Map the payer before choosing an investment layer
Publisher-interest disclosure: Kooya develops AI systems and benefits commercially from AI adoption. This educational analysis is not a personalized investment recommendation or a ranking of securities. Named businesses illustrate disclosed economics and risks; inclusion does not establish investment suitability or endorsement.
This is a framework for financial investing in businesses exposed to AI infrastructure. It is separate from deciding whether to buy AI services for your own company. An operating team asks whether a system improves a workflow; an investor also needs to ask what cash reaches the security holder and what price is being paid for it.
Kooya analysis: draw a chain from the end customer’s budget to an application provider, its model or cloud supplier, the infrastructure operator and the equipment and electricity suppliers. Real businesses overlap these categories. A company may own chips, lease facilities, sell cloud access and distribute applications at the same time. Identify the legal entity and the particular asset or security rather than assigning the whole group a fashionable label.
The same peso or dollar can appear at several points as it is spent along this chain. Adding every participant’s sales or investment announcement together does not measure unique end-user demand. Begin with the source of payment and then examine the contractual obligations that continue if demand slows.
Chips and networking: demand must become deployable systems
NVIDIA’s quarter ended 26 July 2026 filing identifies land, power, building shells and capital as necessary resources for customers to deploy AI infrastructure. It warns that shortages can affect future revenue and financial performance. The filing also describes the possibility that customers postpone adoption because deployment infrastructure or financing is unavailable. These are issuer-disclosed risks, not a forecast that a particular delay will occur.[1]
Kooya analysis: a hardware thesis should connect orders to deliveries, customer acceptance, cash collection and the next upgrade cycle. Ask how concentrated customers are, which commitments remain if demand weakens, and whether a product transition can leave inventory or capacity mismatched. A powerful chip is not an operating data centre by itself.
Separate the economics of the designer, manufacturer, memory supplier, networking supplier and system integrator. Their margins, investment requirements and exposure to a change in architecture need not be similar. Examine the specific issuer’s disclosures rather than using one successful supplier as evidence for an entire basket.
- Evidence to seek: revenue recognition policy, customer concentration, inventory, purchase commitments and product-transition risks.
- Key question: who bears the cost if delivery is ready before the customer’s site or financing is ready?
Cloud: recurring usage meets a large investment bill
Microsoft’s fiscal 2026 annual report says it will continue capital expenditure supporting cloud growth and AI infrastructure. It also attributes pressure on Microsoft Cloud gross margin percentage to AI infrastructure investment and growing AI product usage, partly offset by efficiency gains. More usage and stronger economics therefore cannot simply be treated as interchangeable claims.[2]
Kooya analysis: cloud capacity needs customers who pay enough over time to cover more than the current electricity bill. Read capital expenditure alongside operating cash flow, leases, depreciation and the useful lives assumed for equipment. Compare these consistently across periods. Buying hardware outright and accessing it through a lease can create different reported cash-flow patterns without eliminating the underlying economic obligation.
For a compute-rental business, investigate utilisation, realised pricing, customer duration and concentration. A full booking calendar may still differ from cash collected, and a long contract is only as reliable as its conditions and counterparty. Do not treat management’s backlog or pipeline language as equivalent to already-earned cash.
- Evidence to seek: cash-flow statement, lease commitments, depreciation policy, capacity spending and revenue definitions.
- Key question: how much paid utilisation is needed before expansion generates cash after replacing equipment?
Data centres: distinguish a building from productive capacity
Equinix’s June 2026 quarterly filing describes recurring and non-recurring revenue and warns that power limitations can constrain capacity even where physical cabinet space is available. Its operating-risk disclosures also explain how service interruptions can affect customer obligations and financial results. The property, electrical infrastructure, connectivity and operating service must work together.[3]
Kooya analysis: keep four milestones separate—announced development, energised capacity, contracted capacity and paid utilisation. Ask how the issuer defines each. A large megawatt announcement does not establish the date that tenants can use it, the level of contracted demand or the portion of cash that accrues to the listed company.
