Shares in US technology companies, led by semiconductor manufacturers, tumbled on Tuesday as a combination of macroeconomic anxiety and mounting scepticism about the scale of artificial intelligence spending weighed on the sector that has powered the market's multi-year rally.

The Philadelphia Semiconductor Index, the benchmark that tracks the largest US chipmakers, dropped 5.6 percent in a single session, while Nvidia, the company whose processors underpin most large-scale AI training and inference workloads, declined 2.2 percent. The losses rippled outward through the wider market: the S&P 500 shed 0.6 percent and the technology-heavy Nasdaq fell 1.6 percent, with the damage concentrated among the companies most closely tied to the data-centre build-out.

The retreat was not confined to equities. In the bond market, the yield on 30-year US Treasuries climbed to its highest level since 2007, as investors fretted about government borrowing requirements linked to the war with Iran. Similar moves played out across the developed world, with 30-year German, French and British yields rising to levels not seen in decades, while Japanese 30-year yields approached their highest levels on record.

Chipmakers lead a broad retreat

Semiconductors have become the most direct and liquid proxy for AI enthusiasm, which makes them the first place investors go when conviction wobbles. The sector has delivered extraordinary returns over the past two years as hyperscale cloud providers, sovereign AI programmes and a wave of well-funded start-ups competed for limited supplies of accelerated computing hardware. That concentration of enthusiasm also means the group carries the heaviest burden of expectation when the broader macro backdrop deteriorates.

Tuesday's decline followed a familiar pattern. Higher long-dated bond yields raise the discount rate applied to future earnings, and companies trading on the promise of profits several years out feel the effect most acutely. A semiconductor business valued at many multiples of its current revenue is, in effect, a long-duration asset, and long-duration assets are precisely what suffer when the 30-year yield pushes to a 17-year high.

Analysts were quick to note that nothing in the fundamentals of chip demand changed overnight. Order books for leading-edge accelerators and high-bandwidth memory remain stretched, and the largest buyers have reiterated capital expenditure plans running into the hundreds of billions of dollars. What shifted was the price investors are willing to pay for that growth, and the willingness of portfolio managers to hold crowded positions into a period of rising financing costs.

Memory and storage names hit hardest

Among the sharpest fallers were the memory and storage specialists that have been among the biggest beneficiaries of the AI boom. Sandisk fell 9.8 percent and Western Digital dropped 7 percent, giving back a portion of the extraordinary gains both have posted over the past year.

Those two companies sit at the centre of a genuine supply squeeze. AI data centres consume vastly more storage than traditional workloads, both for training datasets and for the vast archives of inference logs, checkpoints and synthetic data that model developers accumulate. Demand for high-capacity hard drives and NAND flash has outstripped supply, allowing producers to raise prices sharply after years of punishing cyclical downturns.

The violent swings in these names illustrate a broader feature of the current market: investors are rewarding scarcity but punishing any sign that the cycle may be peaking. Memory has historically been the most volatile corner of the semiconductor industry, prone to boom-and-bust dynamics as capacity additions arrive all at once. With manufacturers now committing to new fabrication capacity, some shareholders are beginning to ask how long the current tightness can last.

Bond yields climb to 2007 highs

The macro backdrop did much of the damage. Thirty-year US Treasury yields rising to their highest since 2007 reflects concern about the trajectory of government borrowing, amplified by the fiscal costs associated with the conflict with Iran. When the long end of the curve sells off, it tightens financial conditions for every borrower in the economy, from homeowners to hyperscalers.

The move was global rather than local. Long-dated German, French and UK government bonds all saw yields climb to multi-decade highs, and Japanese 30-year yields approached record territory as the country slowly normalises a monetary policy that spent years anchored near zero. Synchronised sell-offs at the long end tend to reflect shared worries about debt sustainability, term premia and the willingness of investors to absorb an ever-growing supply of duration.

Barclays analysts pointed to a further, sector-specific driver: technology companies themselves have become major issuers of long-dated debt, adding supply precisely where demand has become more price-sensitive. Corporate treasuries that once funded themselves almost entirely from retained cash flows are now tapping bond markets at scale to build out computing capacity.

The record debt behind the AI build-out

The scale of that borrowing is difficult to overstate. Large technology companies are issuing debt at record levels to finance unprecedented spending on data centres, power generation, cooling infrastructure and networking to support AI services. Investment-grade issuance is expected to reach a record $1.9 trillion this year, compared with $1.44 trillion last year, according to Barclays analysts.

That shift matters for equity holders as well as bond investors. Debt-financed capital expenditure converts what was historically a variable cost into a fixed obligation, and it raises the bar for the returns those assets must generate. If demand for AI services grows more slowly than the infrastructure being built to serve it, the resulting excess capacity would weigh on pricing and on margins across the supply chain.

Several factors are converging to make investors more cautious:

• Rising discount rates: higher long-dated yields reduce the present value of profits expected years into the future, which disproportionately affects high-multiple growth stocks.
• Supply of duration: record corporate and sovereign issuance is competing for the same pool of long-term capital.
• Return on invested capital: investors are beginning to scrutinise whether enormous data-centre budgets will earn adequate returns.
• Customer concentration: a small number of hyperscalers account for a large share of accelerator and memory demand, leaving the supply chain exposed to any single buyer trimming plans.
• Growth expectations: disclosures from leading AI developers are being measured against very high bars set during the rally.

Anthropic's numbers and the growth test

Another factor shaping this week's market moves was the disclosure by Anthropic to investors over the weekend that its annualised run rate had reached $65 billion at the end of July. That represents an exceptionally rapid rate of growth for a company of its age, but it may have fallen short of the expectations of some investors who had modelled even steeper acceleration.

Reuters separately reported that Anthropic is providing guidance for 2028 revenue of between $190 billion and $200 billion. Again, the figure is enormous by the standards of enterprise software history, yet it appears to have disappointed investors hoping for even more stellar growth. In a market where valuations have been built on the assumption of near-exponential adoption, guidance that is merely extraordinary can be read as a warning.

The read-through runs well beyond a single company. AI model developers are among the largest buyers of computing capacity in the world, and their revenue trajectories inform how quickly hyperscalers can justify further data-centre investment. When expectations for those trajectories moderate, the effect travels swiftly down the stack, from accelerators to high-bandwidth memory, to storage, to power equipment and cooling systems.

Signals investors are watching now

Attention in the coming weeks will settle on a handful of measurable indicators. The first is the shape of the US Treasury curve, particularly whether the long end stabilises or continues to climb. The second is the flow of investment-grade corporate issuance from technology borrowers, which will show whether treasurers are willing to keep funding construction at current rates.

The third is forward guidance from the major cloud providers, whose capital expenditure plans effectively set the ceiling for chip demand over the next several quarters. Any suggestion of a pause, a stretch-out of timelines or a shift toward leasing rather than owning capacity would be read as a signal that the infrastructure cycle is maturing.

Finally, investors will watch pricing in memory and storage, where current tightness has driven some of the most dramatic share-price gains of the past year. Evidence of new capacity arriving on schedule, or of inventory building in the channel, would test the thesis that scarcity will persist through the next several quarters.

For now, the sell-off looks less like a verdict on artificial intelligence and more like a repricing of the cost of financing it. The technology continues to be deployed at scale, the orders continue to be placed, and the build-out continues. What has changed is that investors are demanding a wider margin of safety for the debt, the duration and the expectations that the boom now depends on.

Source: Silicon UK News