The AI Boom’s Two Engines: Capital Spending, Household Wealth and the Reversal Risk
Reading the counterfactual without mistaking it for a forecast

The artificial-intelligence boom is supporting economic activity through two distinct mechanisms: companies are buying and constructing capital, while households are responding to higher equity wealth. Those mechanisms share an optimistic narrative, but they should not be treated as interchangeable. One concerns current expenditure on infrastructure; the other concerns the present value of expected corporate earnings. A spending boom can therefore provide a modest direct addition to domestic production while simultaneously generating a much larger exposure to changes in financial sentiment. Jason Furman's FRB/US exercise is useful precisely because it separates these channels rather than assuming that every dollar attached to artificial intelligence enters national output in the same way.
The reported estimates put real GDP in Q3 2026 about 0.3% above an economy without the data-center boom, with unemployment roughly 0.1 percentage point lower. Including AI-related equity appreciation increases the corresponding effects to approximately 0.6% and 0.3 percentage point. These are differences from a counterfactual economy, not statements that GDP grew at those rates or that unemployment reached those levels. An economy with strong underlying growth could exhibit these increments, but so could an otherwise weak economy. The baseline is indispensable: the experiment isolates a modeled contribution rather than describing the entire state of the business cycle.
The accompanying chart makes this distinction visible. Its left panel measures the percentage difference in the level of real GDP from an economy without the boom. Its right panel measures the unemployment-rate difference in percentage points. Blue lines represent data centers alone; orange lines add the equity channel. The orange output line sits above the blue line, while the orange unemployment line sits below it. Those opposite visual directions describe the same favorable demand impulse: more production and less unemployment. The two vertical scales are not directly comparable, and a 0.3% GDP difference is conceptually different from a 0.3-percentage-point unemployment difference.
The solid portions of the curves turn into dashed continuation paths. That transition matters more than the visual smoothness of the lines. If data-center investment remains around 1.6% of GDP and equity gains remain near early-November levels, the combined exercise places GDP approximately 0.75% above the no-boom baseline and unemployment approximately 0.45 percentage point below it at the end of 2028. The chart's investment-only path ends near a 0.44% GDP increment and an unemployment reduction around 0.25 percentage point. These approximate chart readings are not additional precision beyond the reported scenario. They show that the investment channel continues to matter even when its immediate domestic content is limited.
A higher level is not a permanently higher growth rate. If the GDP gap rises from 0.6% to 0.75% over a subsequent interval, the incremental growth contribution over that interval is related to the change in the gap, not the full 0.75%. More formally, the approximate growth difference equals the change in the log output gap between the boom and no-boom paths. A stable positive gap can coexist with identical growth rates in the two economies. This distinction prevents a common mistake: adding the endpoint level effect to every year's growth forecast and thereby compounding a stimulus that the experiment does not actually contain.
The FRB/US framework is an appropriate tool for asking how specified expenditure and asset-price paths propagate through a large economy. The Federal Reserve describes it as a large-scale estimated general-equilibrium model used for forecasting, policy analysis and research. That description establishes the framework's purpose, not an official endorsement of this particular application. Private analysis conducted with the model is not a Federal Reserve forecast. The exercise also requires assumptions about expectations, interest rates, price adjustment and policy behavior. Without the underlying configuration and simulation files, readers cannot know how sensitive these particular results are to every closure choice. [Federal Reserve model overview](https://www.federalreserve.gov/econres/us-models-about.htm).
There is also a causal-identification problem that no polished chart eliminates. Equity prices may rise because investors anticipate productivity improvements, lower discount rates, stronger aggregate demand, or unusually high future profits. Some of those developments could have raised spending even without a separately identified AI shock. Treating an equity basket's outperformance as a wealth input is a useful scenario construction, but it is not experimental proof that the entire price gain was caused by one technology. Similarly, an economy without the investment boom might redirect labor, capital and imports elsewhere. The counterfactual must account for displacement if the desired question is net national benefit rather than gross sectoral activity.
