From broad-based growth to scarcity

The starting point is demand.

Bill Demchak, Chairman & CEO of PNC, captured the tone of the quarter: “It’s too broad-based to lay it all on AI… it’s coming from kind of all sectors.”

US Bancorp described growth across almost every commercial category, Fifth Third said confidence was improving broadly, and Honeywell reported broad-based order growth across its short-cycle businesses.

This challenges the idea that the current expansion is solely an AI story. AI remains the most powerful accelerant, but it is increasingly operating on top of a broader industrial recovery.

AI cycle is becoming a physical-infrastructure cycle

At the centre of the investment cycle, demand remains exceptional.

Microsoft CFO Amy Hood said: “Customer demand continues to exceed available capacity.”

NVIDIA now puts a scale on what that demand could imply for investment. CFO Colette Kress expects capex by the five largest hyperscalers to reach nearly USD 800bn in 2026 and USD 1.3tn in 2027.

Jensen Huang also argues that agentic AI could be materially more compute-intensive: “The amount of compute necessary for an agent versus a human using it is probably 15x to 100x.”

The precise multiplier matters less than the mechanism. More capable AI may require more, rather than less, compute, reinforcing the need for additional physical infrastructure.

That investment is spreading well beyond semiconductors. Data centres require electricity; electricity requires generation and grids; grids require equipment and metals; semiconductor fabs require construction, gases, vacuum equipment and specialised machinery.

Chevron CEO Michael Wirth highlighted the bottleneck directly: “Demand far exceeds supply. The grid cannot keep up with the demand from hyperscalers and others, and we see that persisting for years.”

ABB illustrates why the story is broader than AI alone. Data-centre orders grew at a triple-digit rate, yet CFO Christian Nilsson noted: “If we exclude the Data Centers segment, Electrification orders still increased by double digits.”

AI is therefore the accelerant, but the recovery itself is becoming broader.

Supply is becoming the binding constraint

The clearest evidence now comes from Nvidia.

NVIDIA expects the imbalance to persist: “We expect supply to remain a bottleneck at least through the end of fiscal year ’28.”

Jensen Huang put the asymmetry even more clearly: “Even though our demand is much greater than 70%, our supply allows us to confidently deliver 70%.”

This distinction is central to the thesis. Growth is increasingly being determined not by how much demand exists, but by how much capacity can actually be delivered.

The same dynamic is visible elsewhere in the supply chain. Qualcomm says the industry is operating close to full utilisation, while Micron expects DRAM and NAND demand to significantly exceed supply beyond calendar 2027.

Further upstream, Swiss semiconductor equipment vendors VAT and Inficon are already expanding capacity. VAT CEO Urs Gantner described current demand as part of a “long-term technology cycle,” while Inficon CEO Oliver Wyrsch said the company is building capacity to support more than USD 1bn of revenue.

Memory provides perhaps the clearest illustration of scarcity in practice. NVIDIA CFO Colette Kress said: “Tighter memory supply is a symptom of the same demand surge that’s driving our own growth.” She also noted that memory price increases had exceeded NVIDIA’s expectations and were continuing to rise.

The supply shock and the growth impulse therefore have the same origin: exceptionally strong demand.

For example, the price of 8Gb DRAM has risen to around USD 35, roughly ten times its 10-year median and 3.5 times 2018 cycle peak.

Chart 1 — DRAM Scarcity: 8Gb Prices Are Now 10x Their 10-Year Median

Source: Syz Research / Bloomberg

Tim Cook described memory pricing as “a hundred-year flood.” For memory producers, higher prices are a windfall; for downstream customers, they are an input cost. Scarcity travels through the value chain.

Scarcity is constructive — until it hits margins

What we are dealing with is not primarily a traditional negative supply shock. Scarcity is emerging because demand is growing faster than capacity, which is initially constructive for corporate earnings.

High utilisation encourages investment; stronger backlogs improve visibility and scarce inputs gain pricing power. The regime remains favourable as long as companies can absorb higher costs through productivity, pricing and operating leverage.

So far, that is broadly what has happened.

Forward operating margins are at new 15-year highs across the major equity regions and are still accelerating. Over the past three years, margins have improved in 10 of 11 sectors in both the S&P 500 and MSCI Europe, and in 7 of 11 sectors in Japan and Asia ex-Japan.

Table 1 — Global Operating Margins: New Highs Across All Major Regions

Source: Syz Research / Bloomberg

Productivity, pricing and mix have therefore allowed companies to absorb rising costs without sacrificing profitability. However, the first pressure points are emerging precisely where the scarcity framework suggests they should: among companies expanding capacity and among those purchasing scarce inputs.

NVIDIA is one example, with higher memory costs pressuring gross margins before pricing offsets take effect. Apple faces the same memory inflation downstream. One company’s pricing power is another company’s input-cost inflation.

For now, these pressures remain relatively isolated. Global margins are still rising, and the breadth of improvement remains strong. Margins are therefore not yet a warning signal, but they are increasingly the key indicator to monitor.

For now, earnings are winning

The Q2 earnings season has delivered a remarkably consistent message: there is plenty of demand.

Economic momentum is broadening beyond technology, AI investment remains exceptional, and the investment impulse is spreading further into the physical economy. At the same time, capacity constraints are becoming increasingly visible across semiconductors, memory, power and infrastructure.

For now, this remains supportive for aggregate earnings. Forward earnings are still rising faster than equity prices, meaning a greater share of the market advance is being driven by profits rather than multiple expansion.

Chart 2 — An Earnings-Driven Rally: Forward Earnings Are Rising Faster Than Equity Prices

Source: Syz Research / Bloomberg

But scarcity may also cause margin pressure in different parts of the value chain. Three signals will tell us if scarcity is becoming a problem rather than a tailwind.

First, management communication and margin revisions. The earliest warning is likely to be more references to cost pressure, weaker pass-through and customer resistance, followed by broader downward margin revisions.

Second, input prices versus downstream pricing. As long as companies can offset higher memory, power and equipment costs through pricing and productivity, margins should hold.

Third, the pass-through to inflation. If scarcity starts feeding producer and consumer prices more broadly, the risk shifts from individual company margins to rates and equity valuations.

The positioning implication is straightforward: pricing power sits with the owners of the bottlenecks, while downstream companies must continue to demonstrate that productivity and pricing can absorb the pressure.

The first warning is likely to appear in what management teams say before it becomes visible in what they report.