Build now, secure capacity, or acquire later? The strategic trade-off between computing efficiency, infrastructure investment, energy rights, and financial flexibility.
Executive Briefing | For CEOs, CFOs, Boards, and Institutional Investors
Key Context
Anthropic's planned initial public offering (IPO) involves up to approximately $64.35 billion in gross share sales, based on the upper end of its proposed offering range. This represents the size of the public offering, not the company's total equity valuation. Most of the shares are being issued by Anthropic, with a smaller portion sold by existing shareholders.
The offering takes place against a much larger backdrop: Anthropic has disclosed approximately $900.9 billion in aggregate computing and infrastructure-related contractual commitments extending through 2045.
Across the industry, AI developers are pursuing tighter control over semiconductors, computing capacity, and energy infrastructure to secure their long-term competitiveness.
Yet the central investment question extends beyond securing enough computing power.
What if the industry's next competitive advantage comes not from building infrastructure first, but from allocating capital more effectively across successive technological cycles?
This analysis examines the possibility that selective overinvestment, accelerated hardware depreciation, and financial distress could create opportunities for future infrastructure acquisitions.
It also evaluates the opposite scenario: sustained computing scarcity, in which early infrastructure commitments generate durable economic advantages.
Five Strategic Questions
Build now or acquire later? Under what conditions does preserving capital create greater value than immediate infrastructure ownership?
What does AI efficiency actually mean? Why should economically useful outcomes, rather than token volume alone, determine infrastructure investment decisions?
Is proprietary hardware worth its opportunity cost? Can customized chips generate incremental returns sufficient to justify their development?
What is the real recoverable value of a data center? How do energy contracts, interconnection rights, and transfer restrictions affect liquidation and acquisition values?
Who captures value during consolidation? Will early infrastructure owners or financially flexible acquirers ultimately earn superior returns?
The Executive Decision
AI infrastructure demand can continue growing while individual infrastructure investments fail to generate acceptable returns.
The objective is not simply to secure computing capacity. It is to determine which capital commitments remain economically defensible across changing technology, demand, financing, and energy conditions.
1. Anthropic's IPO: The Difference Between Financing Growth and Creating Economic Value
Anthropic's November 10, 2026, amended registration statement identifies 825 million shares in its public offering, with a proposed price range of $70–$78 per share.
Approximately 801.35 million shares are being issued by Anthropic, while approximately 23.65 million are being offered by existing shareholders.
At the maximum proposed price, the combined public offering would raise approximately $64.35 billion before underwriting costs and other deductions. Anthropic's primary issuance would account for approximately $62.5 billion of that amount.
Separately, the company has disclosed a conditional $26 billion private placement involving major technology companies, including Google, Nvidia, AMD, Microsoft, Broadcom, Micron, and TSMC-related investors.
The transaction is therefore predominantly a capital-raising exercise rather than a large-scale shareholder exit.
But access to capital and efficient capital allocation are not equivalent.
Anthropic has disclosed approximately $900.9 billion in aggregate purchase commitments and lease arrangements, largely associated with computing and infrastructure, extending through 2045.
Approximately $451.6 billion is expected to be incurred before the end of 2030, including approximately $70.7 billion during the 12 months following September 30, 2026.
These are commitments over different periods, subject to differing contractual conditions. They should not be interpreted as a single immediate debt obligation.
Their significance lies in the potential divergence between the duration of infrastructure commitments and the economic life of the technologies they support.
Anthropic's revenue is substantially consumption-based. Future demand, pricing, model capabilities, and competitive positioning will continue evolving.
Certain infrastructure agreements, however, include minimum commitments and take-or-pay provisions.
This creates a capital allocation problem:
How much long-term infrastructure exposure is justified when the revenues and technological advantages intended to support it remain uncertain?
The answer cannot be derived from revenue forecasts alone.
2. From Token Efficiency to Capital Efficiency
A central assumption in AI investment analysis is that greater computational efficiency creates a competitive advantage.
That assumption is incomplete unless efficiency is connected to commercially valuable outcomes.
BBIU's earlier research on the Token Efficiency Index (2025) and AI Did Not Democratize Judgment (June 2026) questioned the use of token volume and generated output as proxies for meaningful cognitive or business value.
Those analyses were conceptual and did not establish a validated industry-wide financial metric.
However, their underlying distinction remains relevant:
The economically relevant unit is not a token produced, but a reliable result delivered at an acceptable total cost.
For enterprise AI, total cost may include model inference, computing infrastructure, failed attempts, validation, human correction, operational integration, and error-related losses.
A technically efficient model may be commercially inferior if its outputs require substantially more verification or correction.
Conversely, a model that is more expensive per execution may deliver superior economic value when its accuracy and reliability reduce downstream costs.
The same reasoning applies at the infrastructure level.
