The Companies Building AI Are Starting to Ask Whether It Is Moving Too Fast
For most of the past few years, the artificial intelligence industry has been focused on one question:
How quickly can the technology improve?
Companies have competed to build larger models, stronger AI agents and increasingly powerful computing infrastructure.
But the conversation changed noticeably this week.
Anthropic CEO Dario Amodei called on leading AI developers to slow the rate at which they advance frontier AI capabilities, arguing that increasingly powerful systems could create risks that society may not be prepared to manage.
OpenAI CEO Sam Altman and xAI chief Elon Musk subsequently indicated support for the idea that the industry needs to slow or better control the pace of development.
Investors did not treat the discussion as theoretical.
Asian technology stocks immediately fell.
SoftBank Fell More Than 13%
One of the biggest declines came from Japan.
Shares of SoftBank fell as much as 13.2% during trading.
That matters because SoftBank has become heavily connected to the global AI investment cycle through its exposure to companies and infrastructure associated with artificial intelligence.
Other Japanese semiconductor companies were hit too.
Memory manufacturer Kioxia initially dropped 9.8%, while semiconductor-equipment company Tokyo Electron declined 3.7%.
The selling extended across Asia.
In South Korea:
SK Hynix fell 5.3%.
Samsung Electronics declined 3.7%.
In Taiwan:
TSMC slipped 1.2%.
Chinese semiconductor and AI-related companies also moved lower.
For an industry that has spent years assuming ever-faster AI development, even the possibility of slowing down was enough to make investors reconsider valuations.
Why Would Slower AI Development Hurt Semiconductor Companies?
The connection is relatively straightforward.
The AI boom has driven enormous demand for:
- GPUs
- High-bandwidth memory
- Advanced semiconductor manufacturing
- AI servers
- Networking equipment
- Storage
- Semiconductor manufacturing equipment
- Advanced packaging
- Cloud infrastructure
- Power and cooling systems
Technology companies are investing hundreds of billions of dollars because they expect AI capabilities and demand to continue expanding rapidly.
If frontier AI development slows, even temporarily, investors may question how quickly some of that infrastructure will actually be required.
That doesn’t mean demand suddenly disappears.
But technology stocks are often valued based on future growth expectations, not just today’s revenue.
When expectations are extremely high, even a modest change in future growth can significantly affect valuations.
Saxo Bank strategist Charu Chanana told Reuters that current valuations assume both strong demand and a relentless pace of technological progress, meaning even the possibility of delay can trigger investors to take profits.
Why Are AI Companies Suddenly More Worried?
The concern is not simply that AI models are getting better at answering questions.
The bigger change is AI agents.
Traditional chatbots generally wait for a person to ask a question.
AI agents can increasingly:
- Plan tasks
- Use software tools
- Browse systems
- Write and execute code
- Work through multi-step problems
- Interact with external services
- Coordinate with other AI agents
That makes AI much more useful.
It can also make misuse much more powerful.
Anthropic’s September 2026 threat intelligence report found examples where AI had moved beyond simply assisting attackers and was being used to directly execute or orchestrate parts of cyber operations.
AI Is Changing How Cyberattacks Work
Anthropic said its investigators observed threat actors using AI across different stages of cyber operations, including reconnaissance, tool development, exploitation and data processing.
More importantly, the company said several operations involved multi-agent frameworks executing reconnaissance, exploitation and data exfiltration, with humans mainly selecting targets and reviewing the results.
In one case described by Anthropic, AI agents monitored whether malware was being detected by cybersecurity products.
When detection occurred, the agents could modify and rebuild the malicious tools in an attempt to evade those defences.
This demonstrates why cybersecurity researchers are increasingly concerned about AI agents.
The problem isn’t only that attackers get better answers.
The concern is that parts of an attack that previously required significant human effort can become automated.
Smaller Attackers Could Become Much More Capable
Traditionally, sophisticated cyberattacks required specialised teams.
One person might perform reconnaissance.
Another might build malware.
Another might maintain infrastructure.
Another might analyse stolen information.
AI can potentially reduce that manpower requirement.
Anthropic said one of the most significant trends it observed was that sophisticated attacks no longer necessarily require equally sophisticated attackers.
Its researchers found that AI could reduce the gap between well-resourced state-backed operations and smaller individual operators by giving attackers access to skills and automation that previously required larger teams.
That changes the economics of cybersecurity.
Attackers can potentially do more with fewer people.
Defenders therefore need to automate more as well.
This Doesn’t Mean AI Development Will Stop
The headlines around an AI slowdown can easily create the impression that leading technology companies want to stop developing artificial intelligence.
That is unlikely.
The more realistic discussion is about pacing frontier capability development and strengthening safeguards before releasing increasingly powerful systems.
AI companies still have enormous commercial incentives to innovate.
Governments also see AI leadership as strategically important.
And businesses are already integrating existing AI technology into software, cloud systems, cybersecurity platforms and everyday workflows.
