The OpenAI CEO’s controversial argument exposes a much bigger conflict: AI companies need enormous amounts of memory, consumers are paying more for everyday devices, and the future of local computing could be at stake.
The next time you buy a laptop with 32GB of RAM, consider an uncomfortable question: Does the world really need that memory sitting inside your computer, or would it be more useful inside an AI data centre serving millions of people?
That is essentially the question OpenAI CEO Sam Altman has brought into the technology industry’s increasingly heated debate over memory shortages.
Speaking with Stripe co-founder John Collison on the Cheeky Pint podcast, released on October 29, Altman argued that people keeping large amounts of DRAM in personal computers represents an inefficient allocation of a resource that the world urgently needs.
His reasoning was straightforward. Most computers do not use all their installed memory continuously. Meanwhile, AI data centres can allocate memory across thousands of workloads and keep expensive infrastructure operating at much higher utilization.
“From a global allocation perspective, actually, it’s quite terrible to have everybody have a bunch of the world’s limited supply of DRAM sitting in their own computer,” Altman said.
He added that the memory might be used just 5% of the time and argued that maximum utilization in cloud infrastructure would be preferable while supply remains constrained.
The comments quickly attracted criticism from people who support running artificial intelligence locally, particularly the open-source AI community.
But the controversy deserves a much deeper examination than whether Altman is right or wrong about laptop memory.
His comments expose a fundamental disagreement about the future of computing: Should powerful computing resources be concentrated in a relatively small number of corporate data centres, or should individuals retain the hardware and freedom to run increasingly capable AI systems on their own devices?
And there is an uncomfortable economic question underneath it all: If AI companies are consuming enormous quantities of the world’s memory supply, should ordinary consumers be expected to accept higher prices and fewer choices?
Table of Contents
1. What Sam Altman actually said — and why it matters
Altman’s argument emerged during a broader discussion about the global shortage of computing infrastructure.
Collison challenged the assumption that open-source AI models and local inference would eventually reduce the need for enormous data centres.
Altman disagreed.
His argument was that allowing millions of individuals to maintain large amounts of DRAM in computers that remain underutilized would not necessarily solve the global computing shortage.
A cloud data centre, by comparison, can serve many customers, allocate resources dynamically and maintain much higher utilization across its infrastructure.
Imagine two scenarios.
In the first, 1,000 people each own computers containing 64GB of RAM. Much of that memory remains unused while they browse websites, write documents or leave their computers idle.
In the second, a cloud operator manages a large pool of memory and allocates capacity to different customers as their workloads require it.
From a narrow utilization perspective, the second arrangement can be more efficient.
But there is an important distinction: higher infrastructure utilization does not automatically mean a better computing system for society.
The cloud operator controls the infrastructure. Users depend on its pricing, availability, policies and technical decisions.
A personally owned computer gives its owner direct control over the hardware, even when that hardware is not operating at maximum utilization.
This is where an engineering argument becomes an economic and political debate.
2. The memory shortage is real — but the problem is bigger than people buying too much RAM
To understand why Altman’s comments matter, it is necessary to understand what is happening inside the semiconductor industry.
Artificial intelligence requires enormous amounts of memory.
Modern AI accelerators need rapid access to massive quantities of data while training and running sophisticated models. This has increased demand for high-bandwidth memory, commonly known as HBM.
HBM is a specialized form of DRAM designed to deliver extremely high memory bandwidth. It is particularly important for advanced AI accelerators.
However, HBM and the DDR memory installed in ordinary laptops are not interchangeable products.
They use different architectures, packaging approaches and manufacturing requirements.
The connection is the semiconductor manufacturing ecosystem.
Manufacturers have limited production capacity, capital budgets, engineering resources and packaging capabilities. As demand for high-margin AI memory increases, manufacturers have stronger incentives to prioritize server and AI-related products over lower-margin consumer applications.
This creates pressure throughout the memory supply chain.
According to TrendForce’s September 30, 2026 forecast:
- Conventional DRAM contract prices are expected to increase 10–15% quarter over quarter in the fourth quarter of 2026.
- NAND Flash contract prices are projected to rise 15–20%.
- Server memory demand remains strong.
- PC memory supply could become tighter in 2027 as manufacturers shift more capacity toward server applications.
- Mobile memory availability also remains constrained.
