Average return on invested capital against the return on new capital, applied to two cloud builders, Microsoft and Oracle. Play the short lesson, step through all eight frames, then back-solve what new capital has to earn.
Play runs all eight frames, about 2:25. Pause and Resume keep your place from any frame, Previous and Next move a frame at a time, and Restart goes back to the start. The space bar also pauses. Record mode fills the window with the 100-second video cut (frames 1, 2, 3, 5 and 7, then the end card); press Escape to leave it.
Pick a company and a target. The tool finds the return new capital would need to earn to get there, and compares it with what new capital has earned so far. It starts from fiscal 2026.
Fiscal 2026 profit is taken to reflect capital in place before the lag. Every dollar added after that, including recent dollars still ramping, earns the solved return once the lag has passed. ROIC each year is after-tax operating profit ÷ average invested capital. A return below the average lowers ROIC; whether it creates value depends on the cost of capital, which this tool does not estimate.
Fiscal years end June 30 for Microsoft and May 31 for Oracle. $ billions, approximate.
Sources. Microsoft Form 10-K for fiscal 2026 (filed July 29, 2026); Oracle Form 10-K for fiscal 2026 (filed June 22, 2026) and Form 10-Q for the quarter ended August 31, 2026; Microsoft earnings call transcripts posted on its investor relations site. All measured figures are as of each company’s fiscal 2026 year-end.
Profit. After-tax operating profit (NOPAT) is operating income plus the interest inside operating lease cost, estimated as the average operating lease liability × the weighted-average discount rate, taxed at the reported effective rate. Without the add-back, fiscal 2026 average ROIC is ~30.2% rather than ~30.4% at Microsoft and ~12.0% rather than ~12.7% at Oracle. Oracle’s effective rate was ~12.6%; at 21% its fiscal 2026 ROIC would be nearer ~11–12%.
Capital. Invested capital is net property and equipment, operating lease right-of-use assets, goodwill and acquired intangibles, and operating working capital (current assets less cash and short-term investments, less current liabilities other than debt and lease liabilities). Both companies already include finance lease assets in property and equipment, so they are not added again. Equity investments, deferred taxes and other long-term items are left out. The goodwill-excluded view removes goodwill and acquired intangibles and adds back after-tax amortization. Microsoft’s ~$23B rise in other current assets stays in working capital; the 10-K does not explain it, though separately it attributes a ~$22B increase in other investing outflows mainly to facilitating component purchases.
Financing-like liabilities. Capital spending still unpaid at year-end is financing of capital, not operating working capital, so it is added back to invested capital: Microsoft’s purchases of property and equipment in accounts payable (~$4.3B at fiscal 2024, ~$6.9B at 2025, ~$26.7B at 2026) and Oracle’s unpaid capital expenditures (~$0.6B at fiscal 2023, ~$1.6B at 2024, ~$3.0B at 2025, ~$5.3B at 2026). Microsoft does not disclose the fiscal 2023 figure; the goodwill-excluded fiscal 2024 ROIC is ~86–88% depending on what is assumed for it, shown as ~87%. Oracle received ~$4.6B of customer prepayments with a significant financing component in fiscal 2026, which it treats as financing; its current deferred revenue rose only ~$0.5B, so most of it sits in non-current liabilities, which invested capital already leaves out, and no separate adjustment is made. Neither filing identifies a vendor financing balance. Customer-provided hardware never reaches the balance sheet, so it would raise reported returns without any adjustment being possible.
Return on new capital. Measured only after each company’s last large acquisition: Oracle’s Cerner (closed June 2022) and Microsoft’s Activision Blizzard (closed October 2023). Capital is measured from Oracle’s fiscal 2023 and Microsoft’s fiscal 2024 year-ends. With a lag of L years, profit is measured from L years after that base to fiscal 2026, and capital from the base to L years before fiscal 2026, so the capital window never reaches back into an acquisition year. The measure credits all profit growth to new capital, including growth that needed little capital, such as price increases on existing software, so it can overstate the return on the capital itself; capital not yet in service can make it understate. A figure is shown as n/a when the profit window is under a year or the capital change is under ~10% of the base. Profit is taxed at each year’s reported effective rate, so tax swings move the result: holding each company’s fiscal 2026 rate constant gives ~18% and ~30% for Microsoft (same year, one-year lag) and ~8% and ~18% for Oracle (~12% excluding construction in progress). Oracle’s fiscal 2023 rate was unusually low, at ~6.8%. Microsoft’s fiscal 2024 holds about 8½ months of Activision and fiscal 2025 a full year, so its window includes about 3½ months of acquired profit; Microsoft does not quantify it, and it is likely a small share of the change. Microsoft’s fiscal 2024 average ROIC is also flattered by averaging capital across the acquisition, which is why frame 3 starts the reported line at fiscal 2025.