The unemployment estimates also require care. A reduction of 0.45 percentage point does not mean a 45% fall in unemployment, nor does it establish how many permanent technology jobs were created. Aggregate demand can support employment in restaurants, construction, logistics and other services far from a data center. Labor-force participation, hours worked and productivity affect the relationship between output and unemployment. The chart therefore supports an economy-wide transmission argument, not a simple employment-per-server calculation. Its increasingly negative unemployment gap alongside a flattening output gap also warns against inferring one fixed output-to-employment coefficient from the endpoints.
The most productive reading is consequently conditional and comparative. Under the stated inputs, infrastructure expenditure helps, and adding equity wealth increases the modeled support substantially. That finding is meaningful without claiming that current valuations are justified, that the investment is socially optimal, or that recession risk has disappeared. It identifies where the current demand support may reside and where the economy could become sensitive to disappointment. The rest of the analysis should therefore ask three separate questions: how much spending becomes domestic value added, how reliably equity wealth turns into consumption, and whether productive returns eventually validate the expectations sustaining both channels. Keeping those questions separate is the first requirement for understanding the boom rather than merely extrapolating it.
Why a large investment boom produces a smaller domestic output impulse
The investment numbers initially look inconsistent with the output results. The reported surge reached 1.3% of GDP, yet the estimated Q3 2026 real-output benefit from data centers alone is approximately 0.3%. The main reconciliation is the assumption that roughly 75% of the relevant investment was imported. National accounting measures domestic production rather than the nationality of the purchaser. An American company can undertake enormous capital expenditure without creating an equally large contemporaneous addition to American value added. That is not evidence that the equipment is useless or that the investment has no future return. It is a distinction between where a capital good is produced and where its productive services will subsequently be used.
The expenditure identity Y = C + I + G + X - M clarifies the arithmetic. Imported servers enter investment, I, but their foreign-produced value also enters imports, M. The subtraction prevents foreign output from being counted as domestic output; it does not mean imports independently destroy GDP. Under the simplified assumption that three quarters of the 1.3% investment increment represents foreign production, the immediately retained domestic component is 1.3% × (1 - 75%) = 0.325% of GDP. That rough calculation helps explain the scale of the reported 0.3% effect. It is an accounting illustration, not a replication of a dynamic macroeconomic simulation or an independently estimated investment multiplier.
Several qualifications keep this arithmetic honest. The domestic share may include construction, engineering, installation, power connections, transport margins and other services, not simply domestically manufactured chips. Imported components can also be embedded inside equipment assembled in the United States. The relevant concept is domestic value added, not the location of the final invoice or the headquarters of the vendor. Moreover, nominal expenditure shares cannot be mapped mechanically into chained real GDP contributions without attention to prices and statistical measurement. Rapid changes in equipment quality make that distinction especially important. A faster processor can represent more real capital services even if its dollar price changes little.
There is a timing issue as well. Equipment orders, deliveries, payments, construction work and operational commissioning need not occur in the same quarter. Inventory accumulation can temporarily separate production from final installation. Some spending has already occurred before the facility supplies economically useful computing services. Other expenditures, such as electricity and maintenance, arise only after operation begins and may be classified differently. A model's quarterly output response therefore reflects a sequence of income and demand effects, not a photograph of invoices at one point in time. Comparing one annualized spending share with one quarterly counterfactual output gap is informative only if that difference in concepts remains explicit.
The next layer is the multiplier, which depends on slack, imports, financing and policy. Domestic construction income can support further consumption, but some of that consumption leaks into imports or saving. If skilled electricians and grid equipment are scarce, more spending may raise prices and delay unrelated projects rather than expanding real output proportionately. If interest rates rise in response to stronger demand or inflation, rate-sensitive housing and business investment can be displaced. A sector may consequently experience dramatic expansion while its net economy-wide effect remains smaller. Conversely, idle local labor or underused supplier capacity can allow a larger real response than a simple domestic-share calculation would suggest.