These are forecasts for contract-market prices, not guaranteed increases in every retail product. Nevertheless, they demonstrate that the pressure is not simply a social-media exaggeration.
The problem is particularly important because memory is an essential component in almost every modern electronic device.
It is found in smartphones, laptops, desktops, televisions, gaming systems, networking equipment and industrial electronics.
When manufacturers prioritize one category of customer, the effects can eventually reach many others.
AI is not just competing for GPUs anymore. It is competing for the broader hardware resources that make modern computing possible.
3. The hidden AI tax: How ordinary consumers are already paying for the AI boom
Perhaps the most important part of this story is happening outside the data centre.
Consumers do not directly purchase HBM chips for AI accelerators. But they can still experience the consequences of a market in which memory manufacturers increasingly prioritize AI infrastructure.
India offers a particularly revealing example.
An Indian Express investigation published in June 2026 reported that memory costs had become a dramatically larger component of budget smartphone manufacturing.
According to figures cited from Counterpoint Research, memory could account for approximately 43% of the bill of materials of smartphones priced below Rs 20,000.
This is a significant shift for a market where affordability determines whether millions of consumers upgrade their devices.
The consequences extend beyond smartphones.
Indian manufacturers and industry analysts have reported pressure on laptop prices, smartphone production and entry-level device economics.
TechCrunch’s July 17 report on India’s smartphone market also described how memory shortages were pushing brands to increase prices, reduce their exposure to low-margin models and encourage consumers to consider refurbished devices.
The situation is not limited to India. It is a global supply-chain issue.
For consumers, the consequences can include:
- Paying more for a laptop with the same memory capacity.
- Getting less RAM at a similar price.
- Delaying a smartphone upgrade.
- Choosing a lower-storage or lower-memory configuration.
- Keeping an older computer for longer.
- Finding that budget devices become less attractive compared with refurbished alternatives.
This creates an interesting contradiction.
AI is being marketed as a technology that will make people more productive and improve access to knowledge. Yet the infrastructure supporting that technology is putting additional cost pressure on the very devices people use to participate in the digital economy.
In effect, consumers may be paying an indirect AI tax without ever subscribing to an AI service.
4. The Derek Thompson connection: The AI boom is becoming an infrastructure and political problem
This is where the thinking of writer Derek Thompson becomes particularly relevant.
In his August 2026 essay, The Great American AI Rebellion, Thompson examined a contradiction in the American economy.
AI infrastructure investment has become an important driver of economic growth, but the construction of data centres is increasingly facing public resistance.
His central approach was to put several facts next to one another.
First, the AI infrastructure buildout has become extraordinarily important to economic growth.
Second, the industry needs more data centres to support rising demand for AI services.
Third, communities across the United States are increasingly resisting new data-centre developments.
Thompson explored reporting by Jasmine Sun, which found that opposition was not necessarily driven by hostility toward artificial intelligence itself.
People were also concerned about local electricity and infrastructure demands, corporate transparency, secretive agreements and whether communities were receiving meaningful benefits from developments taking place in their neighbourhoods.
This framework offers a useful way to understand the Altman controversy.
The conflict is no longer confined to the technology companies building AI systems.
It is spreading into the everyday economy.
Data centres need electricity, land, construction materials, cooling systems, networking equipment, processors and memory.
Every additional layer of demand creates potential competition with other users of those resources.
Altman’s comments add another dimension: even memory installed in privately owned computers is now being discussed as a resource that might be better allocated to centralized infrastructure.
Thompson’s broader argument about public resistance becomes relevant because people are increasingly being asked to accommodate an infrastructure expansion whose benefits and costs are not distributed equally.
A technology company may benefit from higher data-centre utilization. A consumer may face a more expensive laptop. A local community may experience additional infrastructure pressure. A small business may struggle to compete for computing capacity.
These are different sides of the same economic transformation.
The second warning: What if the AI infrastructure investment does not pay off?
Thompson’s October 23 discussion, The AI Story Wall Street Doesn’t Want to Hear, raises another important question: Is there enough actual demand for AI services to justify the enormous spending on chips and data centres?
The supply side of the AI boom is easy to see.
Companies are investing in GPUs, memory, networking, buildings and power infrastructure.
The demand side is more complicated.
Someone ultimately has to pay for all that computing capacity.
If businesses and consumers increasingly pay for AI services because those services deliver measurable productivity improvements, the infrastructure investment could prove economically justified.
