AI Infrastructure Cycle
Expansion continues.
Cash collection faces a test.
Connecting semiconductor, server and cloud disclosures
to assess current conditions, conditional quarterly forecasts and historical forecast errors.
Industry activity is rising.
Slower growth does not mean contraction.
Current data support continued expansion in AI infrastructure. However, the next investment differentiator is less the volume of orders than usage, profit and cash collection after delivery. A high industry score does not establish a probability of rising share prices.
Versus the previous quarter +15.0%
Supply-chain sample index, 2020Q4=100
Of the 4 businesses, 3 show accelerating growth
Above 50: acceleration outweighs deceleration
Cloud segment profitability is improving
Financing and payment terms need separate scrutiny
Composite annual growth slowed from 121.6% to 92.3%. Excluding Dell, however, it accelerated from 66.9% to 84.8%. Dell's revenue itself also rose 9.1% quarter over quarter. A decline in aggregate growth does not mean demand is falling across the supply chain.
This analysis covers the nominal revenue cycle of semiconductor, server and cloud businesses connected to AI. The indices do not measure the AI-only market size, shipments or stock returns. The 4 businesses also include non-AI revenue.
Demand is evident in results and upcoming delivery plans.
Financial pressure appears in different indicators.
The calculation centers on quarterly revenue for 4 businesses linked to 116 official earnings disclosures. Growth rates use the prior-year comparatives in the same release. Orders, guidance, profit and cash flow provide cross-checks against actual business conditions.
| Business / latest common quarter | Revenue USD billions | Year over year | Quarter over quarter | Change in annual growth | Interpretation |
|---|---|---|---|---|---|
| NVIDIA Data Center [N] | 89.023 | +116.6% | +18.3% | +24.2%p | Accelerating |
| Dell Servers and Networking [D] | 26.932 | +108.1% | +9.1% | -182.3%p | Decelerating |
| AWS [A] | 42.232 | +36.8% | +12.4% | +8.4%p | Accelerating |
| Google Cloud [G] | 24.768 | +81.8% | +23.7% | +18.4%p | Accelerating |
1. Upcoming production and deliveries also point to expansion
Dell raised its FY27 AI server revenue outlook from $60 billion to $74 billion. This is not year-over-year growth; it is a 23.3% increase in the outlook for the same year. TSMC's Q4 revenue guidance midpoint is $53.3 billion, 14.1% above Q3. [D][T]
Interpretation: Near-term production and delivery plans corroborate strong current results. Orders and backlog, however, are not the same as revenue or cash.
2. Monetization is progressing; cash collection is a separate question
AWS's operating margin improved from 32.9% a year earlier to 39.4%, and Google Cloud's from 20.7% to 35.6%. Amazon's group-wide trailing-12-month free cash flow, meanwhile, was −$7.6 billion. [A][G]
Interpretation: Cloud profitability has improved, but this does not establish that group capital expenditure has been recovered. Segment operating margins cannot measure the return on all AI investment.
3. The supply chain is also exposed to customers' financing capacity
NVIDIA's days sales outstanding rose from 45 to 60 days. The company attributed this to extended payment terms on long-term contracts with certain investment-grade customers. This is not evidence of delinquency, but it shows that revenue growth and cash collection are not moving at the same speed. [N]
CoreWeave's quarterly net interest expense equals 24.9% of revenue. For businesses of this kind, the question is whether usage revenue and financing keep pace with expanding capacity. [C]
Figures for different businesses are not added together. TSMC includes non-AI semiconductors, Dell includes conventional servers, and Google Cloud includes Workspace and TPU product sales. Transactions along the same supply chain are recognized at multiple stages; more observations do not establish an equivalent increase in independent final demand.
One index measures the scale of revenue;
the other measures how broadly growth is accelerating.
A. Supply-chain revenue activity index
Each business's revenue is compared with 2020Q4, then combined as a weighted product. A rise from 100 to 110 means revenue activity in this sample increased 10%. Today's reading of 1,100 is approximately 11 times the base-quarter level. It is not a rating out of 100.
