The AI trade is splitting — is money moving from chips into software?

The AI trade is starting to divide. Semiconductor stocks have dominated the market for much of 2026 as spending on AI accelerators, memory and data centres surged. The Philadelphia Semiconductor Index is still up roughly 58% this year, reflecting just how powerful that rally has been. But momentum has become less consistent across some of its biggest names. 

At the same time, software stocks that spent much of the year under pressure from fears that AI could replace traditional applications have staged a significant recovery. The S&P 500 software sub-index has recently climbed back towards where it began 2026, with companies including Salesforce, ServiceNow and Workday recovering sharply from their mid-year lows. 

Now a new debate over whether the world is developing AI too quickly has given that rotation another catalyst.

Anthropic chief executive Dario Amodei has called for the industry to slow the development of advanced AI models, with OpenAI's Sam Altman, Google DeepMind's Demis Hassabis and Elon Musk among the industry leaders backing greater restraint. 

That raises a question that could matter well beyond one volatile trading session: does a slower AI development cycle hurt the companies selling the computing power while helping the software businesses investors once feared AI would replace?

For most of the AI boom, the easiest part of the investment thesis was infrastructure. More powerful models required more GPUs. More GPUs required more data centres, networking equipment, high-bandwidth memory and power. Hyperscalers including Microsoft, Alphabet, Amazon and Meta responded by committing hundreds of billions of dollars to AI infrastructure.

That spending created a direct earnings tailwind for Nvidia, Broadcom, AMD, Micron and other semiconductor suppliers.

The latest safety debate challenges the assumption that the fastest possible development path will continue indefinitely. For chipmakers, the concern is straightforward. If AI laboratories deliberately stretch development cycles, require more safety testing or delay new models, some of the urgency behind infrastructure spending could ease. That means investors may begin demanding more evidence that enormous capital expenditure programmes can continue generating equally enormous returns.

The semiconductor trade remains strong over longer periods, but leadership has become less uniform. Broadcom and Nvidia have both lost ground from recent highs, while AMD and Micron have held up better.

That matters because the market is no longer rewarding every AI hardware name equally. The next phase may depend less on exposure to AI itself and more on which companies can keep converting huge infrastructure spending into earnings growth.

For Australian traders, that divergence creates opportunities on both sides of the global technology market. CFDs can provide exposure to individual US technology shares and major indices without requiring the entire AI sector to move in the same direction.

Open a Trading Account

      “Trade Tech Stocks with an ASIC-regulated broker. Fast AUD funding via PayID. ”  

Software stocks entered 2026 under pressure from fears that AI agents could replace parts of traditional enterprise software and reduce demand for paid applications.

That view has started to soften. Salesforce, ServiceNow and Workday have all rebounded from mid-year lows as investors reassess whether established software companies can use AI to strengthen their products rather than be displaced by it.

Recent earnings have helped. Salesforce has reported more than US$1.5 billion in recurring revenue from AI-powered products, while major software vendors continue embedding AI into platforms that already serve large corporate customer bases.

The result is a more selective market: investors are increasingly rewarding software companies that can monetise AI, rather than treating the whole sector as vulnerable to it.

The hardware boom benefits from faster development. Each new generation of models can require more computing power, larger training clusters, and more advanced memory and networking.

For traditional software, faster development can create greater disruption.

If AI agents become capable of independently designing workflows, writing code, managing customer relationships and completing administrative tasks, some existing software products risk losing value.

A slower frontier-model development cycle potentially changes that equation.

Established software companies would gain more time to integrate AI into products customers already use, defend their distribution advantages and shift pricing models towards AI-enabled services.

That could turn AI from a replacement threat into another feature that companies such as ServiceNow, Salesforce and Workday sell to existing corporate customers.

It also explains why the software recovery began before the latest calls to slow AI development. Investors have already started questioning whether the disruption narrative became too pessimistic.

Cybersecurity adds another layer to the rotation. AI safety warnings have increasingly centred on the ability of advanced models to conduct cyberattacks, automate malicious activity or exploit computer systems with limited human involvement.

That makes cybersecurity spending potentially more important even if frontier-model development slows.

Companies including CrowdStrike, Palo Alto Networks and Zscaler sit in an unusual position: they can benefit from companies adopting AI, but may also benefit from concerns over the security risks the technology creates.

That helps explain why cybersecurity has recently been one of the strongest parts of the recovering software market.

The result is another potential split inside technology. Investors may become less focused on simply finding companies labelled as “AI beneficiaries” and more focused on which part of the AI spending chain continues to capture revenue.

One factor could still overwhelm the rotation: spending by the hyperscalers.

Microsoft, Amazon, Alphabet and Meta remain committed to building massive AI infrastructure networks. Unless those companies begin reducing or delaying capital expenditure, demand for GPUs, networking chips and memory could remain exceptionally strong.

That means calls for slower AI development should not automatically be interpreted as calls for dramatically lower semiconductor demand.

Safety testing itself requires computing resources. AI inference continues expanding. Companies are still deploying existing models across businesses and consumer products. Competition between the US and China also makes a coordinated global slowdown difficult.

The semiconductor trade therefore faces a different question than it did earlier in the boom.

Instead of asking whether AI demand exists, markets may increasingly ask how much additional spending is already priced into chip valuations and how quickly that spending needs to keep growing.

Three developments could determine whether the split between chips and software lasts.

• AI investment guidance: Any change to capital expenditure plans from Microsoft, Alphabet, Amazon or Meta would provide a clearer signal for semiconductor demand than short-term share-price volatility.

• AI regulation and safety rules: Concrete requirements for model testing, independent evaluation or development limits would give the current debate financial consequences. Without policy changes, the slowdown discussion may remain largely voluntary.

• Software earnings: ServiceNow, Salesforce, Adobe and Workday still need to show that AI is increasing revenue rather than simply increasing development costs. Evidence that customers are paying for AI features would strengthen the sector's recovery.

Together, these catalysts could make the next phase of the AI trade considerably more selective than the last.

The widening gap between different parts of the technology sector means traders do not necessarily need a single bullish or bearish view on AI.

Through Mitrade, eligible Australian clients can use CFDs to take either long or short positions on major US technology shares and indices, allowing different views on semiconductor, software and broader Nasdaq exposure to be expressed separately.

Traders can also use stop-loss and take-profit orders to manage predefined levels, while pending orders can be used when waiting for a particular price before opening a position.

Accounts can be funded in AUD, helping Australian traders avoid the need to manually convert funds before each US-market trade. A demo account is also available for testing strategies without committing real capital.

Mitrade is regulated in Australia by ASIC. CFDs are leveraged products, meaning both potential gains and losses can be magnified, so risk controls remain particularly important when trading volatile technology stocks.

The AI boom may be entering a more selective phase

The latest AI safety debate does not signal the end of the artificial-intelligence investment story.

It may, however, challenge the idea that every part of the trade benefits from the same developments.

Semiconductor companies remain backed by enormous spending on AI infrastructure, but the strongest stocks have already priced in years of rapid growth. Any suggestion that frontier-model development could slow therefore attracts greater scrutiny.

Software companies face the opposite setup. Many were heavily discounted because investors feared AI would destroy their business models. The recovery in ServiceNow, Workday and Salesforce suggests that markets are starting to reconsider that assumption.

If AI development becomes more controlled while enterprise adoption continues, the next winners may be companies that apply AI profitably rather than simply supply the hardware used to build it.

Start trading global technology shares in three simple steps