But if companies build capacity faster than profitable demand develops, the industry could face overcapacity, falling prices and substantial financial losses.
This distinction matters enormously for Altman’s argument.
He is effectively advocating maximum utilization of scarce memory today while also describing a future in which computing becomes dramatically cheaper because supply expands.
In the podcast, he suggested that the five-year total cost of owning and operating a gigawatt of AI infrastructure could fall from roughly $70 billion to $20 billion.
That is an ambitious vision, not an established forecast.
The problem is that building infrastructure takes time, while demand and technology can change quickly.
If the industry overestimates future demand, the same infrastructure that looks essential today could become an expensive financial burden tomorrow.
If demand continues accelerating, the shortage could last much longer.
Either outcome could reshape the economics of consumer and enterprise computing.
5. Why local AI users are particularly angry
Altman’s remarks have touched a nerve because local AI is about much more than saving money on cloud subscriptions.
For a growing community of developers, researchers and enthusiasts, running AI models locally represents a different philosophy of computing.
It offers several advantages.
Privacy and data control
When a model runs entirely on a personal computer, sensitive prompts and documents can remain on the device.
This can matter for lawyers, researchers, journalists, businesses and individuals working with confidential information.
Local processing does not automatically guarantee privacy, but it can reduce dependence on external services.
Freedom from subscription costs
Cloud AI services generally operate through subscriptions, usage limits or API pricing.
Local inference requires an upfront hardware investment, but once the hardware is owned, users can run supported models without paying a cloud provider for every request.
There are still electricity, maintenance and hardware costs.
Independence from corporate decisions
Cloud providers can change their prices, restrict access, discontinue models or introduce new usage limits.
A locally downloaded model can provide greater independence, subject to its licence and hardware requirements.
Offline access
Local models can operate without an internet connection once everything required has been downloaded.
This can be useful in remote locations, unreliable network environments and situations involving sensitive work.
These benefits explain why many users are willing to accept lower performance in exchange for greater control.
Altman’s argument focuses on resource utilization. Local AI advocates are also thinking about ownership, privacy and independence.
Those are not the same objectives.
A system that maximizes memory utilization might not maximize individual freedom.
6. The economic contradiction: OpenAI wants cheaper computing, but who controls the supply?
One of the most interesting parts of the podcast is that Altman did not simply criticize consumers for owning memory.
He also acknowledged that the semiconductor market is not responding to higher prices as efficiently as he would expect.
He described the invisible hand of the market as failing in the current situation.
His explanation was rooted in the historical economics of memory manufacturing.
DRAM manufacturers have experienced repeated boom-and-bust cycles. Investing billions in new fabrication facilities can become disastrous when supply eventually exceeds demand and prices collapse.
Altman argued that manufacturers remember these earlier cycles and remain cautious about expanding capacity, even when today’s AI-driven demand is exceptionally strong.
This is a legitimate economic problem.
Building semiconductor factories is not like opening another retail store.
It requires enormous investment, specialized equipment, advanced engineering, skilled workers and years of development.
New capacity cannot appear immediately because prices have increased.
But there is another side to the issue.
OpenAI is not simply an observer of the market.
It is one of the companies seeking enormous quantities of computing infrastructure.
During the same conversation, Altman discussed OpenAI’s own chip development, its efforts to learn how to build data centres and the possibility of selling its chip in the future.
He also described using robotics to help manufacture components and build infrastructure.
This reveals the strategic direction of the company.
OpenAI wants to reduce the cost of intelligence by gaining greater control over the infrastructure required to produce it.
That is understandable from a business perspective.
However, it creates an obvious tension when the same company argues that individual consumers should reconsider how they use their own computing resources.
The question is not whether OpenAI should be allowed to build infrastructure or whether cloud computing is useful.
It is whether the future of computing should become increasingly dependent on a small number of companies controlling access to its most valuable resources.
7. The overlooked problem: Cloud efficiency can create a new kind of dependency
There is a significant weakness in treating cloud computing as the obvious solution to the memory shortage.
Cloud infrastructure offers efficiency through pooling and shared access.
But centralized systems introduce other costs and risks.
Consider a small business that runs an AI model locally.
It purchases the hardware, installs the software and controls its operating environment.
Now consider the same business relying entirely on a cloud provider.