B. Growth breadth index
If annual growth increases by more than 3 percentage points from the previous quarter, s=+1; if it decreases by more than 3 percentage points, s=−1; otherwise, s=0. 100=all accelerating; 50=acceleration and deceleration balance; 0=all decelerating. A reading of 66.7 does not mean a 66.7% probability of an increase.
| Role | Business | Weight | Reason for inclusion |
|---|---|---|---|
| Compute | NVIDIA Data Center | 1/3 | Compute infrastructure revenue |
| Systems | Dell Servers and Networking | 1/3 | Server deployment revenue |
| Services | AWS + Google Cloud | 1/6 each | Cloud business revenue |
The 1/3 weight per role is an analytical choice, not market share or an estimated optimal weight. Azure is excluded because a consistent history of absolute revenue is difficult to obtain. Changing the current acceleration threshold to 2, 3 or 5 percentage points leaves the reading at 66.7. The long-run index depends on the 4 businesses selected today and should not be generalized to the whole industry.
A rising + B>50: expansion and acceleration coexist. A rising + B<50: revenue increases, but growth slows. A falling + B<50: examine whether activity is contracting. A decline in B alone does not establish falling revenue or share prices.
There is evidence for near-term expansion.
The scale a year from now depends heavily on sustained growth assumptions.
The reference outlook assumes company guidance is achieved and the annual growth rates derived from it are maintained. This is a conditional calculation aligned with near-term business plans, not a validated probability forecast for the market.
| Target quarter¹ | Activity index 2020Q4=100 | Versus latest actual | Historical error reference range² | Quarterly assessment |
|---|---|---|---|---|
| 2026Q4About 3 months | 1,453 | +32.0% | 1,209–1,938 | Near-term disclosures support continued expansion |
| 2027Q1About 6 months | 1,890 | +71.8% | 1,427–2,820 | Verify delivery and installation schedules and continued guidance |
| 2027Q2About 9 months | 2,174 | +97.6% | 1,740–3,453 | Increasingly dependent on sustained high growth |
| 2027Q3About 12 months | 2,281 | +107.3% | 1,758–3,623 | Most dependent on financing and investment recovery |
¹ 2026Q4 combines the cloud quarter ending December 2026 with NVIDIA's and Dell's quarters ending January 2027. Later quarters follow the same convention. The 3-, 6-, 9- and 12-month horizons identify quarter windows relative to the issue month, not revenue on an exact date. The still-unconfirmed 2026Q3 bridge estimate is an activity index of 1,154.
² The shading and ranges are calculated by averaging historical baseline actual/predicted ratios within each target quarter, taking their 10th and 90th percentiles, and applying them to the current conditional outlook. They are neither validation results for the current guidance-based outlook nor 80% probability intervals. Actual outcomes may fall below the lower bound.
Three calculations connecting the outlook
| Input | Calculation | Interpretation |
|---|---|---|
| NVIDIA near-term anchor | Next-quarter total revenue guidance midpoint 108 (±2%) × current DC share 89.023/96.221 = DC revenue of $99.921 billion [N] | Total guidance is company-provided. Holding the DC share constant is an analytical assumption. Dividing by 51.215 for the year-earlier quarter gives an annual multiplier of 1.951. |
| Dell second-half anchor | Full-year AI 74 − first-half AI 32.533 + conventional servers 10.531×2 = Second-half Servers and Networking revenue of $62.529 billion [D] | The full-year AI outlook is company-provided. Constant quarterly conventional-server revenue and allocation across the second half are analytical assumptions. Relative to 24.930 in the prior-year second half, the multiplier is 2.508. |
| Cloud and subsequent quarters | Annual multiplier: AWS 1.368 / Google Cloud 1.818 Each business's year-earlier quarterly revenue × its multiplier | AWS and Google retain the latest disclosed growth rates; NVIDIA and Dell retain rates derived from the anchors above. For calculations extending beyond a year, the prior-year value is an earlier forecast. |
What the numbers mean: 1,453 → 1,890 → 2,174 → 2,281 are revenue activity levels for different quarters. Composite annual growth is the same 97.6% throughout this calculation because a constant growth rate is assumed. It does not mean growth momentum at four separate dates has been independently predicted. Differences in quarter-over-quarter changes also inherit the prior year's quarterly revenue pattern.
