NextFin News - BlackRock is keeping an overweight on artificial intelligence even as valuation, concentration and financing risks rise, but the firm’s position is more selective than a blanket endorsement of every AI stock. Wei Li, BlackRock’s Global Chief Investment Strategist, is directing attention toward scarce inputs such as power, grids, memory, chips and data centers, where demand is visible even though the eventual winners in the model layer remain uncertain.
The tension is straightforward: the physical AI buildout can remain structurally necessary while the shares financing it go through a cyclical repricing. BlackRock’s 2026 outlooks make that distinction explicit. The firm remains overweight U.S. equities and the AI theme, yet its weekly commentary says the theme requires selective and active positioning and that investors must focus on profit durability, not merely another round of earnings beats. The question is no longer whether companies are spending on AI. It is whether that spending will create durable cash flows before the cost of capacity overwhelms the return.
As of Aug. 4, 2026, the clearest operating evidence still comes from the semiconductor supply chain. NVIDIA reported $81.6 billion of revenue in the first quarter of fiscal 2027, up 85% from a year earlier. Its fiscal 2026 annual review reported $215.9 billion of total revenue, up 65%, and $194 billion of Data Center revenue, up 68%. Those figures confirm powerful demand for compute. They do not, by themselves, prove that every customer, model developer or infrastructure project will earn an adequate return.
That is why BlackRock’s AI stance matters. It is a structural conviction wrapped in cyclical risk controls.
The Bull Case Has Moved From Models to Bottlenecks
BlackRock’s most important change is not simply staying bullish. It is changing the unit of analysis from the prospective model winner to the inputs every model requires.
Its 2026 Midyear Global Outlook says the AI buildout is accelerating and bringing binding constraints forward. It keeps U.S. equities overweight and identifies power, grids, memory, chips and data centers as bottleneck exposures. The broader Global Investment Outlook says investors do not need to know which AI model wins to know that AI requires power, memory, chips and other scarce capacity. This is a more durable argument than assuming that one software platform will capture the entire economic rent.
The reason is transmission. Model competition can reduce prices and compress margins quickly. Electricity interconnections, advanced memory, networking equipment and suitable data-center sites cannot be replicated at the same speed. When supply is constrained, the owners of capacity may capture economic rents even while downstream companies compete aggressively.
NVIDIA’s reported results show the first leg of that chain. The company’s $81.6 billion quarterly revenue and $194 billion of fiscal-year Data Center revenue are supplier-level evidence that buyers are paying for accelerated computing now. The 68% annual growth in Data Center revenue is also a warning: the market no longer needs proof that AI infrastructure has demand; it needs proof that demand can keep expanding fast enough to support the capital committed across the ecosystem.
“We seek broad exposure to the AI buildout through an overweight in U.S. equities,” BlackRock’s 2026 Global Investment Outlook says.
BlackRock’s investment commentary says it expects another $5 trillion to $8 trillion of AI-related capital expenditure through 2030 and identifies the power grid’s ability to support escalating compute demand as critical. The International Energy Agency’s official Energy and AI report likewise treats electricity as a necessary input to data-center AI and models how the technology could affect energy demand, security and affordability.
Those sources do not establish that all related equities will outperform. They establish that AI is no longer only a software-budget story. It is a demand shock for electricity generation, transmission, cooling, construction, semiconductors and financing. That broadening creates the first second-order effect: AI capex increases demand for industrial and utility inputs, which can raise the cost of AI itself.
The bullish case therefore rests on a bottleneck, not on enthusiasm. If demand for compute continues to outrun the delivery of power and capacity, infrastructure suppliers can benefit even if the model layer becomes more competitive.
Why Higher Risks Do Not Yet Break the Structural Thesis
The current episode is structural in infrastructure and cyclical in valuation. That split is essential because it explains how BlackRock can acknowledge higher risks without abandoning the theme.
The cyclical evidence is familiar across three prior investment waves. Telecom companies overbuilt fiber in the late 1990s, leaving excess capacity after the bubble burst even though internet usage continued to grow. Shale producers expanded supply rapidly, and commodity prices later weakened as new output arrived. Cloud infrastructure operators also spent ahead of utilization, forcing competition on capacity and price before demand caught up. In each case, the technology or resource remained important, but capital spending and asset prices still corrected.