It no longer needs to purchase as much hardware. However, it must depend on external infrastructure.
Its operating costs become connected to subscription prices, API charges, internet connectivity and the provider’s service policies.
If the cloud provider raises prices, changes its terms or restricts access, the business has limited immediate alternatives.
There are also questions about data residency, regulatory compliance, outages and vendor lock-in.
This does not mean cloud computing is inherently less secure or less reliable. Large cloud providers can invest far more in security, redundancy and infrastructure than most individuals or small businesses.
But centralized efficiency has a trade-off: it concentrates operational control.
A society that depends entirely on a few infrastructure providers could become more vulnerable to decisions made by those providers.
This is particularly important as AI systems move beyond answering questions and begin handling business operations, software development, financial workflows and personal information.
The more essential AI becomes, the more important it is to have alternatives.
8. Why the future is probably neither entirely local nor entirely cloud-based
The debate is often presented as if one model must defeat the other.
That is a false choice.
Cloud computing and local computing solve different problems.
A large AI model performing complex reasoning across enormous datasets may benefit from specialized data-centre infrastructure.
A personal assistant processing private documents may be better suited to a local device.
A company might use cloud infrastructure for demanding workloads while keeping sensitive operations on its own hardware.
A smartphone might run smaller models locally and connect to a cloud model when it needs additional capabilities.
This hybrid approach is likely to become increasingly important.
The industry is already moving toward smaller, more efficient models, quantization, model compression and hardware optimized for on-device inference.
These developments can reduce memory requirements and make local AI more practical.
However, efficiency improvements can also create a rebound effect.
When running AI becomes cheaper, people may use it more frequently, for longer tasks and across more applications.
The resulting increase in demand can consume much of the efficiency gained.
This is one reason why better models do not necessarily mean lower overall demand for computing infrastructure.
The question is not simply how much memory one AI request requires.
It is how many billions of requests society will eventually make.
9. What happens next? Three possible futures for computing
The current memory shortage could lead to several different outcomes.
Scenario 1: AI infrastructure wins the allocation battle
Memory manufacturers continue prioritizing high-margin AI and server products.
Data-centre operators secure large supply agreements, while consumer electronics manufacturers face higher input costs.
Smartphone and laptop makers respond by increasing prices, reducing entry-level configurations or redesigning products around lower memory requirements.
Cloud AI becomes more powerful, but consumers face greater costs for owning capable personal hardware.
This is the scenario that local AI advocates fear.
Scenario 2: Manufacturing capacity catches up
Higher prices encourage semiconductor companies to invest in additional production capacity.
New facilities, improved manufacturing processes and advances in memory packaging gradually increase supply.
Competition improves and prices eventually stabilize.
Altman’s argument could then become less relevant because the world would have more memory available for both data centres and personal computers.
However, factory construction and qualification take time. Even a major investment announcement does not immediately translate into usable production.
Scenario 3: Computing becomes more efficient and distributed
A different possibility is that the industry changes the way it uses memory.
Smaller models, more efficient inference, improved hardware architectures and hybrid computing reduce the amount of expensive infrastructure needed for individual tasks.
Personal devices become capable of running increasingly sophisticated AI models.
Cloud systems continue to handle demanding workloads, but local computing remains an important part of the ecosystem.
This would preserve some of the advantages of both approaches.
The most desirable outcome for consumers is arguably a combination of the second and third scenarios.
More manufacturing capacity and more efficient computing would reduce the need to treat personal RAM and cloud memory as competing priorities.
10. What this means for consumers, particularly in India
For Indian consumers, the debate has a direct financial dimension.
India is an extremely price-sensitive electronics market.
A relatively small increase in the cost of a smartphone can influence purchasing decisions, especially in the entry-level and mid-range segments.
If memory costs remain elevated into 2027, manufacturers could face difficult decisions.
They may increase prices, reduce specifications, discontinue less profitable models or push customers toward more expensive product categories.
For consumers, a few practical considerations become important.
Do not automatically choose the lowest-RAM configuration. If you intend to keep a laptop for several years, insufficient memory can make the device frustrating as applications become more demanding.
Check whether memory is upgradeable. Some laptops have soldered RAM, making future upgrades impossible.
Compare total ownership costs. A cheaper laptop with insufficient memory might become a poor investment if you need to replace it sooner.