The near-term assessment is therefore 'continued expansion'; the 6–12-month assessment is 'conditional on sustained high growth.' Longer-term figures serve as a reference path against which to check each quarter's results, rather than investment targets.
We reconstructed forecasts at monthly dates.
The ability to identify upward turning points remains unproven.
From March 2021 through October 2026, we reconstructed 68 monthly observation dates × 4 forecast horizons. Only information published by each date was used. Excluding outcomes still in the future, evaluation covers 18–21 target quarters per horizon. Repeated monthly forecasts of the same quarter are averaged within that quarter first, so repeated observations do not inflate performance.
| Forecast horizon | Target quarters / monthly records | Baseline Level error³ | Growth-trend model Level error³ | Baseline Activity direction accuracy | Always predict an increase Direction accuracy |
|---|---|---|---|---|---|
| 3 months | 21 / 62 | 15.6% | 16.0% | 84.1% | 90.5% |
| 6 months | 20 / 59 | 19.9% | 22.0% | 91.7% | 95.0% |
| 9 months | 19 / 56 | 21.2% | 25.0% | 91.2% | 94.7% |
| 12 months | 18 / 53 | 23.0% | 30.5% | 100.0% | 100.0% |
³ Level error is the mean of |predicted index / actual index − 1|; lower is better. The baseline holds each company's latest annual growth rate constant; the trend model also incorporates recent changes in growth. Activity direction is classified by whether the future actual index moves beyond ±2% of the latest common index known at issue. The current guidance-based outlook does not inherit this historical performance.
What the validation supports
A high proportion of correct upward calls alone does not establish predictive skill. The baseline's 100% 12-month direction accuracy matches an 'always up' forecast. The more complex trend model also failed to beat the baseline on full-period level error. At each of the 3- and 6-month horizons, 3 monthly records were classified as actual declines; the baseline missed all of them. These are repeated records of the same contraction, not 3 independent decline events. The evidence therefore does not justify approving this as an automated trading or peak-and-trough forecasting tool.
Actual contraction differed from growth normalization
| Historical period | Revenue activity index | Growth breadth index | Lesson for interpretation |
|---|---|---|---|
| 2022Q3 → 2023Q1 | 159.6 → 152.7 (−4.3%) | 25.0 → 33.3 | Revenue activity actually fell. This is the contraction observed in the sample. |
| 2024Q2 → 2025Q1 | 385.0 → 431.7 (+12.1%) | 50.0 → 16.7 | Fewer businesses showed accelerating growth, but revenue kept rising. Reading lower breadth as a recession would be a mistake. |
These are retrospective case studies. Economic quarters and earnings release dates differ, so the cases are not counted as having predicted a decline a specified number of months ahead. Too few evaluated decline records exist to generalize downside-detection ability.
Additional model comparisons, recent performance and monthly reconstruction
Do TSMC and capital expenditure improve forecasts?
We also compared a model that gives equal weight to company revenue paths and a regression on monthly indicators. It uses revenue growth and its change, annual growth in TSMC's latest three months of revenue and its change, capital expenditure growth, and the forecast horizon for each observation. It is retrained using only actual values known at each date.
| Horizon (months) | Full period: combined / simple baseline | Since 2025: combined / simple baseline | Target quarters: full / recent |
|---|---|---|---|
| 3 | 23.8 / 31.7 | 28.7 / 26.0 | 13 / 6 |
| 6 | 31.9 / 43.3 | 38.4 / 36.1 | 11 / 5 |
| 9 | 26.6 / 39.5 | 47.1 / 43.7 | 9 / 4 |
| 12 | 32.2 / 42.1 | 56.6 / 47.5 | 7 / 3 |
The table reports mean absolute error in annual growth, in percentage points. It is not directly comparable with index-level error (%) in the main table. Each cell compares matched issue dates and target quarters. Because the combined model requires training data, its evaluation sample is shorter: 7–13 target quarters.