Consider a deliberately hypothetical project with a total cost of 100 units. Suppose imported equipment accounts for 75 units and domestic construction and services for 25. Paying the domestic contractors does not establish a 100-unit current contribution to domestic production, and counting their subsequent spending without subtracting displaced activity would overstate the multiplier again. But the project may later provide computing services worth more than its operating and capital costs. Those future services can add domestic value even though the original hardware was imported. The proper assessment separates the current demand impulse, the future stream of capital services, and the distribution of income between local workers, asset owners and foreign suppliers.
The continuation assumption raises the investment share to roughly 1.6% of GDP, but a stable high flow of spending is different from continuously accelerating spending. Capital accumulation can continue while its contribution to the change in aggregate demand diminishes. In an illustrative constant-price economy, moving investment from 100 to 120 adds 20 to expenditure growth; keeping it at 120 thereafter does not add another 20 each period. Replacement spending further complicates interpretation. If technological obsolescence is rapid, maintaining the same effective capital stock may require large gross purchases. High headline expenditure can then coexist with much smaller additions to net productive capacity and less incremental construction employment than during the initial buildout.
Import dependence also changes the geography of risk. Foreign suppliers may capture a substantial share of the initial production benefit, while American households retain a substantial share of the equity exposure. Domestic demand can therefore become financially sensitive to a boom whose physical supply chain is global. A reversal could weaken overseas equipment orders before it fully affects domestic consumer services, or disrupt domestic projects through supplier distress. Exchange rates, tariffs and export restrictions can alter costs without increasing usable computing capacity. None of these possibilities is quantified by the 75% assumption alone. They explain why domestic-content estimates deserve sensitivity analysis rather than treatment as immutable technological constants.
Finally, the public-policy question should not be reduced to maximizing the domestic share of every purchase. Imported capital may improve efficiency, lower costs and accelerate adoption; costly localization can increase measured domestic expenditure while reducing economic returns. The relevant test is whether the available combination of equipment, energy, labor and complementary investment creates valuable services at sustainable prices. For macroeconomic monitoring, the priority is to distinguish gross commitments from actual domestic production and net capacity creation. That distinction explains why a spectacular spending cycle need not dominate GDP immediately. It also prevents the opposite error of dismissing imported capital as economically irrelevant when its eventual contribution depends on how productively it is deployed.
The household wealth channel and the economics of concentrated gains
The larger financial channel begins with a counterfactual asset-price calculation. Seven AI-related companies reportedly gained about $11 trillion more in market value than they would have gained by tracking the rest of the equity market since ChatGPT's launch. Estimated American household ownership of those excess gains is $7.6 trillion. These figures are not corporate revenue, cash transferred to households, or additional national saving. They are changes in the market value of claims on future earnings relative to a benchmark. Their macroeconomic relevance arises because households may change spending when the value of their balance sheets changes, even before they sell the assets.
The benchmark matters. Comparing the seven companies with the rest of the market removes some common equity appreciation, but it does not perfectly isolate artificial intelligence. Differences in profitability, market power, duration, business mix and interest-rate sensitivity can all generate relative performance. Calling the residual AI-related is therefore an economically motivated classification rather than a clean causal estimate. The household ownership calculation introduces another layer: direct holdings, pooled funds and retirement accounts do not all translate into equally accessible spending power. Foreign ownership and institutional claims also affect how much of the capitalization gain can influence current domestic household expenditure.
A useful theoretical starting point is consumption smoothing. A household that expects higher lifetime resources can afford a somewhat higher consumption path, but it generally will not spend an entire permanent wealth gain immediately. A temporary gain deserves an even smaller response. This is why the relevant parameter is a flow response per dollar of wealth rather than a dollar-for-dollar conversion. In the simple expression ΔC = m × ΔW, m is the annual consumption response and ΔW is the relevant wealth change. The formula is a sensitivity tool, not a claim that one coefficient fits every household, asset category or market regime.
