Software stocks are in oversold territory....here's a round-up of tech's biggest headlines
Arm's transformation and Qualcomm's memory problem. Read on and catch up
Software stocks fell 15% in a single week, pushing the sector 29% below its September highs and into territory that, by at least one measure, has never been this extreme. The proportion of software stocks in oversold territory eclipsed 70%, and the iShares Expanded Tech Software ETF relative to the S&P 500 hit its most oversold reading on record.
The catalyst was Alphabet. Google's parent company announced a fiscal year capital expenditure forecast of $185 billion, roughly $65 billion more than analysts had expected. The stock dropped nearly 5%, dragging the broader market with it. The S&P 500 fell 1.2%, the Dow shed 1.2%, and the Nasdaq slid 1.4% to its lowest point since November.
This was not a contained correction. It was a market-wide repricing of what the AI buildout actually costs.
Alphabet's cloud is growing at 48%. The market doesn't care
The irony of Alphabet's selloff is that the company's underlying business is performing well by almost any conventional measure. Google Cloud grew 48% year over year, comfortably ahead of Microsoft Azure's 39% growth. Under normal circumstances, that number would have sent the stock higher.
But $185 billion in planned spending overwhelmed the growth story. Investors are recalibrating what they are willing to pay for a company that is ploughing cash into infrastructure at this pace, regardless of how strong the revenue trajectory looks. The momentum heading into earnings had been strong, which made the reaction sharper: this was a "take profit" moment dressed up as a capex panic.
The concern is not unique to Alphabet. Amazon, expected to report shortly after, saw its stock drop 4% ahead of earnings on fears that its own capex guidance would follow a similar pattern. Wall Street estimates put Amazon's 2026 spending at around $125 billion, a 22% increase from 2025. The $500 billion Stargate initiative announced in January, once dismissed as aspirational, is starting to look credible, and that is making investors nervous rather than excited.
ARM's transformation into an AI infrastructure company
While the market punished the companies writing the cheques, one beneficiary of the spending surge stood out. ARM, the chip architecture company majority-owned by SoftBank, reported that its data centre business is growing at 100% year over year.
CEO Rene Haas described a clear inflexion point approaching: the moment when ARM's revenue base shifts from being primarily handset-driven to data centre-driven. That crossover is arriving sooner than expected. More than 50% of hyperscale data centre operators now use ARM-based CPUs, and the rollout of NVIDIA's Vera Rubin platform, which uses 88 ARM-based processors, is accelerating the shift.
Jensen Huang's announcement that NVIDIA will sell its ARM-based CPU as a standalone product was, in Haas's framing, a trajectory-changing moment. It opens a new market for ARM royalties beyond the integrated systems where its chips have traditionally lived.
The memory shortage that is squeezing other chipmakers has had minimal direct impact on ARM, thanks to the breadth of its licensing model. ARM collects royalties across handsets, data centres, automotive and IoT. It does not manufacture chips or depend on any single supply chain. That diversification, long seen as a weakness compared to more focused competitors, is now functioning as a shield.
Qualcomm's memory problem is a demand problem in disguise
Qualcomm told a different story. The company forecast $11 billion in sales for the current quarter, but CEO Cristiano Amon was blunt: the number could have been higher if memory supply were not constrained.
The memory shortage is reshaping how the entire handset industry plans its production. Original equipment manufacturers are building their forecasts around memory availability rather than consumer demand, an unusual inversion that is compressing volumes at the lower end of the market. Amon argued that the premium and high-tier segments remain resilient, both because consumers are willing to absorb price increases and because OEMs are prioritising their best-margin products when allocating scarce components.
Qualcomm's diversification push is gaining traction. The company posted a record quarter in automotive, with revenue hitting $1 billion for the second consecutive quarter, anchored by a broad partnership with Volkswagen Group. Robotics and connected devices are growing. And the data centre business, built around an inference-focused architecture that does not require high-bandwidth memory, is expected to start contributing meaningful revenue in fiscal 2027.
But the stock fell 8%. In this market, a story about future growth does not offset a constrained present.
$18 billion in software loans just became distressed
The selloff in public equities is only part of the picture. In the leveraged loan market, nearly $18 billion in software debt has been pushed into distressed territory over the past four weeks, the highest level since October 2022.
The question at the centre of this dislocation is existential for parts of the software industry: which companies will survive the arrival of AI-native competitors, and which will see their revenue bases eroded? A new platform called Frontier, designed to help enterprises deploy AI agents with guardrails and data access controls, represents exactly the kind of tool that could chip away at incumbent software providers.
Some loan investors are pushing back against the panic. Aries' CEO has publicly defended the safety of software loans, arguing the market is indiscriminately punishing companies with strong fundamentals alongside genuinely vulnerable ones. There are software companies generating billions in revenue with accelerating earnings, not declining ones.
But the mechanics of the loan market could amplify the damage. Collateralised loan obligations, which are significant buyers of leveraged loans, have limits on how much distressed debt they can hold. If the selloff pushes more loans below those thresholds, CLOs could become forced sellers, creating a cascading effect that has little to do with the underlying health of the businesses.
Established players like West Law and Lexis-Nexis are not sitting still. Both have been aggressive adopters of AI within their own products, and their decades of proprietary data give them a moat that generic AI tools cannot easily replicate. For new entrants to compete, they would need to demonstrate that their systems no longer hallucinate and can reliably flag where human review is needed. That bar has not yet been cleared.
Crypto's parallel reckoning
The risk-off mood extended well beyond tech stocks. Bitcoin fell below $68,000, its lowest since November 2024 and roughly 45% below its October high. Ethereum and Solana, tokens that were supposed to underpin a new internet-native financial system, are struggling to maintain relevance.
The timing is notable. Gold and silver are rallying. If Bitcoin's primary value proposition is as digital gold, its performance relative to actual gold is becoming difficult to explain. Crypto exchange Gemini announced plans to cut about 25% of its workforce, a signal that the industry's own operators are losing confidence in near-term growth.
The market is asking a simple question with no simple answer
Across every segment of this selloff, the same tension is playing out. AI is clearly working. Cloud revenue is surging, productivity tools are improving, and the infrastructure being built will almost certainly be used. But the gap between "AI works" and "this level of spending is justified" remains wide, and investors are no longer willing to fund the difference on faith alone.
The companies at the centre of this cycle, Alphabet, Amazon, Meta, and Microsoft, are not startups burning through venture capital. They are among the most profitable businesses ever built. That gives them staying power. Whether it gives them the right to spend $185 billion in a single year without the market flinching is a different question entirely.