Consider refurbished hardware. A well-maintained previous-generation laptop can sometimes offer better value than a new entry-level model.
Avoid buying hardware purely for AI marketing. Evaluate the applications you actually use rather than paying a premium for features that may not provide meaningful benefits.
For people interested in local AI, sufficient system memory and compatible hardware will remain important considerations.
The right amount depends on the model, its quantization, context length and workload. More RAM alone does not guarantee that a computer can run every AI model.
11. What This Means for Consumers, Particularly in America
For American consumers, the AI memory shortage is no longer just a semiconductor industry problem. It is increasingly becoming a household spending issue.
The United States is at the centre of the AI infrastructure race. Major technology companies are investing billions of dollars in data centres, advanced processors and memory systems to support increasingly powerful AI models. But the same semiconductor supply chain also supports the laptops, smartphones, gaming consoles and connected devices Americans use every day.
As demand from AI companies grows, consumers are discovering an uncomfortable consequence: the technology boom promising to make computing more powerful could also make ordinary computing more expensive.
American laptop buyers could pay more for the same performance
One of the clearest effects is appearing in the personal computer market.
Memory is a fundamental component of a computer. It determines how smoothly a machine can handle multiple applications, demanding software and increasingly sophisticated AI workloads.
When DRAM prices rise, manufacturers face a difficult choice: increase retail prices, absorb lower profit margins or reduce specifications.
Research firm Gartner projected in February 2026 that combined DRAM and SSD prices could increase 130% by the end of the year compared with 2025. It estimated that this could push PC prices up 17% and smartphone prices up 13%.
These are forecasts rather than guaranteed retail price increases, but they illustrate the potential scale of the problem.
For American shoppers, the consequences could include:
- Paying more for a laptop with the same amount of RAM.
- Getting 8GB or 16GB instead of a more generous memory configuration at a similar price.
- Finding fewer attractive options in the entry-level laptop market.
- Paying additional costs for gaming PCs and workstation upgrades.
- Delaying a computer replacement because the upgrade no longer makes financial sense.
Gartner also warned that the traditional sub-$500 PC category could disappear by 2028 if memory costs remain elevated.
That would matter particularly to students, families, schools and small businesses that depend on affordable computers.
Gaming could become a more expensive hobby
The shortage also has implications for American PC gamers.
Gaming computers rely on several types of memory, including system RAM and graphics memory. While these are different components, both are exposed to pressures across the wider semiconductor supply chain.
When memory prices increase, building or upgrading a gaming PC becomes more expensive.
This can affect:
- DDR4 and DDR5 system memory kits.
- Graphics cards using GDDR memory.
- SSDs used to store increasingly large games.
- Prebuilt gaming desktops.
- Gaming laptops and high-performance workstations.
Consumer Reports documented an example of how quickly the market changed: a 32GB DDR5 memory kit that cost $90 on September 11, 2025, was priced at $269.99 by November 21 and $349.99 when the publication checked again in December.
That is an individual retail example, not a universal market price, but it illustrates how rapidly the cost of a seemingly ordinary PC upgrade can change.
For gamers, the consequences go beyond paying more for hardware. Higher costs could encourage people to keep older systems longer, delay upgrades or choose lower-specification machines.
Smartphones and connected devices are not immune
The memory shortage is also affecting the economics of smartphone manufacturing.
American buyers could face higher prices, fewer attractive budget configurations and manufacturers passing more component costs on to customers.
The effects are not necessarily limited to phones.
Modern televisions, streaming devices, smart-home products, networking equipment and other electronics all rely on semiconductor memory.
Even products that do not advertise AI features can become more expensive because their manufacturers compete for components within the same supply ecosystem.
This is an important distinction: consumers do not have to buy an AI product to feel the financial effects of the AI infrastructure boom.
Cars and medical equipment could also face pressure
Perhaps the least discussed consequence is that memory chips are not exclusively technology-company components.
Modern vehicles depend on semiconductor memory for infotainment systems, driver-assistance technologies, navigation, control systems and other electronic functions.
Medical equipment, telecommunications infrastructure and industrial systems also rely on memory chips.
In June 2026, Reuters reported that American business groups representing automakers, retailers, medical-device manufacturers and telecommunications companies had warned US government departments that rising AI-related memory demand was increasing costs and threatening supplies.
The warning matters because it demonstrates that the shortage is not simply a dispute between laptop buyers and AI companies.