A proprietary semiconductor program may reduce computation costs but still produce inadequate returns if the investment diverts resources from higher-value software development or enterprise distribution.
This introduces the opportunity cost of capital.
Developing cheaper computing infrastructure is not necessarily the same as creating a more valuable AI business.
For a company deciding whether to build custom chips, the relevant comparison is the incremental value created by proprietary hardware against the best alternative investment of the same capital, engineering talent, and time.
Historical BBIU research:
3. The Overinvestment Hypothesis: What the Telecommunications Cycle Teaches
There is a historical precedent worth examining.
During the late 1990s and early 2000s, telecommunications companies invested heavily in fiber-optic networks in anticipation of rapidly increasing internet traffic.
The long-term demand thesis was broadly valid.
Internet usage continued expanding.
Yet many infrastructure investments became financially unsustainable.
Technological improvements increased the transmission capacity of existing networks, while demand developed more slowly than some operators anticipated.
The resulting excess capacity and declining prices damaged highly leveraged business models.
Global Crossing filed for bankruptcy protection in January 2002, following substantial investment in international fiber-optic infrastructure.
A 2003 report by the United States International Trade Commission documented how technological improvements, excessive deployment, and insufficient demand contributed to failures among major network operators.
The important lesson is not that AI infrastructure will necessarily repeat the telecommunications collapse.
AI computing infrastructure differs from fiber networks in depreciation, technological dependencies, energy requirements, and contractual structure.
The lesson is more specific:
Strong long-term industry demand does not guarantee adequate returns on every infrastructure investment undertaken to serve that demand.
The same industry can produce successful technology adoption, failed infrastructure operators, and attractive acquisitions for later investors.
This is the scenario BBIU proposes to evaluate—not assume.
4. The Counterargument: Computing Capacity Remains Scarce
A credible assessment must also recognize evidence against an imminent infrastructure glut.
CBRE reported that vacancy across major North American data-center markets fell to approximately 1.4% during the first half of 2026.
More than 80% of capacity under construction was already preleased.
Average asking rents increased across major deployment categories, while power constraints continued delaying new supply.
These indicators suggest that high-quality computing infrastructure remains scarce in several important markets.
They do not support describing the current market as experiencing generalized oversupply.
Under these conditions, early infrastructure commitments may generate advantages that cannot be reproduced quickly.
Companies with reserved capacity may launch models sooner, support enterprise demand more reliably, avoid future price increases, and preserve flexibility to deploy new workloads.
Minimum-spend and take-or-pay contracts are therefore not inherently poor investments.
When demand remains sufficiently high, they can provide predictable access to scarce resources and potentially attractive effective computing costs.
The strategic relationships involving Anthropic's technology partners also matter.
Suppliers that are simultaneously shareholders may have incentives to support Anthropic's expansion, although these relationships can also introduce commercial dependencies and conflicts of interest.
The cost of waiting must include the possibility of losing technological momentum, customers, and market position.
Therefore, the appropriate comparison is not aggressive investment versus financial caution.
It is the risk-adjusted economic value of two competing investment strategies.
5. Quantifying the Choice: Build Now or Acquire Later?
Consider a hypothetical AI infrastructure project.
The company requires comparable computing capacity and is evaluating two alternatives.
Strategy A — Build immediately
The company commits $100 million today to establish the required infrastructure.
Strategy B — Preserve capital and acquire later
The company rents equivalent computing capacity during a three-year transition period, paying $12 million annually.
At the end of year three, it acquires an existing facility for $70 million and incurs an additional $15 million in integration, technical adaptation, and contractual remediation costs.
Using a 10% annual discount rate, the present value of Strategy B's costs is approximately $93.7 million.
Under these assumptions, Strategy B has a present-value cost advantage of approximately $6.3 million over building immediately.
But that result is highly sensitive to acquisition pricing.
Illustrative Acquisition-Price Sensitivity
$50 million future purchase price: approximately $78.7 million in total present-value costs. Waiting saves approximately $21.3 million.
$70 million future purchase price: approximately $93.7 million in total present-value costs. Waiting saves approximately $6.3 million.
$90 million future purchase price: approximately $108.7 million in total present-value costs. Building immediately is cheaper by approximately $8.7 million.
In this simplified example, the break-even future acquisition price is approximately $78.4 million.
Above that threshold, the modeled financial benefit of waiting disappears.
These are illustrative assumptions, not actual market valuations. The comparison assumes equivalent computing performance, availability, and business outcomes under both strategies, with no additional operating-cost differences after acquisition. It also assumes the future asset can be acquired at the stated price and that the transition costs are fully captured.
A complete net present value assessment must additionally account for taxes, revenue timing, maintenance, depreciation, financing structure, utilization, probability of acquisition, and the cost of capacity shortages.