For illustration, applying an assumed annual response of 2% to $7.6 trillion yields about $152 billion of additional spending; 4% would imply about $304 billion. These are illustrative calculations, not Furman's reported consumption estimates, and neither number can be added mechanically to the modeled GDP effect. Some expenditure purchases imports, some substitutes for other spending, and monetary or supply responses can offset part of the impulse. The comparison nevertheless shows why a modest propensity to consume can matter when multiplied by a very large wealth stock. The macroeconomic size comes from the product of the exposure and the behavioral response, not from an implausible assumption that households liquidate all their gains.
Empirical research supports the existence of a stock-wealth channel while rejecting an overly simple representative-household interpretation. Di Maggio, Kermani and Majlesi use Swedish household data to study consumption responses to stock returns and find substantial differences across the wealth distribution. That is evidence that ownership composition can matter; it is not a ready-made coefficient for contemporary American AI shareholders. Different institutions, taxes, liquidity needs and portfolio concentration can change the response. The relevant lesson is methodological: an aggregate wealth number needs to be weighted by who holds it and how those households adjust spending. [Stock Market Returns and Consumption](https://www.nber.org/papers/w24262).
Concentration can dampen the immediate aggregate response while increasing dependence on a narrow group of spenders. Affluent households often have greater capacity to smooth losses and may consume a smaller fraction of additional wealth. At the same time, their absolute spending can be large enough to support travel, leisure, professional services and durable purchases. Workers supplying those activities need not own the original stocks to benefit indirectly. This transmission makes it possible for a concentrated financial gain to support broad service employment without producing an equally broad improvement in household balance sheets. An apparently resilient consumption total can therefore conceal very different experiences across income and wealth groups.
Liquidity and mental accounting further complicate the relationship. A gain inside a retirement account may be treated differently from a gain in a taxable brokerage account, even if both increase measured net worth. Households can finance consumption from current income while saving less, rather than selling appreciated shares. They may also borrow against assets, converting market value into spendable liquidity at the cost of leverage. Taxes and transaction costs can restrain realization, while anticipated bonuses or equity compensation can strengthen the link between technology valuations and current expenditure. None of these mechanisms means unrealized gains are equivalent to disposable income; they explain how balance-sheet changes can still influence decisions.
Persistence is especially important for the continuation scenario. Holding equity gains near early-November levels does not require prices to keep rising at their previous pace. Households can adjust spending gradually to a permanently higher wealth level, allowing demand support to build even after appreciation stops. Conversely, households that regard gains as fragile may wait before changing behavior. Federal Reserve research distinguishes direct wealth effects from the possibility that stock prices merely signal future income, and finds that direct responses can unfold across several quarters. That supports attention to lags rather than a single contemporaneous correlation. [Does Stock Market Wealth Matter for Consumption?](https://www.federalreserve.gov/econres/feds/does-stock-market-wealth-matter-for-consumption.htm).
This channel also creates potential overlap in a sector-by-sector narrative. Higher technology valuations may stimulate consumption, lower corporate financing costs and encourage capital spending at the same time. Adding independently estimated effects from each pathway can double count feedbacks unless a coherent model handles their interactions. The reported difference between the combined and investment-only scenarios is therefore more informative than summing unrelated headline estimates. By Q3 2026, adding equity gains increases modeled GDP support from around 0.3% to 0.6% and unemployment support from around 0.1 to 0.3 percentage point. That comparison indicates a meaningful financial contribution without identifying a universal wealth multiplier. It leaves the central investment question intact: whether the earnings and productivity behind the wealth stock ultimately materialize.
From demand stimulus to durable productivity and defensible valuations
A demand contribution does not settle the productivity question. Building facilities raises expenditure before those facilities demonstrate useful returns, and equity appreciation can support consumption before the earnings behind it are realized. The reported macroeconomic effects should therefore be interpreted as a bridge between expectations and activity, not proof that the bridge reaches a profitable destination. For the boom to become durable, computing capacity must produce services that customers value enough to cover hardware, energy, financing, maintenance and complementary organizational costs. The difficult issue is not whether useful AI applications exist, but whether the incremental returns on the entire investment cycle justify its scale and the market capitalization attached to it.
