The AI trade is no longer moving as one block. On 8 September, shares of major software companies fell as investors digested OpenAI's GPT-6 Astra, while parts of the semiconductor complex benefited from fresh infrastructure deals. The contrast matters because both groups are labelled AI stocks even though the economic exposure can be almost opposite.
Reuters reported declines of roughly 4% to 5% in Salesforce, Intuit and ServiceNow, with the broader software and services group down about 1.4%. Qualcomm rose after detailing its multi-generation AI data-centre collaboration with Amazon. The S&P 500 finished 0.58% lower amid additional pressure from oil, inflation concerns and geopolitics.
Software is being valued on what AI can replace; chips on what AI still has to buy
For application software, increasingly capable models create a pricing question. If an AI agent can complete a workflow across multiple tools, customers may need fewer seats, fewer point solutions or less manual work inside an incumbent application. That does not make established software obsolete, but it raises the burden on vendors to prove that their data, workflow integration and distribution remain valuable when the model layer improves.
For chips, networking and data centres, stronger models can have the opposite near-term effect. More capable systems often require substantial compute to train and serve, while hyperscalers are investing in custom silicon and high-speed interconnects to lower unit costs. The infrastructure suppliers therefore monetize the expansion of AI usage even when the application layer is being disrupted by it.
Our view: stop using 'AI stocks' as a single investment category
Global Markets Review's view is that the useful distinction is between AI toll collectors and AI incumbents under attack. Some companies earn more when AI workloads consume additional silicon, power, networking or cloud capacity. Others must defend an existing software profit pool against models that can automate more of the work their products were paid to organize.
The categories can overlap. Microsoft, Amazon and Alphabet operate infrastructure while also owning application businesses. The market will increasingly price those internal offsets rather than awarding every AI-exposed company the same multiple. The 8 September session is one day, not a permanent regime, but it offers a clean illustration of the split.
| Exposure | Example market reaction | Core question |
|---|---|---|
| Enterprise software | Salesforce, Intuit, ServiceNow fell about 4-5% | Can AI compress seats, pricing or workflow value? |
| Semiconductors | Qualcomm rose after Amazon collaboration | How much incremental compute demand becomes revenue? |
| Cloud platforms | Mixed exposure | Do infrastructure gains outweigh application disruption? |
| Broad market | S&P 500 -0.58% on 8 Sep | AI was one driver alongside oil, inflation and geopolitics |
Frequently asked questions
Why did software stocks fall after GPT-6 Astra?
Investors are reassessing whether more capable AI agents could automate workflows that support software seat counts and pricing. Reuters linked the 8 September declines in several software names to renewed AI disruption concerns.
Why can AI chip stocks rise at the same time?
More AI usage can increase demand for compute, networking and data-centre infrastructure even while the application layer faces greater competition.
Did AI cause the entire S&P 500 decline?
No. Reuters also cited oil prices, inflation concerns and geopolitical risk in the 8 September session.