If delaying investment causes a model-launch delay, customer losses, or expensive emergency capacity purchases, Strategy B may become substantially less attractive.
Likewise, if the future facility requires additional electrical or regulatory work, its apparent acquisition discount may disappear.
The purpose of the model is not to predict acquisition prices. It is to establish the conditions under which waiting creates value—and the conditions under which it destroys it.
6. Energy Infrastructure: Why Physical Assets and Operating Rights Must Be Valued Separately
Computing equipment and energy infrastructure do not necessarily follow the same economic cycle.
Accelerators and servers may depreciate rapidly as new architectures improve computational efficiency.
Energy infrastructure, by contrast, can retain value because electricity connections, transformer capacity, cooling systems, and regulatory approvals may require lengthy development periods.
But the value of those physical assets depends on more than their replacement cost.
A data center may possess expensive equipment and substantial electrical installations without guaranteeing that a future owner can preserve every associated operational right.
Electricity supply contracts, grid interconnection arrangements, capacity reservations, permits, and financial guarantees may be subject to assignment restrictions, change-of-control clauses, or other contractual and regulatory conditions.
These provisions can become critical during insolvency or restructuring.
A buyer acquiring an operating company may preserve certain contractual relationships, while an asset purchase may require additional transfers or approvals.
Neither structure automatically eliminates the risk.
Legal analysis by Mayer Brown has identified contractual portability as a material factor in AI infrastructure acquisitions, particularly where access to computing capacity depends on long-term commercial agreements.
The same principle must be evaluated separately for electricity agreements and grid interconnection rights, which are governed by their own contracts and regulatory frameworks.
This leads to an important valuation distinction:
The cost of constructing a facility is not necessarily the value recoverable when that facility must be sold.
7. The Contractual Exit Trap: How Asset Values Can Diverge During Distress
Consider a company that has invested $1 billion in data-center infrastructure.
Subsequently, lower utilization, debt pressure, or technological changes force it to seek a buyer.
The facility remains physically intact.
However, the buyer discovers that certain electricity or infrastructure arrangements cannot be transferred without consent, additional guarantees, regulatory approvals, or renegotiation.
The potential acquisition value now depends on several unresolved questions:
Can the buyer preserve electricity access? Will the contracted capacity remain available? Are existing tariffs and commercial terms transferable? Will lenders or counterparties impose additional requirements? Could delays make the installed equipment economically obsolete before the facility resumes full operation?
These uncertainties can reduce the seller's recoverable value.
But they do not automatically create value for the acquirer.
A purchase discount may merely compensate for additional costs, delays, and risks.
The opportunity emerges when the buyer can resolve these constraints more efficiently than the distressed seller.
A financially stronger acquirer may have greater capacity to provide credit guarantees, negotiate with counterparties, finance upgrades, or restructure operations.
In that situation, an asset that has become economically unattractive to its original owner may become viable under new ownership.
The distinction is fundamental:
Financial distress can destroy value for the original investor while creating acquisition opportunities—but only when the buyer can recover operational value at a cost below the discount obtained.
This also changes the responsibilities of infrastructure investors and lenders.
Residual value should be assessed not merely through expected equipment depreciation or physical replacement cost, but through the durability and transferability of the rights required to operate the asset.
For a prospective acquisition, legal and operational due diligence should precede—not follow—the final valuation decision.
8. The Strategic Alternative: Secure Scarcity, Preserve Flexibility
The evidence suggests that infrastructure investment should not be treated as an all-or-nothing decision.
A company may be justified in securing long-lead-time infrastructure while postponing investments in components with shorter technological lives.
Anthropic's own registration statement describes a related strategy under Anthropic Direct.
The company intends to separate commitments for data-center capacity from the timing of accelerator deployments, enabling infrastructure to be secured before committing to particular computing hardware.
This approach reflects a fundamental economic distinction.
Energy availability, facility development, semiconductors, model architecture, and enterprise applications have different investment horizons.
Treating them as a single capital commitment may create unnecessary financial rigidity.
A more flexible architecture could involve securing essential energy and facility capacity early, using contracted computing where economically justified, and preserving discretion over future hardware investments.
It could also preserve the ability to evaluate acquisitions if distressed assets become available.
However, contracted capacity itself may involve substantial long-term obligations and switching costs.
Financial flexibility must therefore be assessed across both owned and contracted infrastructure.
The optimal objective is not minimum capital expenditure. It is maximum risk-adjusted value from each commitment, including the strategic value of capacity availability.
9. Three Scenarios That Could Change the Investment Decision
Scenario A — Sustained Capacity Scarcity
AI adoption and computational demand expand faster than available infrastructure.
High-quality computing capacity remains scarce, and early commitments protect market access.
Under this scenario, building or securing infrastructure ahead of demand may create superior economic returns.