Read ownership and financing carefully. A property owner, a colocation operator and a joint-venture investor can have different rights to the same facility’s cash flows. Check construction expenditure, maintenance obligations, debt maturity, interest costs and partner distributions. A project can have growing demand while its equity holders face refinancing or additional capital needs.
- Evidence to seek: commissioning dates, operational capacity definitions, occupancy or utilisation measures, customer contracts and ownership percentages.
- Key question: which entity receives rent or service revenue, and which entity must fund the next stage?
Power and cooling: a bottleneck is not an automatic investment return
The IEA’s 2026 energy-and-AI assessment describes slow grid connections and supply constraints affecting electricity equipment. It also cautions against assuming a general uplift for the whole energy sector from AI: the exposure differs across technologies and businesses. Its forward-looking scenarios describe conditional possibilities, not guaranteed demand or equity returns.[4]
Kooya analysis: examine the contract that turns electricity demand into revenue for the company you are evaluating. The relevant business might sell equipment, construct a connection, generate electricity or operate a regulated network. Each has different payment timing, cost exposure and obligations. An equipment order can create a working-capital burden before cash is received; an electricity contract can allocate fuel and price risk differently from a spot-market sale.
Cooling and electrical upgrades also require site-specific execution. Ask whether the proposed facility can support the intended rack density and who pays for modifications. Avoid extrapolating an overseas grid bottleneck into a claim about a particular Philippine utility or property without local disclosures.
- Evidence to seek: orders versus recognised revenue, connection agreements, completion dates, funding requirements and contract risk allocation.
- Key question: does extra demand improve cash flow for this business, or require more capital before any benefit arrives?
Applications determine whether customers keep paying
Kooya analysis: the application layer connects infrastructure to a customer outcome. Its economics depend on the price customers will repeatedly pay and the full cost of serving them. More model calls can indicate useful adoption, wasteful workflow design or a promotion; the usage count alone does not resolve which.
Trace the application’s payments to infrastructure suppliers and the supplier’s exposure back to those customers. Faster or cheaper inference may improve an application’s margins, reduce the price it can charge, expand demand, or combine all three. The outcome depends on competition and customer behaviour; it is not safe to assume that every layer captures the same benefit.
For an infrastructure investment, this becomes a demand-quality question. Ask whether the end use has recurring budgets and measurable value, how sensitive those budgets are to price, and whether a small number of model or application companies account for a large share of projected demand. Financial support and commercial purchases between connected counterparties deserve careful reading in the relevant filings.
Apply the map to Philippine exposure and shared downside
PLDT Enterprise’s VITRO service page illustrates the operational side of local data-centre exposure: it describes data-centre services and peering with telecommunications and internet providers. The page is a service description, not evidence of investment returns, current asset valuation or a security’s direct ownership of every advertised facility.[5]
Kooya analysis: when reviewing a Philippine listed company or proposed project, trace the exact subsidiary, facility, partner interests and financing structure through its latest filings. Distinguish revenues earned from local colocation or connectivity from a claim of direct exposure to AI training. For foreign securities, separately evaluate the account route and the effect of currency movements on the peso value you ultimately need.
Run a common downside scenario across the whole chain: a large customer delays expansion; equipment delivery continues; power connection takes longer; borrowing costs remain payable; application revenue grows more slowly than expected. This is a stress-test scenario, not a prediction. Record which holdings lose cash in the same scenario. Owning a chip supplier, a cloud company and a facility operator can still concentrate you in one spending cycle.
Use our comparison of AI stocks versus funds to assess how the value-chain exposures fit into the investment vehicle you are considering.
- Build a one-page investment map: payer, asset owner, source of cash, required reinvestment, financing and main failure condition.
- Recheck valuation separately: a strong operating business can still be an unattractive investment at a price that assumes too much success.
- Use scenario ranges to expose assumptions; do not present them as projected returns.