just four big tech firms plan to spend $650 billion in AI infrastructure in 2026, up about 70% from the prior year. 4
The new reality changes both firm strategy and how it should be executed, even for small and medium-sized enterprises. Imagine a $2 billion SaaS firm whose new AI feature suddenly doubles usage. Customer enthusiasm is rising but so are cloud bills. There are questions about response-time, service standards, and uninterrupted uptime. Enterprise customers start asking where their data is processed, whether it is kept secure and confidential. Other customers question the resilience of the service and whether performance can be guaranteed during unexpected peak periods. What looked like an ordinary technology decision quickly becomes a strategic question about capacity, control, cost, and risk. That is the new world this article addresses. It also explains what managers, boards, and investors should do about it.
BYOC: Bring Your Own Compute
We define an AI-driven company as one whose competitiveness and the profitability of serving each query depend on large-scale AI training and model use. 5 For these firms, compute is not just a tool. It is a production input, much like inventory, factory capacity, or logistics in a traditional business. BYOC means deliberately securing compute capacity, through contracts or ownership, rather than treating compute as an always-available utility. This section explains why ownership of compute is becoming attractive for some firms but may not be necessary for others. Then, we lay out a continuum of options, from rent to full ownership.
Why AI Firms Need Ownership
Ownership matters when it creates and protects competitive advantage while reducing operational risk. Three forces push firms toward ownership.
When Certainty of Capacity Becomes a Moat, and Uncertainty Becomes a Business Risk
For years, the public cloud felt readily available. Firms could assume that capacity would show up when needed. At AI scale, that assumption is weakening. Demand is rising faster than supply. AI-related computing needs are growing even faster (roughly four times each year) than Moore’s law (two times every eighteen months).6,7
When firms own, or tightly control, capacity, they can plan a multi-year hardware pipeline, build dedicated capacity, and schedule workloads with fewer interruptions. That predictability speeds learning, enables faster iteration, and improves model quality. Microsoft emphasized that staying ahead in both model training and computing is a critical competitive advantage. 8 Alphabet similarly links competitiveness to timely deployment of new capacity. 9
In contrast, if firms rent capacity, they face shortages, queuing, and delivery delays. Regional availability is uneven. Compute providers’ priorities can change. Product roadmaps can become hostage to someone else’s allocation decisions. Outages also matter. Survey evidence suggests that outages can be expensive even for medium-sized enterprises. 10 For AI products with real-time user interaction, downtime can trigger churn, reputational damage, and contractual penalties. Delays and uncertainty can slow product launches and weaken service quality.
When Designing HW and SW as One System Improves Operating Performance
The strongest argument for ownership is the ability to improve the whole system, from chips to applications, as one integrated operation. Ownership lets firms design hardware, write software, and run day-to-day operations as a coherent system. That is difficult to replicate with commodity rented capacity. A useful analogy is manufacturing. Recall the difference between a collection of general-purpose machines and a vertically integrated factory fine-tuned for just one product.
This concept is aptly described in Microsoft’s “from silicon to software to systems” strategy for optimum utilization of Azure AI workloads. 11 Alphabet frames it as a “full-stack approach” emphasizing vertical integration across “hardware, compilers, models and products” to drive efficiencies across training, serving, and developer productivity. 12 Alphabet’s program illustrates how its TPU supercomputers train models for its specific needs 50 times faster than general-purpose supercomputers. 13
Those choices also shape how much useful output a firm gets from its compute time, hardware spending, and electricity use. Small efficiency gains matter. When companies spend billions and operate on thin margins, even one basis point (one-hundredth of a percent) of improvement can compound into a large advantage.