It has implications for multiple industries that support the American economy.
The American consumer faces a difficult choice: Buy now or wait?
For households planning to replace a laptop, upgrade a gaming PC or purchase a smartphone, the timing of a purchase could become more important.
However, there is no guarantee that buying immediately will always be cheaper. Retail promotions, existing inventory, product launches and manufacturer pricing decisions can temporarily offset higher component costs.
Consumers should therefore focus on actual product value rather than panic-buying.
A few practical steps can help:
- Check memory upgrade options: A laptop with replaceable RAM may offer greater flexibility than one with permanently soldered memory.
- Compare configurations carefully: Look at RAM and SSD capacity, not just the advertised processor.
- Consider refurbished devices: Previous-generation hardware may offer better value than a new entry-level model.
- Avoid unnecessary upgrades: If your current computer performs well, replacing it solely because of fears about future prices may not be worthwhile.
- Think about long-term requirements: Buying too little RAM to save money today can make a laptop frustrating to use in a few years.
The bigger concern is that persistent memory inflation could change the traditional American consumer electronics market, where buyers have become accustomed to getting more computing power for roughly the same amount of money.
That expectation may become harder to sustain.
12. The bigger question: Is computing becoming a privilege rather than a basic capability?
There is a philosophical question beneath this entire controversy.
For decades, personal computing has followed a relatively simple principle: People buy hardware, install software and decide what they want to do with their machines.
The cloud changed that relationship by moving more computing into centralized infrastructure.
Artificial intelligence is accelerating the transition.
Today, the most capable AI systems increasingly depend on specialized processors, enormous memory pools and infrastructure that ordinary consumers cannot realistically purchase or operate.
This creates a potential divide.
On one side are companies with access to enormous computing budgets, long-term chip contracts and specialized infrastructure.
On the other are individuals and small businesses that must purchase increasingly expensive hardware or rent computing capacity from those companies.
If the gap continues widening, access to advanced computing could become increasingly dependent on purchasing power and corporate relationships.
That does not mean personal computers will disappear.
But it does raise an important question about the direction of technological progress.
Should the next generation of computing make individuals more independent, or make them increasingly dependent on infrastructure controlled by a handful of powerful companies?
The answer will influence not just the AI industry but the broader digital economy.
Conclusion: The real shortage is not just memory — it is control over computing
Sam Altman’s argument deserves serious consideration.
When memory is scarce, sharing expensive resources across many users can improve utilization. Cloud infrastructure has genuine economic advantages, and the semiconductor industry needs more manufacturing capacity to support growing demand.
But utilization is not the only measure of a healthy computing ecosystem.
Personal computers provide privacy, independence, resilience and ownership. Those benefits cannot be fully captured by measuring how many hours a memory chip remains active.
Derek Thompson’s analysis of the AI infrastructure boom provides an additional warning: the physical expansion of artificial intelligence is becoming an economic and political issue, not merely a technological one.
The AI industry needs more data centres. It needs more memory, electricity, processors and manufacturing capacity. Yet the people expected to accommodate this expansion are also the people buying the phones and computers whose prices are being affected by the same infrastructure race.
There is a contradiction worth watching.
AI companies want to make intelligence cheaper and more accessible. But if the path toward that future makes personal computing increasingly expensive and dependent on centralized providers, the benefits may not be distributed as widely as the technology’s promises suggest.
Perhaps the real question is not whether your laptop’s RAM is being wasted.
It is whether the future of computing should be designed primarily around maximizing the efficiency of corporate infrastructure — or preserving people’s ability to own, control and use their own technology.
Because the most valuable resource in the AI era may not simply be memory. It may be the freedom to decide who gets to use it.
Sources and further reading
- Primary source: Cheeky Pint – Sam Altman interview and full transcript, October 29, 2026.
- Derek Thompson: The Great American AI Rebellion.
- Derek Thompson: The AI Story Wall Street Doesn’t Want to Hear — discussion of AI infrastructure investment and demand.
- TrendForce: AI Server Demand Sustains Memory Contract Price Increases in 4Q26.
- The Indian Express: Why AI Is Making Budget Smartphones More Expensive.
- TechCrunch: AI-driven memory crunch jolts India’s smartphone market.
- Reuters: Samsung’s October 29, 2026 results and outlook on AI memory demand and supply constraints.