A standard capital-accumulation relationship helps organize the distinction: K_next = (1 - δ)K + I, where K is productive capital, I is investment and δ is economic depreciation. Gross spending can be large while net accumulation is modest if equipment becomes obsolete quickly. Economic depreciation also need not match accounting depreciation. A machine may remain operational but lose rental value when newer hardware supplies substantially more computation per dollar. Conversely, older equipment may retain profitable uses for less demanding workloads. Investors need evidence on utilization, realized pricing and useful life rather than inferring economic value from either a physical asset count or a chosen depreciation schedule alone.
Capital deepening and total factor productivity are also different concepts. More equipment per worker can increase output per hour without demonstrating that the economy has become more efficient at combining all inputs. A genuine efficiency improvement requires more useful output relative to the full bundle of capital, labor, energy and other resources. This matters because a construction-driven demand impulse can coexist with weak near-term measured productivity, while a later diffusion phase could improve productivity after construction growth slows. Neither pattern is logically contradictory. The challenge is to measure the timing and cost of adoption rather than treating the macroeconomic cycle and the technology's ultimate potential as the same object.
Research on general-purpose technologies offers a relevant, but not unconditional, explanation for delayed gains. Brynjolfsson, Rock and Syverson's productivity J-curve emphasizes complementary intangible investments, including redesigned processes, human capital and new business arrangements. Such adaptation can precede visible improvements in measured performance. This provides a reason not to judge a technology solely from its first spending wave. It does not guarantee that every project will succeed or that all current valuations are reasonable. A lag explanation becomes unfalsifiable if no subsequent evidence is ever required. [The Productivity J-Curve](https://www.nber.org/papers/w25148).
At the firm level, complementary investment means more than buying access to a model. Businesses may need cleaner data, revised permissions, integration with existing software, quality assurance, employee training and changes to responsibility for errors. A tool that saves time on a narrow task may not increase final output if another bottleneck remains binding. For example, reducing drafting time from ten hours to five does not double completed transactions if verification still takes fifteen hours and cannot be accelerated. The economically relevant outcome is the end-to-end improvement after implementation and oversight costs. Task-level demonstrations are informative, but extrapolating them directly to aggregate productivity risks ignoring organizational constraints and the share of work actually affected.
Revenue capture creates a second gap between social usefulness and investor returns. Cheaper AI services can generate large benefits for users while competition compresses suppliers' margins. A supplier can enjoy rapid volume growth yet earn disappointing cash returns if prices decline faster than unit costs or if each capacity expansion requires another costly upgrade. Downstream companies may capture some gains, consumers may receive others, and scarce power or network access may collect rents elsewhere. Broad economic benefits therefore need not validate the valuations of the particular companies that initially attracted investment. An investment thesis should specify who captures value, not merely demonstrate that value exists.
Discounted cash flow makes the valuation hurdle explicit. A simplified perpetual cash flow of 100 valued at a required return of 8% and growth of 3% is worth 100 / (0.08 - 0.03) = 2,000 units. If the required return rises to 9% with growth unchanged, the value falls to roughly 1,667. If growth instead falls to 2% at the original required return, the denominator also becomes 6%, producing the same simplified value. These hypothetical numbers are not valuations of any named company. They illustrate why long-duration equities can lose substantial value from relatively small changes in assumptions even when operations remain profitable and the underlying technology continues to improve.
The macroeconomic interaction is subtle. A credible productivity improvement can increase expected earnings and wealth while reducing unit costs, potentially allowing demand to expand with less inflation pressure. But stronger expected investment returns can also raise the demand for capital and the equilibrium interest rate. The net effect on bond yields, equity multiples and policy is therefore not predetermined. A favorable real-economy story can coexist with a lower valuation multiple if discount rates rise enough. Similarly, successful competition can make the economy more productive while reducing incumbent rents. Treating every positive technology development as automatically positive for every AI-linked security conflates production, distribution and valuation.