When Sovereignty, Resilience, and Trust Become Product Differentiators
Many AI products must compete on trust as much as capability. More and more customers and regulators care about privacy, security, resilience, and jurisdiction. Regulations increasingly require companies to keep data within specific geographic boundaries and legal jurisdictions. Some regulators have already made sovereignty, meaning control over where data is processed and under which legal jurisdiction, a first-order consideration. For example, the EU’s GDPR establishes an extensive compliance regime for personal data processing and cross-border transfers, shaping how AI services can store and process user data. 14
Trust is also a market differentiator in consumer AI. Regulated sectors dealing with sensitive data (public sector, healthcare, finance, critical infrastructure), demand infrastructure control (including key management, operational access controls, and location). Apple’s Private Cloud Compute (PCC) is a salient example of an infrastructure choice driven by concerns for privacy. 15
Bring Your Own Power (BYOP)
The old assumption—electricity is always available on demand—no longer holds.
Power is increasingly the binding constraint for AI, even when chips are available.
If you cannot energize capacity, secured compute exists only on paper.
AI workloads are power-dense, so small model expansions can trigger large load increases.
Lead times for major power equipment can run five to seven years (for example gas turbines).
Grid capacity is local, and connecting new loads is slow and requires regulatory permissions.
Adding a few hundred megawatts may require grid upgrades that take years.
Some large customers may need on-site power solutions or accept curtailment.
Treat power access as a strategic input, not a facilities afterthought.
A Continuum of Rent-to-Ownership Strategies
Ownership is not a universal solution. Consider three firms facing the same AI opportunity but very different operational needs: a mid-sized SaaS company that faces volatile demand; a healthcare or financial-services provider that needs tight control because of data, reliability, or regulatory constraints; and a hyperscaler whose demand is increasing exponentially and must plan for cost, speed, and resilience, or lose the AI race. The continuum below helps managers decide where they belong on the rent-to-ownership continuum.
A Simple Ownership Decision Tool: How to Judge the Need for Ownership
To decide when ownership is worth considering, we suggest a simple scoring tool. This table lists the main factors and suggested weights. Each factor is framed as a question, but the answer need not be binary. Assign an internal score from one to five, where five is closer to “yes” and one is closer to “no.” Suggested weights reflect our assessment of typical AI-driven firms; you should adjust them based on their context. Higher scores suggest moving one step closer to ownership on the continuum.
Do not ignore basic economics. If cloud usage starts requiring multi-year rental commitments that resemble ownership, then your rental choice is no longer “asset light” in any meaningful sense. The only difference is paying in installments instead of upfront. At that point, you should consider owning as a way to secure capacity. Ownership may offer more control over design, location, and priority access.
Capital Requirements and Operational Complexity
Need and favorable economics are not enough. Ownership works only if you can execute. Use the checklist to assess readiness.
Can You Execute on Ownership?
1)
2)
3)
4)
5)
6)
7)
For a Mid-Sized AI-Driven Company
Treat compute as a strategic input, not routine IT overhead.
You probably should not build data centers.
Renegotiate cloud contracts before growth forces expensive emergency buying.
Track response-time requirements, outage exposure, and dependence on a single vendor or region.
Use committed cloud capacity, dedicated hosted infrastructure, or such hybrid arrangements when they improve predictability.
Move closer to ownership only if customer requirements, reliability risk, data-control demands, or economics justify it.
Implications for Corporate Leaders
For many digital-native executives, this new world will feel unfamiliar. It is closer to running a utility-like operation than running a software business. That shift changes what management of erstwhile asset-light digital natives must focus on. We offer several recommendations for leaders of AI-driven firms, ordered by importance, as they transition to hybrid models and, in some cases, ownership.
CEO: Capacity Assurance Becomes an Advantage
How much capacity must you control before growth becomes hostage to someone else’s availability, pricing, or outages? Strategy now starts with capacity. For a small firm, that may mean renting. For a hyperscaler, it may mean building. Either way, executives should ask: If AI usage doubles, can we serve it without outages, rationing, or margin collapse?
Five Things the CEO Should Do
Rewrite the strategy story: move from “infinite software scale” to “assured capacity” (compute, power, reliability).