The mean-reversion mechanism was the same: new supply arrived, pricing power weakened, and investors lowered the value of future cash flows. AI can follow that pattern if hyperscalers build capacity faster than customers adopt paid applications, if cheaper models reduce revenue available to developers and suppliers, or if financing costs delay projects.
The structural evidence is different. Power generation, transmission corridors, semiconductor fabs and data centers involve long lead times, location constraints and permitting. A fall in a stock price does not create those assets. Even if model prices fall, lower prices can expand usage and increase total inference workloads. That is why cheaper AI is not automatically bearish for the infrastructure chain.
“We remain overweight on the AI theme, but it requires selective and active positioning,” BlackRock’s weekly investment commentary says.
The quote also identifies what has changed. Earlier in the cycle, earnings beats and continued spending could validate the broad theme. Now the market needs evidence that spending produces recurring revenue and acceptable returns on invested capital. The second-order test is whether the buildout creates a durable service economy or merely accelerates a race among a small number of balance sheets.
That distinction separates infrastructure from inventory. Infrastructure produces a long-lived service and may earn contracted or regulated returns. Inventory becomes vulnerable when the next generation arrives before the previous one pays for itself. AI facilities contain elements of both, which is why the theme can remain strategically favored and tactically fragile.
The Market’s Unpriced Problem Is Profit Capture
The conventional question is whether AI is a bubble. The more useful question is where the economic rent settles when the spending wave matures.
If models become cheaper and more capable, users may adopt them faster, lifting total demand for compute. That favors scarce inputs and can support a wider group of suppliers. It also threatens the pricing power of model developers. A lower price per query can be bullish for utilization and bearish for revenue per query; the result depends on whether volume grows faster than price falls.
This is the expectation gap in BlackRock’s positioning. The market can be right about AI’s economic importance and wrong about the identity of its winners. A broad U.S. equity overweight captures the probability that some winners will emerge in the country’s technology and capital markets. A bottleneck focus narrows exposure to inputs for which demand is visible regardless of which model wins.
The cross-asset transmission is just as important. More AI capex increases demand for electricity, equipment and construction. If supply cannot respond, power prices and inflation-sensitive income can rise. Higher nominal growth can support some equities while keeping long-term bond yields elevated. That weakens the diversification benefit of long-duration government debt and raises the discount rate applied to long-duration technology cash flows.
AI can therefore be both a growth engine and a source of higher term premia. The same investment boom that supports earnings can make the financing environment less forgiving. A high-growth company with distant cash flows is exposed to both the success of AI and the interest-rate consequences of funding its physical expansion.
BlackRock’s preference for shorter-duration income alongside its AI and infrastructure view fits that mechanism. The firm is treating the buildout as a source of opportunity while recognizing that higher yields can persist if capital intensity, fiscal spending and supply constraints keep inflation above the pre-pandemic pattern.
The implication for equities is dispersion. Hardware and infrastructure companies can benefit from orders but still suffer if customers delay projects. Utilities can gain from load growth but face political pressure over rates and grid costs. Semiconductor companies can report rapid revenue growth while margins attract new competitors and export controls reshape their addressable markets. Software companies can gain productivity while facing disruption if customers substitute AI tools for established products.
Sector labels can also conceal concentration by economic driver. A portfolio holding a chipmaker, a data-center operator, a utility and an industrial supplier may look diversified while remaining exposed to the same capex cycle, power bottleneck and financing condition. The relevant question is who pays, who funds and who bears the risk if utilization disappoints.
The Strongest Counter-Thesis Is a Capital-Expenditure Bust
The strongest case against BlackRock’s position is not that AI has no value. It is that the industry is building too much capacity ahead of verified demand, and that the adjustment will hit the very bottlenecks being treated as scarce.
A capital-expenditure bust would begin with customer economics. Hyperscalers could slow orders if AI services fail to produce returns that justify the cost of chips, data centers and electricity. Model providers could cut prices to compete, forcing investors to lower revenue assumptions. Credit markets could then demand more compensation from data-center developers, creating a feedback loop: weaker utilization reduces financing capacity, and weaker financing delays the capacity needed for future demand.