Evidence should accordingly move along a sequence from installed capacity to utilization, from utilization to paid demand, and from revenue to sustainable cash generation. Analysts should examine whether recurring customer payments are growing independently of promotional credits, whether customers renew, and whether gross margins survive hardware replacement and electricity costs. Commitments among suppliers and customers should not be assumed to represent independent final demand if they depend on the same financing pool. At the macro level, a widening set of adopting industries and improving output per hour would be more convincing than additional announcements concentrated among infrastructure providers. No single metric resolves the question, but the sequence provides tests that can challenge the narrative.
This is where the source's principal long-run warning becomes decisive. If AI fails to generate enough productivity and revenue to support current investment and valuations, both demand engines become vulnerable at once. If adoption succeeds, however, an eventual moderation in construction need not imply an economic failure: the economy could move from building capacity to using it productively. The distinction is between a healthy transition in the composition of growth and a collapse in the expected returns that financed the buildout. Assessing that distinction requires cash-flow discipline and economy-wide evidence, not allegiance to either a permanently bullish technology story or a blanket claim that all investment booms must end in crisis.
A joint reversal is a different shock from an investment slowdown
A reduction in data-center spending would remove one source of demand. A simultaneous equity correction would weaken another, and their joint occurrence could matter more than either viewed independently. The source describes an illustrative downside in which the buildout ends and the broader stock market falls about 30%, with growth slowing to roughly 1.5%. In that model baseline, the result falls short of recession. This is a conditional simulation, not a minimum growth guarantee. It should not be translated into a claim that the economy can absorb any AI-related collapse without serious damage, especially because the exercise does not fully capture financial-system stress or leverage unwinding.
The definition of the equity shock is crucial. A 30% decline in the broader stock market is not the same as a 30% decline confined to seven AI-related companies, nor is it necessarily a reversal of exactly $7.6 trillion of household gains. The affected asset base, household ownership share and timing determine the wealth loss. Mixing those definitions would produce misleading consumption arithmetic. It is equally important not to read 1.5% growth as a 1.5% contraction or a 1.5-percentage-point loss relative to an unspecified baseline. The reported number is a growth outcome under particular assumptions, and its distance from recession depends on the rest of the economy and the definition of the growth interval.
Co-movement can arise from a common earnings disappointment. If expected future demand for AI services falls, firms may cut equipment orders while investors simultaneously revise earnings and valuation multiples. Lower stock prices can then raise the cost of capital or reduce the attractiveness of issuing equity, reinforcing the investment slowdown. Households can respond to weaker wealth by spending less, reducing demand for businesses far beyond technology. Those firms may cut hiring, further weakening income and spending. The initial shock is therefore sectoral, but the propagation is not. The common narrative that encouraged both capacity expansion and wealth creation can become a common source of contraction.
The variance of aggregate exposure offers a compact way to express the issue. If two channel shocks are A and B, then Var(A + B) = Var(A) + Var(B) + 2Cov(A,B). Positive covariance increases total uncertainty even before allowing for nonlinear feedback. This is not an estimated covariance for the current boom; it is a reminder that diversification across labels is not diversification across economic drivers. A household might hold technology shares, work for a supplier, own property in a construction-heavy region and receive equity-linked compensation. Each exposure looks different on paper, but all may depend on the same assumption about future computing demand.
Financial amplification begins when market losses interact with contractual obligations. The financial-accelerator literature associated with Bernanke, Gertler and Gilchrist explains how weaker borrower balance sheets can raise external financing costs and amplify an initial shock. Its relevance here is conceptual: asset-price declines can matter beyond an ordinary wealth effect when collateral and credit supply respond. It is not evidence that a particular current borrower is distressed or that all data-center financing is fragile. Establishing actual vulnerability requires information about debt, liquidity, covenants and who bears losses. [The Financial Accelerator in a Quantitative Business Cycle Framework](https://www.nber.org/papers/w6455).
A hypothetical leveraged balance sheet shows the asymmetry. An owner with assets worth 100 and fixed debt of 60 has equity of 40. If asset value falls by 30%, assets become 70 while debt remains 60, leaving equity of 10: a 75% equity loss. Cash-flow shortfalls or collateral requirements can force sales before the investment's long-run value is resolved. An unlevered owner of the same asset instead bears the price loss without an automatic debt-service crisis. This example is not a description of the seven companies. It shows why aggregate market capitalization cannot reveal the resilience of borrowers, intermediaries or project vehicles that sit elsewhere in the financing chain.
