March Market Insights: Hardware vs. Software

Written by: Edward Wilhelm

A familiar pattern continued to emerge in markets this February. Companies building the infrastructure of artificial intelligence (AI) continue to post remarkable numbers, while companies building software on top of that infrastructure are navigating a more uncertain moment.

The tension between AI hardware and software companies came into sharper focus last month, and it carries real implications for investors thinking about the AI trade.

NVIDIA Sets the Bar. Again.

NVIDIA reported fourth-quarter results last week that were, by any measure, stellar. Revenue reached $68 billion for the quarter, up 73% from a year ago, and the company guided its next quarter to $78 billion, far ahead of Wall Street's $72 billion estimate.

For the full year, NVIDIA generated nearly $216 billion in revenue.

The hyperscalers are on pace to collectively spend almost $700 billion on AI infrastructure this year, and that demand flows directly to companies supplying the hardware powering artificial intelligence.

Yet the stock dropped roughly 5% the day after earnings. It was a somewhat puzzling moment that reminds us just how high expectations are. The stock price, at any given moment, reflects what the market already believes.

Software's Uncomfortable February

The contrast with the software sector this month has been stark.

The iShares Expanded Tech-Software ETF (IGV), which holds companies like Salesforce (CRM), Palo Alto Networks (PANW), and Intuit (INTU), fell more than 20% in February alone.

The pressure isn’t coming from slowing growth or rising rates. Instead, it is coming from advances in artificial intelligence itself.

Anthropic released two major model upgrades this month. Claude Opus 4.6 added the ability to coordinate multiple AI agents simultaneously across complex professional tasks. Claude Sonnet 4.6 followed, bringing dramatically improved computer-use abilities.

OpenAI continued pushing forward as well.

The market is doing the math: if an AI agent can draft a slide deck, process invoices, manage a workflow, and review code, then the platforms selling those capabilities as separate subscriptions may face genuine pressure. That pressure shows up in earnings guidance and in how investors are pricing the sector.

The Age of Agentic AI: Capability and Caution

The term dominating AI conversations last month was agentic AI.

An AI agent does not just answer a question. It takes action on your behalf, inside your systems, with real consequences. The shift from AI as a tool you interact with to AI as a system that acts independently is one of the most significant developments in artificial intelligence’s short history.

Nothing illustrated the risks of that shift more vividly than a story that went viral last month.

Summer Yue, the Director of AI Alignment at Meta’s Superintelligence Labs, connected an open-source AI agent called OpenClaw to her work email to help organize it.

She gave it clear instructions: do not take any action without her explicit approval.

The agent acknowledged the instruction, then ignored it.

OpenClaw began deleting her entire inbox. She sent commands to stop from her phone, but the agent continued. The only intervention that worked was physically running to her computer and unplugging it.

Her post on the incident drew nearly ten million views, and she later called it a “rookie mistake.”

The irony is hard to miss. The person responsible for keeping AI systems aligned temporarily lost control of her own AI agent.

But the broader implication matters more than the irony.

According to a recent industry report, 81% of organizations have deployed AI agents past the planning phase, while fewer than 15% have full security approval for those deployments.

The technology is moving faster than the governance frameworks built to contain it.

That is not a reason to avoid AI exposure in an investment portfolio, but it is a reason to be thoughtful about which companies are building responsibly versus which ones are simply building fast.

What This Means for Investors

The AI infrastructure buildout remains robust, and NVIDIA’s results confirm it. Hyperscaler capital expenditure plans reinforce it further.

That part of the story remains intact.

This infrastructure story is also personal to how we manage portfolios. Our position in SMH, the VanEck Semiconductor ETF, is grounded in the idea that when the world’s largest companies are competing to spend more on AI each quarter, the companies enabling that computing power tend to benefit in a durable way.

On the software side, the picture is more selective.

Some platforms will successfully integrate artificial intelligence and retain their value to customers. Others may be displaced.

February was a reminder that the most powerful technologies tend to create opportunity and disruption in equal measure.

As always, if you have questions about how these developments may affect your portfolio, please contact an advisor. We are watching these trends closely and are happy to discuss your specific situation.