The Biggest Software-Chip Split on Record. But the Trade Makes Little Sense

A historic split between software and chips

Monday produced an extraordinary market split. Software outperformed semiconductors by 10.67 percentage points in a single session, the widest gap on record between the two sector ETFs. $iShares Semiconductor ETF (SOXX.US)$ fell 5.6%, while $CrowdStrike (CRWD.US)$ jumped 13.8% and $Palo Alto Networks (PANW.US)$ gained about 13%.

The market's logic was simple: if frontier AI labs slow model development, demand for GPUs and AI infrastructure should weaken. At the same time, more AI safety risks should mean more cybersecurity spending. Investors sold hardware and rushed into software.

That logic has a major problem

Slower frontier model development does not automatically mean lower compute demand. Training is still continuing, while more safety testing, reinforcement learning, evaluations and monitoring can also consume substantial compute. The real warning signs for AI hardware would be Capex cuts, lower GPU orders or delayed data center projects. So far, calls for safer AI development are not the same thing as cuts to AI infrastructure spending.

The bullish cybersecurity argument has an even bigger problem.

Symbols

Yes, more powerful AI should expand the cybersecurity TAM. More autonomous agents, identities, data access and attack surfaces will force companies to spend more on security. But a larger cybersecurity TAM does not automatically mean existing cybersecurity vendors capture all of that value.

First, higher security spending does not prove that today's products can keep pace with frontier models. Palo Alto Networks says its own testing found a step-change in offensive capability, with frontier models able to discover vulnerabilities and chain exploits at a scale that many defenders are not prepared for. Slowing future frontier development would not erase capabilities that already exist.

$Boston Scientific (BSX.US)$ is a useful warning. Industry reporting indicates that CrowdStrike Falcon had already been deployed in its environment before the recent cyber incident, while Boston Scientific has officially confirmed that $CrowdStrike (CRWD.US)$ is involved in the investigation. Yet the attack still disrupted on-premise systems, manufacturing, order processing and shipping badly enough to affect the company's financial outlook. That does not prove Falcon itself technically failed, but it shows why investors should be careful with the assumption that buying a leading EDR platform makes enterprises immune to major AI-era cyber incidents.

Second, part of the expanding AI security TAM may ultimately be captured by the frontier model companies themselves.

This is already starting to happen. Palo Alto Networks is using Anthropic's Claude Security and OpenAI's frontier cyber models inside its defensive products and services. CrowdStrike is also building its AgentWorks ecosystem around frontier models from companies including Anthropic and OpenAI. In other words, even leading cybersecurity companies increasingly rely on the same frontier AI labs for the most advanced reasoning capabilities.

That raises an important valuation question. If OpenAI or Anthropic provides the intelligence layer, while $CrowdStrike (CRWD.US)$ or $Palo Alto Networks (PANW.US)$ mainly provides telemetry, endpoint distribution, security data and workflow orchestration, how much of the incremental AI security economics ultimately belongs to the cybersecurity vendor, and how much goes back to the model provider through model usage, tokens and platform distribution?

AI can make the cybersecurity market much bigger while still making the competitive position of traditional cybersecurity vendors more complicated.

Higher rates are an even bigger problem for software

The macro backdrop makes the rotation harder to justify. The U.S. 10-year Treasury yield moved above 5%, its highest level since October 2023, while markets were heavily pricing another Fed rate hike.

Higher discount rates are especially painful for long-duration software stocks because a larger share of their valuation depends on cash flows many years into the future.

The problem is that software was hardly cheap before Monday's rally.

CrowdStrike is still growing FY27 revenue by roughly 25%, yet its valuation is around 39x trailing sales and more than 140x forward earnings based on current market data. Palo Alto Networks, with roughly low-20% revenue growth, trades at around 24x sales and close to 80x forward earnings. Even after a major correction, $Zscaler (ZS.US)$ still trades at roughly 8x sales.

Symbols

Those are demanding multiples in any environment. They become even harder to justify when the risk-free rate is rising.

Meanwhile, parts of the AI hardware chain are still delivering materially faster revenue and earnings growth, large current cash flows and highly visible infrastructure demand. The market effectively responded to higher rates and AI uncertainty by selling companies with strong near-term earnings visibility and buying some of the longest-duration assets in technology.

That is a difficult fundamental trade to defend.

Summary

Monday looked more like a narrative rotation than a fundamental rotation.

The U.S. market increasingly feels less anchored to slow-moving fundamentals and more driven by positioning, headlines and short-term narratives. That makes price swings much larger than the underlying changes in earnings or long-term demand.

For investors, the lesson is simple: when software and hardware can diverge by more than 10 percentage points in one day, chasing what just went up and panic-selling what just went down can be one of the biggest risks in the market.

$NVIDIA (NVDA.US)$ $Advanced Micro Devices (AMD.US)$ $Broadcom (AVGO.US)$ $Micron Technology (MU.US)$ $SanDisk (SNDK.US)$ $SK hynix (SKHY.US)$ $Lumentum (LITE.US)$ $First Trust Exch Traded Fund 2 Nasdaq Cybersecurity Etf (CIBR.US)$

816K Views