Contract structure determines whether a spending reversal is gradual or abrupt. Long-term leases, minimum purchase commitments and take-or-pay energy arrangements can keep cash obligations high after expected demand falls. Those obligations are not all debt payable immediately, and their legal and economic treatment differs. Nevertheless, they can reduce flexibility and shift losses among equipment owners, operators, tenants and lenders. Cancellation rights, parent guarantees and residual asset values become more important than headline commitment totals. A supplier with a diversified customer base may absorb a cancellation; a project dependent on one tenant may not. Detailed mapping is necessary before making a systemic claim in either direction.
Liquidity can also transmit stress without an immediate insolvency problem. Investors facing redemptions may sell their most liquid holdings, spreading price pressure beyond the original sector. Margin calls can force position reductions even when expected long-run cash flows remain intact. Private assets may be marked slowly, delaying recognition while public markets adjust rapidly. Conversely, strong cash reserves, low short-term refinancing needs and patient funding can interrupt these mechanisms. The right question is therefore not simply whether leverage exists, but where it is concentrated, how quickly liabilities can be called, and whether funding remains available when asset values fall.
Policy and supply conditions influence the endpoint. A demand-led correction with easing inflation could permit monetary support, although lower policy rates do not guarantee that damaged borrowers regain market access. A reversal combined with sticky inflation, power constraints or adverse trade shocks could leave less room to cushion demand. Fiscal responses depend on political and budget constraints, while automatic stabilizers work with their own timing. The source's reported 1.5% growth result should therefore be read within its modeled environment. Without full scenario documentation, assuming unlimited policy accommodation would be as unwarranted as assuming that no stabilization occurs.
The disciplined conclusion is neither complacency nor a prediction of crisis. The base simulation suggests that losing a visible growth engine need not automatically produce recession if other demand remains resilient. Its stated omissions mean that the severe tail cannot be assessed from the central path alone. A useful stress test must combine investment weakness, wealth losses, financing conditions and the ability of households and firms to absorb losses. The most dangerous case is not merely fewer new buildings; it is a collapse in expected returns that depresses investment, household consumption and credit availability together. That distinction is what turns a sectoral monitoring exercise into a macro-financial risk assessment.
A monitoring and allocation framework for a conditional boom
A practical framework should preserve the distinction between what has been estimated, what is being assumed and what remains unknown. The Q3 2026 figures describe modeled contributions under a particular counterfactual. The end-2028 figures describe a continuation scenario with investment near 1.6% of GDP and equity gains near early-November levels. Neither set supplies a probability distribution over future outcomes. Assigning portfolio weights directly from the chart would therefore require additional judgments about persistence, valuation, correlations and loss tolerance. The analytical task is to update those judgments with evidence rather than attach false precision to a smooth projected line.
Monitoring the physical investment channel starts with realized activity rather than announcements. Equipment deliveries, construction put in place, energized capacity and grid connections reveal different stages of the same pipeline. A large order backlog may indicate strong demand, but it can also reflect supply bottlenecks or commitments that remain cancellable. Spending growth can weaken because projects are delayed, because equipment prices fall, or because desired capacity is genuinely being reduced. Those interpretations have different implications for domestic employment and future computing supply. Analysts should reconcile expenditure, quantities and commissioning dates before concluding that a change in nominal capital spending represents an equivalent change in real activity.
The domestic-output question then requires attention to import content and displaced activity. A higher domestic construction share can strengthen the immediate GDP impulse, but it may also intensify local labor and power constraints. Falling imported equipment prices can reduce nominal spending while expanding real capacity. Conversely, tariffs can raise the bill without increasing productive services. The reported 75% import assumption is therefore a sensitivity parameter worth tracking, not a shortcut that can permanently translate any future spending number into a GDP estimate. A coherent update would identify what changed in the supply chain and whether the model's domestic-value-added assumptions still describe the actual investment mix.