The primary risk is underinvestment.
Scenario B — Selective Infrastructure Consolidation
Overall AI demand continues growing, but individual operators fail to achieve sufficient utilization or financial resilience.
Debt refinancing becomes more difficult, and distressed sales create opportunities for well-capitalized acquirers.
Under this scenario, preserving capital and contractual flexibility may produce significant strategic value.
The primary risk is excessive early capital commitment.
Scenario C — Technological Repricing
Advances in algorithms, accelerator efficiency, or model architecture change computing economics faster than originally anticipated.
Some hardware loses economic value while energy-connected, technically adaptable facilities retain strategic relevance.
Under this scenario, investment timing and infrastructure adaptability become particularly important.
The primary risk is technological obsolescence before capital recovery.
These scenarios are not mutually exclusive. Different regions, operators, and asset classes may experience them simultaneously.
The appropriate strategy depends on which risks are material to a specific investment.
10. BBIU Decision View: The Real Competitive Advantage May Be Capital Allocation
The current AI infrastructure cycle is frequently discussed through investment totals, semiconductor performance, model benchmarks, and anticipated revenue growth.
These are important indicators.
But they do not establish the economic quality of an individual investment.
A company can possess superior technology and still allocate capital inefficiently.
Another company can operate without proprietary hardware and achieve stronger returns through better software, differentiated services, favorable computing contracts, or strategically timed acquisitions.
Likewise, preserving capital may appear prudent but prove expensive if insufficient computing access causes irreversible competitive losses.
The central strategic challenge is not choosing between expansion and restraint. It is determining which commitments create durable economic value under changing market conditions.
This requires evaluating model quality, total computing costs, technological depreciation, financing conditions, energy access, contractual durability, and alternative uses of capital as interconnected variables.
A decision that appears attractive under one assumption may become unfavorable when another variable changes.
That is where conventional investment evaluations can become vulnerable.
An infrastructure project may be technically viable, financially fundable, and commercially promising while still presenting an unacceptable combination of technological, contractual, and timing risks.
Three Questions Every Board Should Answer
1. What evidence demonstrates that building or owning infrastructure now generates superior risk-adjusted returns compared with contracted capacity and future acquisition alternatives?
2. How would the investment perform if computing prices, hardware efficiency, or customer demand evolved materially differently from the base-case forecast?
3. What is the recoverable value of the infrastructure if a financial restructuring or change of ownership disrupts critical operating contracts and energy-access rights?
These questions should be resolved before significant capital commitments become difficult to reverse.
Why This Matters for Institutional Decision-Makers
The investment choices facing AI developers, infrastructure owners, and institutional investors increasingly cross traditional organizational boundaries.
Finance evaluates projected returns.
Engineering evaluates technical performance.
Legal teams evaluate contractual obligations.
Energy specialists assess supply reliability and connection feasibility.
Commercial teams project future demand.
Yet the investment decision depends on how these variables interact.
BBIU's independent strategic red-team approach integrates these perspectives to challenge assumptions that may remain untested within conventional functional reviews.
For organizations evaluating proprietary hardware development, large-scale infrastructure investment, long-term computing agreements, or strategic acquisitions, the objective is to identify which assumptions could materially change the decision before capital becomes committed.
That includes testing alternative investment paths, quantifying financial sensitivities, identifying contractual vulnerabilities, and examining scenarios that contradict the prevailing investment narrative.
The purpose is not to predict an inevitable AI infrastructure collapse or argue that expansion should stop.
It is to distinguish investments justified by durable economic advantages from investments driven primarily by the fear of falling behind.
BBIU — Challenge Your Assumptions. Before You Decide.
Independent strategic intelligence and decision-specific red-team assessments for companies and institutional investors.
www.biopharmabusinessintelligenceunit.com
Selected References
Anthropic, PBC. Form S-1/A, filed November 10, 2026. U.S. Securities and Exchange Commission. Primary filing.
Reuters. Investors Flock to Anthropic IPO but Question Durability of AI Lead. November 13, 2026. Reuters.
CBRE. North America Data Center Trends H1 2026. August 27, 2026. Market research.
U.S. International Trade Commission. Recent Trends in U.S. Services Trade: 2003 Annual Report. Historical telecommunications infrastructure analysis. Report.
Knowledge at Wharton. Factors Behind Global Crossing's Failure. February 13, 2002. Historical case.
Mayer Brown. Portability of AI Compute Infrastructure in AI Acquisitions. June 12, 2026. Legal analysis.
Research note: Financial examples are hypothetical and provided to illustrate the capital allocation framework. They are not Anthropic forecasts or observed acquisition valuations. Strategic scenarios represent conditional BBIU analysis. Contractual rights, transferability, and insolvency treatment depend on the relevant jurisdictions and individual agreements.