Over the full period, the combined model's error was 23.5–32.8% lower than the simple baseline's. Since 2025, however, it was 6.2–19.1% higher at every horizon. There is no evidence that adding more data always improves forecasts. We do not selectively attribute these results to the current outlook.
Recent performance of the same simple baseline and trend model
In the recent period, baseline level error is 14.1–22.5%. With only 3–6 target quarters per horizon, the sample is insufficient for precise performance comparisons. We do not select only favorable results from the full or recent period.
| Horizon (months) | Target quarters | Baseline level error | Trend-model level error | Baseline annual-growth error |
|---|---|---|---|---|
| 3 | 6 | 14.1% | 13.2% | 26.0%p |
| 6 | 5 | 18.8% | 18.3% | 36.1%p |
| 9 | 4 | 22.5% | 22.2% | 43.7%p |
| 12 | 3 | 21.2% | 31.9% | 47.5%p |
This view shows the baseline that can be reconstructed consistently at every date. Future periods without reported results remain 'Unconfirmed.' These are research reconstructions made now, not a record of forecasts actually published at those dates.
Did forecasts for the same quarter improve over time?
Comparing the earliest and latest forecasts, absolute level error declined in 7 of 21 target quarters. Because different forecast horizons are compared together, this is not a causal estimate of the benefit from additional information alone.
Revenue → usage and profit → cash.
Watch where the connection weakens.
We define in advance the conditions for maintaining or lowering the current expansion assessment. These are monitoring rules, not historically optimized buy/sell thresholds. One indicator at one company does not change the assessment of the entire supply chain.
| Priority | What to monitor | Signals supporting continued expansion | Signals to lower the assessment | Meaning |
|---|---|---|---|---|
| 1 · Deliveries | NVIDIA's next quarter, Dell's annual AI outlook, TSMC Q4 | Guidance maintained and achieved; deliveries on schedule | Guidance cuts and delivery delays across several companies | Are orders converting into revenue? |
| 2 · Growth breadth | Activity index A and breadth index B | A rises; B remains above 50 | B drops below 50, followed by a decline in A | Is slower growth becoming an actual revenue contraction? |
| 3 · Cash collection | NVIDIA's DSO trend after 60 days, receivables and operating cash flow | Cash collection improves alongside revenue growth | Further lengthening of collection periods alongside slower revenue growth | Is more financial support needed to sustain contract growth? |
| 4 · Customer economics | Cloud usage revenue, segment profit and capacity utilization | Usage revenue and profit rise together | Profitability weakens despite higher utilization; investment plans are cut | Can customers cover the cost of new capacity? |
| 5 · Financing costs | Interest, maturities and new financing terms at debt-funded AI clouds | Operating cash generation covers more interest and expansion spending | Worsening refinancing terms alongside expansion delays | Can funding constraints stop deliveries and investment even when demand exists? |
| 6 · Supply execution | Delivery schedules for power connections, memory, advanced packaging and cooling | Installation and commissioning align with delivery plans | Bottlenecks repeatedly delay deliveries or revenue recognition | Can weak orders be distinguished from supply and installation delays? |
Industry assessment
Current state: expansion. Near-term order and production plans are favorable. A lower breadth index alone does not establish a downcycle. Lower the assessment when guidance cuts, revenue contraction and deteriorating cash collection occur together.
Equity-market assessment
Industry growth and rising share prices are separate questions. One must also assess whether high expectations are already priced in, whether earnings forecasts are rising and whether discount rates are changing. The current dataset lacks histories of share prices, valuations and consensus estimates, so it produces neither share-price targets nor probabilities of appreciation.
For investors, the key is to go beyond asking whether AI spending will keep rising. Who converts revenue into profit and cash, and who also assumes the burden of financing customers? These differences matter within the same expansion. This is an analytical interpretation of linked disclosures, not a conclusion about the valuation appeal of a particular stock.