Monitoring the financial channel requires more than observing a headline equity index. Relative performance of the seven-company basket, changes in household exposure, retirement-account accessibility and the persistence of gains all affect the potential consumption impulse. Broad consumption data should be compared with labor income, saving behavior and spending categories sensitive to affluent households. A rise in travel or luxury spending alongside weak mass-market demand could be consistent with concentrated wealth support, but it would not prove causality. Interest rates, demographics and deferred consumption can produce similar patterns. Evidence becomes stronger when several independent measures tell a consistent story over time.
Labor-market monitoring should likewise look beyond a single unemployment rate. Hours, hiring rates, vacancies, participation and sectoral payrolls can distinguish improving productivity from weakening labor demand. Strong output with stable employment may reflect efficiency gains; falling hours and declining orders may signal an approaching demand slowdown. Construction jobs can weaken even as downstream adopters benefit from cheaper computing. Service employment can remain resilient while asset owners reduce saving rather than current spending. The modeled 0.45-percentage-point unemployment benefit by end-2028 does not prescribe a unique sectoral composition, so the data need to be interpreted in the context of the transmission channel being tested.
Three qualitative scenarios help organize the evidence without pretending to know exact probabilities. In a productive continuation, utilization and paid demand validate new capacity while benefits spread to adopting firms. In an orderly normalization, construction growth slows and valuations become less exuberant, but profitable usage and broader household income prevent a sharp demand break. In an adverse joint reversal, weaker returns cause spending cuts and equity losses while funding conditions tighten. The source's growth result around 1.5% belongs to a specified downside exercise, not automatically to every version of the third scenario. Each scenario requires its own assumptions about leverage, inflation and policy response.
For cross-asset analysis, the shared driver is more important than the instrument label. Equity in infrastructure suppliers, credit to project operators, property near construction hubs and household income linked to technology compensation can all load on the same investment cycle. A diversified portfolio of securities may therefore remain concentrated in one economic narrative. Conversely, businesses that use cheaper computing without bearing the original capital cost may have a different exposure. Even that distinction is not sufficient by itself: downstream firms can lose demand in a broad slowdown. Exposure mapping should include both the direct technology link and the sensitivity to aggregate consumption and financing conditions.
Government bonds illustrate why hedging cannot be reduced to a slogan. A demand shock that lowers inflation can support high-quality duration and make it a useful offset to risky assets. A supply-constrained environment with persistent inflation may weaken that relationship, while a credible productivity acceleration can alter real rates in another direction. Credit combines rate sensitivity with default and liquidity risk, so it is not equivalent to government duration. Cash provides flexibility but carries reinvestment and opportunity costs. The relevant portfolio question is how each asset behaves across the specified states, not whether a familiar hedge worked during a different historical episode.
Risk management should emphasize the ability to survive imperfect timing. A plausible long-run productivity story does not eliminate the possibility of a large interim drawdown, and a correct concern about overinvestment does not establish when prices will adjust. Position sizing, liquidity and diversification of funding sources can matter more than a precise narrative. Stress tests should distinguish mark-to-market losses from cash obligations and examine whether a household or institution would become a forced seller. Those exercises do not require a prediction of imminent crisis. They recognize that investment horizons are constrained by liabilities, governance and liquidity rather than determined solely by confidence in a technology.
The central conclusion returns to the two engines. Data-center spending has supported demand, but the 1.3%-of-GDP surge translates into a smaller direct domestic impulse when roughly three quarters is imported. AI-related equity gains can add substantial consumption support because the household wealth exposure is enormous, even when spending responses are modest. Continued support through 2028 is conditional on investment and valuations holding up, while a joint reversal can remove both channels. The model's nonrecessionary downside is informative but incomplete where leverage and financial stress matter. Durable success ultimately requires productivity and revenue to replace expectations as the foundation of returns. Until that evidence accumulates, the appropriate stance is disciplined conditional analysis: separate accounting from economics, wealth from income, technological usefulness from investor value, and an illustrative scenario from a forecast.



Comments