Central banks hold rates as AI scarcity broadens and diversification moves up capital allocation agenda.
Investment research this week remained broadly constructive on risk assets, though the case for simply following market momentum grew less compelling. The US Federal Reserve and Bank of England both held rates steady, yet each decision drew three dissenting votes for an immediate increase—revealing a hawkish undercurrent beneath the headline holds. Artificial intelligence continued to dominate market performance, but MSCI's analysis showed that returns remained concentrated in a remarkably narrow group of semiconductor and technology-hardware companies. BlackRock, Eastspring Investments and Nuveen consequently emphasised the physical infrastructure behind AI: power, data centres, Asian manufacturing networks and regulated utilities.
The Federal Open Market Committee voted nine to three on July 29 to retain the federal funds target range at 3.5 to 3.75 percent. Beth Hammack, Neel Kashkari and Lorie Logan favoured a 25-basis-point increase instead. The statement described economic activity as solid, with strong productivity growth and capital investment, while acknowledging that inflation remained above target partly because of supply-driven increases in sectors including energy. The unusual division reinforced a key message from pre-meeting research: an unchanged rate should not be mistaken for a return to predictable or conventionally dovish policy.
Before the decision, UBS argued that the Federal Reserve's review of its communications and balance-sheet frameworks could prove more consequential than any single meeting. Less forward guidance may increase short-term rate volatility, while a smaller central-bank balance sheet could eventually require private investors to absorb more long-dated government debt, placing upward pressure on term premia and making duration risk more important. UBS consequently favoured quality short- and medium-maturity bonds, complemented by selective exposure to emerging-market and high-yield debt.
The Bank of England followed on July 30 with another divided hold. The Monetary Policy Committee voted six to three to maintain Bank Rate at 3.75 percent, with three members preferring an increase to 4 percent. UK consumer-price inflation had fallen to 2.6 percent, the labour market was loosening and underlying disinflation remained evident. However, the committee judged that inflation risks were tilted to the upside because of volatile energy prices and the possibility that a prolonged shock could eventually affect wages and broader price-setting behaviour.
Taken together, these decisions leave investors with a more complicated rate environment. Policy rates may remain unchanged, but internal dissent, energy volatility and reduced reliance on forward guidance increase the probability of abrupt repricing along yield curves. The investment implication is not necessarily to abandon duration, but to distinguish more carefully between income earned at the front and middle of the curve and the inflation, fiscal and supply risks embedded in longer maturities.
MSCI's Investment Trends in Focus: Midyear Update 2026, published on July 29, concluded that AI had decisively overtaken geopolitics as the principal force driving markets. The rally survived conflict-related energy disruption and wider geopolitical upheaval because it was supported by genuine earnings expansion rather than simply higher valuation multiples. The less reassuring finding was that the earnings were concentrated in a small number of AI-infrastructure leaders rather than distributed across the broader equity market.
Through July 22, semiconductors and technology hardware together accounted for approximately 76 percent of the MSCI ACWI Investable Market Index's 11.26 percent year-to-date gain. Their combined index weight increased from 18.9 percent to 24.9 percent, while software and services substantially underperformed as investors repriced businesses considered vulnerable to AI-driven disruption. Korea, supported by its memory-chip manufacturers, was the strongest national market in the analysis, while the Netherlands benefited from its exposure to semiconductor-equipment company ASML.
MSCI also found that beta and momentum had driven the equity rally more strongly than factors linked to company fundamentals, growth expectations or profitability. It noted that comparable episodes in which high-beta leaders rose while fundamental factors remained weak have historically been unusual and temporary. This does not establish that the AI rally must reverse, but it places greater pressure on earnings to broaden and on today's market leaders to continue delivering at a pace that justifies their growing index concentration.
The concentration issue extends into private markets. MSCI highlighted the risk that major AI-related listings could improve distributions from venture-capital portfolios while simultaneously increasing overlap between public and private holdings. At the same time, higher real interest rates continue to raise the discount-rate burden on expensive companies, leaving AI-related earnings growth to outrun rather than eliminate valuation risk.
BlackRock Investment Institute maintained its overweight position on the AI theme but argued that investors should focus less on another round of headline earnings beats and more on whether current profit levels are sustainable. The emergence of capable, lower-cost and open-weight models increases competition at the model layer and may weaken the pricing power of frontier-model developers. At the same time, cheaper models could accelerate adoption and create additional demand for the infrastructure required to train, host and deploy them.
BlackRock therefore regarded earnings calls as particularly important. Investors need to assess hyperscaler capital-expenditure commitments, free-cash-flow pressure, debt-financing requirements and evidence that AI spending is producing economic returns. They must also examine how companies outside the technology sector are managing model costs and choosing between proprietary, open-weight and locally controlled systems. BlackRock's conclusion was that cheaper AI may change which businesses capture the economic rent without undermining the wider investment case.
Its preferred exposure remains areas where supply is structurally constrained: power generation and transmission, memory, semiconductors and data-centre infrastructure. These positions do not remove valuation or execution risk, but they rely less heavily on predicting which individual model or application provider will ultimately dominate.
Eastspring Investments approached the AI opportunity through the server and manufacturing ecosystem. It argued that demand should be supported by expanding adoption across industries, rising inference workloads, continuing hyperscaler investment and sovereign programmes intended to establish national AI capabilities. Unlike model training, which occurs periodically, inference requires processing each time an AI application is used, making successful adoption a potential source of recurring infrastructure demand.
Eastspring projected that the AI-infrastructure market could reach USD 758 billion by 2029. Servers accounted for 98 percent of measured AI spending in the second quarter of 2025 and recorded year-on-year growth of 173.2 percent, illustrating how much of the current cycle is anchored in physical compute, memory, networking, power and cooling rather than software alone.
Asia's importance derives from the breadth of its manufacturing network. Taiwan leads in foundry production and advanced packaging; South Korea dominates memory; Japan supplies critical equipment and materials; Southeast Asia supports assembly and deployment; and China contributes components and manufacturing scale. These roles form an integrated ecosystem that has developed over decades and would be difficult for another region to reproduce quickly.
Eastspring nevertheless stressed that exposure to growing volumes does not automatically translate into attractive shareholder returns. Pricing power may migrate as architectures evolve, with potentially stronger economics in memory, advanced foundry and packaging, interconnects, power and cooling where bottlenecks remain binding. The research identified power as the most critical physical constraint and argued that active selection is required to distinguish companies able to protect returns.