That counter-thesis attacks the structural call at its foundation. Scarcity is valuable only when customers can pay for access. If productivity gains arrive slowly, power and data-center constraints may become stranded investments rather than sources of pricing power. The historical analog is not only telecom overbuilding; it is any infrastructure race where capacity arrives before the cash flows meant to support it.
The answer is that current evidence still shows real supplier revenue, not only promotional spending. NVIDIA’s $194 billion of fiscal 2026 Data Center revenue and $81.6 billion quarterly revenue are operating-demand signals, although they do not settle the question of end-user profitability. BlackRock’s shift toward scarce inputs reduces dependence on identifying a single model winner. It does not eliminate the risk that a broad capex slowdown pulls suppliers down together.
The falsifying signal is specific: if NVIDIA’s Data Center revenue growth falls below 30% year over year for two consecutive quarters while the largest cloud operators reduce AI-related capital-expenditure guidance by at least 10% from their prior plans, the claim that AI scarcity remains the dominant investment mechanism would be wrong. That combination would show that demand had stopped outrunning supply and that the bottleneck thesis had become a capacity thesis.
A second warning signal is macroeconomic. If long-term Treasury yields rise materially while AI-related capital spending slows, markets would be showing that financing pressure is overwhelming the growth impulse. BlackRock’s view depends on the buildout generating enough economic value to absorb its funding cost. If funding costs rise and the revenue engine weakens, cyclical and structural risks would point in the same direction.
BlackRock is not saying higher risks are irrelevant. It is saying risk should change how the theme is expressed. That is narrower and more testable than the claim that AI stocks can rise regardless of valuation.
What the Position Means Across Time Horizons
In the short term, sentiment and liquidity can dominate fundamentals. A disappointing earnings guide, a delay in a major data-center project or a rise in real yields can trigger a rotation out of high-duration AI equities even if long-run electricity and compute trends remain intact. The likely response would be dispersion rather than a clean exit: crowded beneficiaries would carry the largest multiple risk, while companies tied to contracted infrastructure or essential components could prove more resilient.
Over the medium term, the decisive variable is profit durability. Investors will need to distinguish capex that creates recurring customer revenue from capex that merely shifts market share. NVIDIA’s quarterly growth establishes the strength of current supplier demand, but future reports must show whether inference, enterprise adoption and sovereign computing broaden the revenue base beyond a small number of buyers.
Over the long term, the infrastructure thesis is structural if electricity supply expands, grids connect new load in time and AI usage grows faster than the cost of serving it. Those conditions would support BlackRock’s emphasis on scarcity and move the opportunity set beyond the model layer into power generation, transmission, memory, networking, cooling and construction.
The base case is a selective continuation of the AI buildout, with rising dispersion and periodic valuation resets. Its trigger is continued growth in data-center demand and sustained capital expenditure by major cloud operators, alongside evidence that AI services generate revenue outside the largest technology companies.
The upside case is a productivity-led broadening. Cheaper and more capable models increase usage while power and grid investment unlock capacity. Market leadership would then broaden from a narrow group of mega-cap technology companies to infrastructure, industrial and selected international suppliers.
The downside case is a capex digestion cycle. Customers slow orders, model prices fall faster than workloads rise and higher yields expose leverage embedded in new infrastructure. The trigger is the falsifying combination of sub-30% annual Data Center growth at NVIDIA for two quarters and a 10% or greater reduction in cloud AI capex guidance.
The next reports to watch are not only model launches or headline spending totals. Utilization rates, contracted power, customer concentration, free cash flow after capital expenditure and the spread between AI-service price declines and workload growth will test whether scarcity is producing economic rent or simply imposing cost.
BlackRock’s call is best understood as structural conviction wrapped in cyclical risk controls. It remains overweight because the physical buildout is still under-supplied, but it is moving away from the assumption that the most famous AI companies automatically capture the value created.
The AI trade is no longer a single bet on intelligence; it is a contest over who can finance and power the infrastructure before the returns arrive.
Explore more exclusive insights at nextfin.ai.