Even if development of the most advanced frontier models became slower, adoption of today’s AI systems could continue growing rapidly.
There is already plenty of capability available that most organisations have not fully implemented.
The AI Industry Could Shift From Training to Using What Already Exists
This could create an interesting shift in the industry.
For the past several years, much of the attention has focused on training bigger AI models.
But businesses ultimately create economic value when they actually use those models.
That means more attention could move towards:
- AI inference
- Enterprise AI applications
- AI agents
- Industry-specific AI
- Cybersecurity AI
- AI PCs
- Edge AI
- Automation
- AI integration with existing business systems
In other words, slower frontier-model development doesn’t necessarily mean slower AI adoption.
It could simply mean businesses spend more time applying the technology that already exists.
This Matters for Malaysia Too
Malaysia may seem far removed from debates between Silicon Valley AI laboratories.
It isn’t.
Malaysia is increasingly connected to the physical infrastructure supporting the global AI industry through:
- Semiconductor manufacturing
- Electrical and electronics production
- Data centres
- Cloud computing
- AI infrastructure
- Networking
- Cybersecurity
If global AI infrastructure investment slows significantly, some parts of that supply chain could feel the impact.
But there is another side to the story.
Malaysia’s AI strategy increasingly emphasises AI adoption and locally developed applications, rather than simply constructing more infrastructure.
That becomes even more important if the industry starts focusing less on endlessly increasing model size and more on extracting practical value from existing AI technology.
For Malaysian businesses, the opportunity is therefore not necessarily building the world’s next frontier foundation model.
It may be building useful services on top of the models already available.
Cybersecurity Could Become One of AI’s Most Important Markets
Ironically, the same risks causing AI companies to call for stronger safeguards could create significant demand for AI cybersecurity.
If attackers increasingly automate reconnaissance and exploitation, defenders will need systems capable of analysing threats at similar speeds.
Security Operations Centres already receive enormous volumes of alerts.
AI could help security teams:
- Analyse suspicious behaviour
- Correlate events across systems
- Investigate alerts
- Prioritise vulnerabilities
- Analyse malware
- Summarise incidents
- Detect unusual activity
- Accelerate response
The cyber arms race could therefore become increasingly AI-versus-AI.
Human security professionals will remain responsible for oversight and decision-making, but automated systems may perform more of the initial analysis.
The Semiconductor Boom Is Not Necessarily Over
Monday’s market decline should also be kept in perspective.
The long-term demand for computing remains enormous.
AI is already being integrated into:
- Cloud platforms
- Smartphones
- PCs
- Enterprise software
- Cars
- Robotics
- Cybersecurity
- Manufacturing
- Healthcare
All of those applications require computing hardware.
Even if the pace of frontier-model development slows, businesses still need chips to run existing models.
AI inference itself can require enormous computing resources when applications serve millions of users.
So the question is not necessarily whether semiconductor demand disappears.
The question is how quickly demand continues growing, and which parts of the semiconductor ecosystem capture the most value.
Investors Are Starting to Ask a Different Question
During the early AI boom, the investment thesis was relatively simple.
AI demand is growing.
Therefore, companies supplying AI infrastructure should grow too.
The calculation is becoming more complicated.
Investors now have to consider:
- AI safety regulation
- Infrastructure costs
- Electricity constraints
- Returns on AI capital expenditure
- Competition between chipmakers
- Custom AI processors
- Model efficiency
- Public opposition to infrastructure expansion
- Whether companies can monetise AI services
Reuters quoted T. Rowe Price portfolio manager Sebastien Mallet making an important distinction: AI can fundamentally change the world without guaranteeing that every investment being made in the sector today will produce attractive returns.
That may become one of the biggest questions surrounding the next phase of the AI boom.
China Does Not Agree With the Slowdown Narrative
The debate also has a geopolitical dimension.
China’s state-backed Global Times criticised Amodei’s argument, describing the proposal as a “Cold War playbook” aimed at restricting China’s technological development.
This shows why coordinating a global AI slowdown would be extremely difficult.
AI is not simply a commercial technology.
Governments increasingly view it as a strategic capability.
If one country slows development while another continues accelerating, governments may worry about falling behind economically or militarily.
AI safety therefore becomes connected to international competition.
That makes agreement much more difficult.
Closing Thoughts
For years, almost every AI headline has been about going faster.
Bigger models.
More GPUs.
Larger data centres.
More investment.
More capabilities.
The reaction across Asian technology markets on 14 September shows that investors are beginning to consider another possibility:
What happens if the AI industry deliberately slows down?
The immediate market reaction may prove temporary.
But the underlying debate will not disappear.
AI systems are becoming more capable, AI agents are increasingly autonomous and real-world misuse is becoming more sophisticated.
The industry now has to balance two objectives that do not always align:
building more powerful AI — and making sure society can safely use it.
For Asia’s semiconductor and technology companies, that debate is no longer abstract.
It is already showing up in their share prices.