New Delhi: OpenAI chief executive Sam Altman has sparked debate after questioning whether large amounts of computer memory should remain installed in personal computers when the global supply of DRAM is under pressure from the rapid expansion of artificial intelligence (AI) data centres.

Speaking on the Cheeky Pint podcast, Altman argued that memory installed in individual computers may remain underused for much of the day. He suggested that the world’s limited memory supply could be utilised more efficiently in cloud data centres, where computing resources can be shared across multiple users and workloads.

His remarks come amid a global memory shortage that is pushing up costs for smartphones, laptops and other electronic devices. The growing demand for high-performance memory from AI companies has intensified competition for production capacity, raising concerns about the availability and affordability of consumer electronics.

What did Sam Altman say about PC memory?

During the podcast discussion, Altman was asked about supply constraints affecting the technology industry and the growing interest in running AI models directly on personal computers.

This approach, known as local inference, allows users to run AI models on their own hardware rather than sending requests to remote cloud servers. Open-source AI models have helped increase interest in this method, particularly among developers and users who want greater control over their computing environments.

Altman questioned whether widespread local AI computing was the most efficient response to the current memory shortage.

He said it was “quite terrible” from a global allocation perspective for people to have large amounts of the world’s limited DRAM supply sitting in personal computers while being used only around five per cent of the time.

His argument is based on resource utilisation. A computer may contain 16GB, 32GB or more of RAM, but its memory capacity is not necessarily fully occupied throughout the day. Cloud data centres, on the other hand, can allocate computing resources to different customers and applications as demand changes.

Altman argued that maximising utilisation in cloud infrastructure would make better use of memory during a period of constrained supply.

Why AI data centres need so much memory

AI data centres require substantial quantities of memory to support the computing systems used to train and operate large AI models.

High-bandwidth memory (HBM) is particularly important for advanced AI accelerators, while conventional DRAM is also used across servers and other computing equipment. These memory technologies serve different purposes, but manufacturers face growing demand across the broader market.

Companies building large AI infrastructure projects are competing for supplies and production capacity. OpenAI has also announced partnerships with South Korean memory manufacturers Samsung and SK hynix to support its Stargate infrastructure ambitions.

The scale of investment in AI computing has prompted memory manufacturers to prioritise products and production capacity that serve data centre customers. This has contributed to concerns about supply availability for consumer electronics manufacturers.

The pressure is reflected in market forecasts. India Today cited TrendForce expectations that conventional DRAM contract prices could increase by 10–15 per cent in the fourth quarter of 2026.

Smartphones and laptops could become more expensive

The memory shortage is not limited to companies operating data centres. Consumer electronics manufacturers also depend on DRAM and other memory components to produce computers, smartphones, gaming devices and related products.

When memory prices rise, manufacturers may face higher production costs. They can absorb some of the increase, negotiate supply agreements or pass additional expenses on to buyers.

The pressure could also influence product specifications. Manufacturers may adjust memory configurations, change product pricing or prioritise higher-margin models when component supplies become difficult to secure.

India Today cited an IDC forecast that global smartphone shipments could decline by 16.7 per cent in 2026, while average selling prices could rise by 27.6 per cent to $581. These are industry projections rather than confirmed outcomes for every manufacturer or market.

For Indian buyers, sustained increases in memory costs could make laptops and smartphones more expensive, particularly in segments where manufacturers operate with limited pricing flexibility.

Local AI offers benefits beyond memory utilisation

Altman’s comments have also brought attention to the advantages of running AI models locally.

Local inference can offer users greater control over their data because some processing can take place directly on their devices. It can also reduce dependence on an internet connection and provide access to models without sending every request to an external service.

Developers may prefer local models for experimentation, customisation and testing. Businesses handling sensitive information may also consider on-device or locally hosted AI systems to reduce the need to transfer data to third-party cloud platforms.

However, local AI has its own limitations. Running larger models can require substantial memory, processing power and electricity. Not every personal computer is equipped to handle demanding AI workloads.

The choice between local and cloud computing therefore depends on the application, cost, privacy requirements, performance and available hardware.

A debate over who should use limited memory resources

Altman’s comments have attracted criticism because they came at a time when AI infrastructure expansion is contributing to intense demand for memory components.

Some users have questioned why consumers should be expected to give up local computing capacity when AI companies are among the major buyers competing for memory supplies. Others have raised concerns about increasing dependence on cloud services and recurring subscriptions.

The debate also highlights an important distinction: memory installed in consumer computers cannot simply be removed and transferred to a data centre. Consumer devices and AI servers use different hardware configurations and memory products, even though they are part of the wider semiconductor supply chain.

Altman’s comments are therefore an argument about how future memory production and computing resources should be allocated, rather than a practical proposal to take RAM away from existing PCs.

What it means for computer buyers

For consumers, the immediate concern is the possibility of higher prices and tighter availability of memory components. People planning to purchase laptops or assemble desktop computers may need to pay closer attention to memory capacity and upgrade options.

Buyers should consider their actual requirements rather than assuming that the highest available RAM configuration is necessary for every use case. At the same time, sufficient memory remains important for demanding applications such as video editing, gaming, software development and local AI.

The larger question raised by Altman is whether the future of computing will increasingly depend on centralised cloud infrastructure or continue to include powerful personal devices capable of running AI independently.

As AI companies expand their data centres, that debate is likely to become more important for technology manufacturers, developers and everyday consumers.