Construction in progress. Oracle discloses it (~$40B at fiscal 2026 year-end, ~$48.5B by August 31, 2026); Microsoft does not. The excluded view is shown for Oracle at the same-year measure only, because a lag already corrects for capital not yet earning.
Leases not yet started. Microsoft ~$329.1B, commencing fiscal 2027 to 2033; Oracle ~$260B at May 31, 2026 (~$288B by August 31), commencing fiscal 2027 to 2029. These are treated as total payments. The stress view converts them to present value at each company’s own ratio of finance lease liability to total finance lease payments (~74% Microsoft, ~67% Oracle) and adds them evenly over the commencement years. This is a stress view, not a forecast.
Useful life. Microsoft’s chief financial officer said on the fiscal 2026 fourth-quarter call that from the start of fiscal 2027 datacenters and office buildings move from 15- to 25-year lives, with a minimal benefit to fiscal 2027 operating income and a larger effect on reported capital spending as more future leases become operating leases. The fiscal 2026 10-K still shows buildings at 5 to 15 years, and the 10-Q for the September 2026 quarter does not report it as a change in estimate. The same transcripts describe roughly two-thirds of quarterly capital spending as short-lived assets in the last three quarters of fiscal 2026, up from roughly half in the first.
Back-solve. With a lag of L years, fiscal 2026 profit is taken to reflect capital in place L years earlier. Capital added since then, including fiscal 2026’s additions, earns the solved return once the lag has passed, which keeps the back-solve consistent with how return on new capital is measured. An earlier version gave recent capital no credit, which overstated the return needed under a lag. Holding an average requires new capital to earn about the average, so a figure below it means ROIC falls, not that the spending destroys value; that depends on the cost of capital, which this lesson does not estimate. Capital growth is an input, not a forecast. Concepts follow Koller, Goedhart and Wessels, Valuation (McKinsey & Company, 7th edition). Not investment advice. Not affiliated with any company named.
Since fiscal year-end (as of October 31, 2026; not used in the calculations). Microsoft’s 10-Q for the quarter ended September 30, 2026 shows leases not yet started of ~$365B, now commencing fiscal 2027 to 2034; cash capital spending of ~$39B, about double a year earlier; operating lease assets obtained of ~$6.5B against ~$1.4B a year earlier and finance lease assets of ~$5.2B against ~$9.1B, consistent with the shift to operating leases; and ~$26.5B of equipment purchases still in accounts payable. Oracle’s 10-Q for the quarter ended August 31, 2026 shows leases not yet started of ~$288B, construction in progress of ~$48.5B, unpaid capital expenditures of ~$6.2B and ~$11.4B of further customer prepayments with a financing component. The lesson will roll forward to fiscal 2027 when both 10-Ks are filed.
Oracle’s guidance. At its October 28, 2026 Investor Day, Oracle guided to fiscal 2030 revenue of $225B or more at a non-GAAP operating margin of 40% or more, with capital spending peaking in fiscal 2027–2028 and free cash flow of $30B or more in fiscal 2029. That is, in effect, a claim that returns on today’s capital arrive with a lag. As a rough check in the back-solve’s profit-growth mode, ~35% yearly profit growth from fiscal 2026 to 2030 needs new capital to earn roughly 8–11% if capital grows ~40% a year, or roughly 17–18% at ~25% (same year to one-year lag), against ~7–17% measured. Guidance is non-GAAP while this lesson uses GAAP profit, and capital growth won’t be constant, so read this as a scenario, not a finding. Microsoft has given no comparable long-range guidance.