You may have heard about AI. Better still, you may already be using ChatGPT, Claude or Gemini while reading this. Artificial intelligence has become a part of everyday life for almost anyone connected to the digital world. Although AI has existed for decades, recent advances in large language models (LLMs) have taken its capabilities far beyond what many imagined possible.
Yet despite the rapid progress, companies such as OpenAI, Anthropic, Meta and Google are nowhere near finished. In many ways, they have only begun exploring what AI can do. The ambition is for AI to become capable of thinking and reasoning like humans, or potentially surpassing human abilities, without requiring constant prompting. It is an exciting and unsettling prospect, but whenever that point arrives, it will come with a significant price.
The world is increasingly discovering something that most people already understand: there is no such thing as a free lunch. With AI, however, the cost of entry is becoming particularly high for both individuals and businesses.
What makes the situation potentially concerning is that the expense is not being absorbed by just one group. Everyone involved is paying for it in some form. The financial impact differs from one company or individual to another, but the cost of building, operating and using AI is being felt across the ecosystem.
Here, the cost largely refers to the money required to keep the entire AI infrastructure running. It is not concentrated in one part of the process. From developing AI models to selling and using them, every stage requires substantial investment. This creates a cycle that is difficult to break unless AI models become significantly more efficient or companies find sustainable ways to monetise the technology.
A major share of this investment is flowing into AI data centres, which effectively house the computing power behind services such as ChatGPT, Claude and Gemini. These facilities are extraordinary engineering achievements, but at their core they depend heavily on two components: computing hardware such as GPUs and TPUs, and memory such as HBM and DRAM.
Even smartphones use a much smaller version of similar technology. AI data centres, however, require enormous quantities of RAM for processing and inference, and the available supply is limited.
Only a small number of companies manufacture memory chips at scale, including Samsung, SK Hynix and Micron. Major AI companies have used their financial strength and influence to secure large quantities of chips through advance orders. Some manufacturers, including Micron, have even stepped away from the consumer market for now, reducing the amount of RAM available for products such as smartphones.
Other manufacturers that continue supplying memory for phones and computers have consequently faced tighter availability, leading to intense competition over who gets the available supply and at what price.
Even Apple has not escaped what some in the industry have dubbed the “RAMpocalypse” or “RAMageddon”. Tim Cook recently compared the mismatch between RAM demand and supply to a “hundred-year flood”, before Apple increased MacBook and iPad prices by as much as 85%. iPhone prices have not yet risen, although speculation continues that the upcoming iPhone 18 could launch at a higher price.
Apple has reportedly also been engaging with the White House over the possibility of conducting business with Chinese memory manufacturers such as CXMT to ease some of the supply pressure.
But Apple is far from being the only company affected. Soaring memory and storage prices are creating challenges across industries, and Zoho has now raised concerns as well. The B2B SaaS sector is experiencing the same pressure that has been hitting consumer electronics.
Zoho co-founder and chief scientist Sridhar Vembu frequently uses social media to share observations and updates that can either inspire optimism or offer a reality check. One of his recent posts did the latter, reinforcing concerns that India Today Tech has been highlighting for months.
The short version is simple:
Hey RAM! The memory crunch is coming for everyone
“Memory prices (are) up 500% in 12 months and 10x the lowest level. Memory prices, along with AI token prices, have made business very difficult,” Vembu wrote on X. “We have held back from raising prices, but it is becoming hard.”
The post did not indicate whether Zoho is preparing to increase its prices. However, it represents a warning sign. For a company that has traditionally preferred a “boring and predictable” approach to business, raising prices would presumably be considered a measure of last resort. Yet the mounting pressure is becoming increasingly difficult to ignore.
Earlier this year, Ramprakash Ramamoorthy, Director of AI Research at Zoho, told India Today Tech just how quickly enterprise hardware expenses were rising.
“Generic servers have seen their market prices surge by four times compared to December 2025,” Ramamoorthy said. He also noted that obtaining the necessary components had become increasingly difficult because demand was substantially outpacing supply, with procurement delays potentially lasting multiple quarters.
The issue extends beyond individual products and technologies to the way businesses structure their finances. Traditionally, the financial model for software companies such as Zoho was relatively straightforward: employee salaries represented the largest expense, operating costs such as offices and travel came next, and AI inference remained a comparatively small and manageable cost.
That hierarchy is now changing.
“Inference cost is now a line item and not a rounding error,” Ramamoorthy had said. He explained that for a typical software company, employees would historically account for the largest expense, followed by operational costs and then server expenses, including inference. That order is changing rapidly, with inference costs potentially becoming the biggest expense for software companies in the near future.
Vembu’s comments indicate that companies attempting to avoid passing these expenses on to customers are increasingly seeing pressure on their margins. His reference was an industry report compiled by Tom’s Hardware, which suggested that high-density server memory, including 128GB enterprise DDR5 modules, had increased in price by as much as 500% over the previous year, approaching ten times historical lows.
A growing amount of industry data points towards an unprecedented increase in memory costs. There is also little indication that the situation will ease soon, with some reports suggesting the supply crunch could continue into next year and beyond.
That raises an obvious question: what can companies do?
For Zoho, one answer has been to develop more of its own technology and hardware, including the Nathu La server. But the company is also looking at the problem from a software-engineering perspective.
Vembu explained that programming languages were traditionally created around the assumption that memory was effectively abundant and inexpensive. According to him, that assumption no longer holds.
“A technical note: for a long time, programming languages were designed with the assumption that memory is 'free.' That era has now ended,” Vembu said. “We need highly memory-efficient languages and smarter compilers, even for AI to use. Safety of code and productivity in writing code cannot come at the expense of memory bloat. That is my area of research.”
As infrastructure costs threaten to overtake payroll and businesses struggle to maintain their existing momentum, the winners of the next phase of software may not simply be companies using the largest AI models or producing code the fastest.
Instead, efficiency could become the defining advantage. Companies that develop smarter compilers, memory-efficient programming languages and bare-metal architectures capable of extracting maximum performance from available hardware may be better positioned for what comes next.
In other words, the age of inexpensive and inefficient computing may be coming to an end, while an era focused on leaner and more efficient engineering could be taking its place.