Formulas, sources and reproduction requirements
| Official source | Use | Release date | Covered period | Original |
|---|---|---|---|---|
| NVIDIA | Calculation inputs, guidance and collection terms | 2026-08-26 | Q2 FY27 / ended July 26 | [N] |
| Dell | Calculation inputs, orders and annual guidance | 2026-09-01 | Q2 FY27 / ended July 31 | [D] |
| Amazon | AWS calculation inputs, segment profit and group cash flow | 2026-07-30 | Q2 2026 / ended June 30 | [A] |
| Alphabet | Cloud calculation inputs, segment profit and business definition | 2026-07-22 | Q2 2026 / ended June 30 | [G] |
| TSMC | Cross-check of recent manufacturing activity and near-term guidance | 2026-10-15 | Q3 results and Q4 guidance / USD | [T] |
| CoreWeave | Cross-check of financing pressure at debt-funded AI clouds | 2026-08-11 | Q2 2026 / ended June 30 | [C] |
Calculation definitions and information timing
Composite annual growth
Gq = Σiwigi,q · annual growth = 100 × (exp(Gq) − 1)
Growth calculated from comparatives in the latest release can differ from the year-over-year change in the activity index, which links each quarter's initially reported revenue, because historical figures may be restated. The files distinguish these definitions. The index is in nominal dollars; inflation, exchange-rate and product-mix effects are not removed.
Trend-model comparison formula
ĝi,q+j = gi,q + di(1 − 0.5j)
R̂i,q+j = Ri,q+j−4 × exp(ĝi,q+j)
This comparison model incorporates recent changes in growth over two quarters to a limited extent. Its full-period level error exceeded that of constant annual growth at every horizon. The current outlook therefore uses a simple baseline connected to guidance. This choice was made after examining historical data, so this is not an independent, previously unused validation sample.
Release timing and monthly records
At each month-end in KST, data are used only after 23:59:59 on the day following the official release date has passed. This conservative convention avoids treating a date-only disclosure as information available intraday. The current issue timestamp is 2026-10-18 23:43:10 KST. The observed index stays unchanged in months without new quarterly data; no nonexistent monthly revenue is created.
The target is the quarter containing 'issue month + horizon − 1 month.' When the four businesses report in different months, forecasts use each company's latest information, while the observed index uses the latest quarter confirmed for all four. Monthly comparisons begin in March 2021, after the 2020Q4 base-quarter data became available.
Historical disclosures were collected again in 2026. This is not a real-time database that can fully establish preservation of every original publication and the absence of subsequent revisions. Companies, weights and models were reviewed on the same historical record, leaving selection bias and retrospective design risk. All records are labeled as reconstructed research to avoid presenting false forecasting performance.
Basis and limits of the outlook; requirements for further validation
Current guidance is not a confirmed result. NVIDIA's DC share and the assumptions for Dell's conventional-server revenue may change. The original constant-growth estimate for NVIDIA's next-quarter DC revenue was $110.9 billion, above even the $110.16 billion upper end of total company revenue guidance. We therefore did not use it in the current reference scenario, which assumes company guidance is achieved. Actual results can exceed guidance, so we do not call it an accounting impossibility.
Claiming comparable historical performance for the current guidance-linked model requires histories of total revenue, business mix, quarterly guidance and annual guidance available at each monthly date. We did not fill missing historical records with today's information. Nor do we imply that TSMC and cash-flow data were inputs to the historical forecast formula when they were not.
To broaden its use as a forecasting model, inputs and forecasts must be frozen at each future issue date and compared with actual results for the same target quarters. Evaluation should cover guidance-model errors, improvement over the simple baseline, downside detection, false alarms and empirical coverage of forecast ranges. Equity forecasting requires separate validation with total returns including dividends, market-relative returns, contemporaneous valuations and earnings-estimate revision histories.
Report, monthly forecasts and source files
Download the report HTML and CSV files for quarterly indexes, monthly forecasts, the current outlook and official sources. Records for all 68 monthly observation dates are embedded in the HTML and can be explored offline.
The CSV files distinguish target quarters and issue dates, forecasts and actual results. Use the unrounded CSV values for calculations rather than rounded screen values. This distribution bundle contains the report and data exports; it does not include the full executable model source.