Plan digital and physical capabilities together. Digital cannot do without physical, and the cost of physical can only be justified with the need for digital.
Decide what you must control versus rent: protect mission‑critical workloads; keep elasticity for peaks and experiments.
Treat power as a core input: lock in supply and price where feasible; plan for grid and permitting timelines early.
Earn the social license to operate: engage communities early on including water, noise, land use, and emissions that come with data centers; assess renting or owning.
Human Resources: Hybrid Talent and Operating Culture
How do you build leaders and teams that can manage both software speed and physical operating constraints? As firms move from pure software to hybrid operations, the talent mix changes. Even firms that do not own data centers need people who can negotiate capacity, manage vendors, and run 24/7 services.
Five Things HR Should Do
Hire for hybrid leadership: leaders should bridge software, infrastructure, energy, and compliance.
Build basic literacy in AI infrastructure constraints: teach product and engineering leaders about power, cooling, and lead times.
Reward stewardship, not just speed: recognize reliability, utilization, and cost discipline alongside shipping velocity.
Build local workforce pipelines where you locate or build: partnerships reduce bottlenecks and improve regulator relationships.
Protect operations from burnout: formalize rotations and reduce continuous deployments; resilience depends on people systems.
CFO: Costing and Capital Discipline Return
When does renting stop being flexible and start becoming an expensive substitute for ownership?
AI increases the importance of cost and margin calculations of each customer query and AI task. Whether you rent or own, margins depend on the cost of serving users at the required speed and reliability.
Five Things the CFO Should Do
Make workload economics nonnegotiable: track cost per query, customer task, or AI output the way a retailer tracks gross margin.
Use an investment lens for capacity choices: compare committed cloud, dedicated/managed, colocation, and ownership on cash recovery and option value.
Do not let accounting optics drive internal decisions: measure economic returns, not just expensed vs. capitalized categories. 16
Plan depreciation and refresh cycles: AI hardware obsolescence is faster; budget accordingly in cash burn and profit calculations.
Manage power and capacity volatility: move from surprises to contracts, hedges, and clear risk limits.
COO: Operations and Supply Chains Matter Again
For two decades, firms optimized for code velocity. Now they must also manage lead times for obtaining and installing hardware and physical bottlenecks. What should be prioritized first when chips, power, cooling, or cloud capacity become scarce? Overly dependent on one cloud, one region, one supplier, or one power source? Should firms monitor capacity risk? Small firms may do this through vendor management; hyperscalers have no choice but to do it through build-and-run programs.
Five Things the COO Should Do
Build capacity planning into operations: forecast demand; set contingency plans; define “what runs first” on scarce days.
Treat energy as part of the supply chain: build capability in procurement, permitting, and grid constraints.
Manage long‑lead items as supply chain (chips, networking, transformers, switchgear, cooling) and plan buffers and alternates.
Upgrade site selection criteria: power availability and cooling economics can dominate proximity to talent clusters.
Create clear shortage response plans: decide which workloads get reduced first, how to switch to another region if one goes down, and how practice drills will keep customers safe.
Board of Directors: New Governance and Scorecards
How do you govern a company undergoing transition—from asset light to asset dependent? Does the company have the necessary expertise, fundraising plans, governance and monitoring mechanisms for the transition (even firms that mostly rent need these assessments)?
Five Things the Board Should Do
Refresh board expertise: add capability in energy markets, infrastructure programs, and operational risk.
Require a quarterly capacity risk review: planned vs. delivered capacity, utilization, bottlenecks, and exposure to delays.
Stress‑test concentration and dependency: reliance on one cloud, one region, one supplier, or one power source.
Demand transparency on the full cost of scale: funds raised and spent, contract commitments, refresh cycles, and upgrade needs.
Treat ESG as capacity resilience: water, emissions, and community impacts affect permitting and uptime, not just reputation.
Whether firms rent, use a hybrid strategy, or their own, the leadership task is the same: secure capacity, impose cost discipline, and manage operational risk. The conclusion spells out the paradigm shift those actions are designed to meet.
A New Paradigm Emerges
OpenAI has unleashed a powerful force, one that may end the era of costless digital scaling. The genie is out of the bottle. For digital natives, the laws of physics and chemistry are back in the boardroom. The race is still defined by software, platforms, and algorithms. Yet it is increasingly defined by chips, power grids, cooling, water, land, permitting, and operational limits. Leaders must accept a new reality: scalability is no longer infinite. Preserving the strengths of the digital model now requires better costing, capital discipline, forward planning for compute and power needs, and operational excellence.
This shift matters across the entire continuum, from small and medium-sized enterprises to hyperscalers. SMEs do not need to become infrastructure owners to compete. But they should treat compute as a strategic input, not as a generic IT line item. For many, the right answer is smart renting: committed cloud capacity, dedicated hosted infrastructure, leased data center space, or hybrid arrangements that provide predictable performance without building facilities. The managerial challenge is to choose contracts that function more like capacity insurance than spot purchasing. It also means monitoring early warning signs, shortages, response-time demands, customer demands for data control, and growing power-related constraints, that may justify moving one step closer to ownership and control.
For large enterprises adopting AI at scale, the challenge is governance. They must translate AI ambitions into capacity plans, risk limits, and clear accountability. They need dashboards that track utilization, exposure to cloud pricing and availability, vendor concentration, and the realism of power and permitting timelines—before AI becomes mission-critical and shortages become business outages.
For hyperscalers and frontier AI builders, ownership and tightly integrated infrastructure design will become more essential. But even for them, the point is not ownership for its own sake. The point is disciplined execution: standardizing and tuning the system, securing scarce inputs, and turning reliability and cost per query or AI output into durable advantages. In this new paradigm, “silicon to software to systems” is not just a slogan. It is becoming a management mantra.
Sustainability also moves from a reputational issue to a scaling constraint. The next generation of consumers, employees, and regulators will reward firms that combine renewable energy with sustainable profits. In the AI era, ESG is not just about optics. As tech firms expand data centers and power use, they consume city-scale water and energy. They also create local impacts, from air emissions to water and noise pollution. ESG reputation would increasingly shape the ability to secure local workers, land, and power, as well as to build, operate, and scale. Leaders must revisit sustainability as part of capacity strategy, not as a separate report.
The practical implication is not that every firm must become a utility. The implication is that winning tech firms must rediscover what made great industrial companies successful: disciplined execution, operational excellence, capital planning, and the ability to secure scarce inputs. AI is pulling the digital world back into the physical world. The winners will be firms that can manage both at once—intangible assets and physical constraints, algorithms and infrastructure, growth and reliability.
This shift also has implications for talent. We expect MBA hiring to make a comeback among digital natives. For years, many tech firms prioritized technologists and engineers, often bypassing MBAs. Visionary technologists rose to CEO and senior functional roles. But as AI firms build-and-run data centers, secure power, manage permitting, consider cost economics and gross margins, negotiate contracts, and engage local communities, they become hybrid organizations that manage physical and intangible assets together. That is exactly what MBAs are trained for: finance, operations, supply chains, regulation, strategy, accounting, communications, and organizational design in one integrated toolkit. Consumer giants such as Unilever and P&G have long relied on MBAs and the “general manager” model because they run hybrid operations at scale. In the back-to-the-future AI era, those practices may seep into digital natives as well.
Footnotes
Notes
Author Biographies
Anup Srivastava holds Canada Research Chair in Accounting, Decision Making, and Capital Markets and is a Professor at Haskayne School of Business, University of Calgary. (
Vijay Govindarajan is the Coxe Distinguished Professor at Dartmouth College’s Tuck School of Business (
Luminita Enache is an Associate Professor of accounting at Haskayne School of Business, University of Calgary (