RAM and storage high prices are putting many companies at risk, now Zoho boss Sridhar Vembu too is saying it

Hey RAM! The memory crunch is coming for everyone and you can blame AI for it.

In Short

• AI data centres need vast memory supplies, intensifying pressure across hardware markets
• Major chipmakers are prioritising enterprise demand, tightening availability for consumer devices
• Apple has already raised some device prices amid the supply mismatch

You may have heard of this thing called AI. Better yet, you might be using ChatGPT, Claude, or Gemini as you are reading this. Artificial intelligence has become ubiquitous in life for almost everyone who is digitally connected. Even though AI has been around for decades, recent breakthroughs in LLMs (short for large language models) have pushed capabilities far beyond human imagination.

But despite all the rapid advancements, AI companies be it OpenAI, Anthropic, Meta, or Google aren’t done yet. They have barely scratched the surface. The goal is that AI should, and it just might in the coming days, become so smart that it is able to think and reason like humans, or even out-think and out-reason them, without any prompting. It is fascinating and scary, but it will happen when it happens. Before that happens, the world is getting to know—which is already known by most people—that there are no free lunches. In the case of AI, the cost of admission is a tad bit too high for individuals and businesses alike.

What’s different—and potentially alarming—is that this cost whether big or small isn’t limited to just one stakeholder. Everyone must bear it. The balance sheets may look different, but we are all paying for AI in one way or another. Cost here refers almost exclusively to the cash flow needed to run the whole operation. This cost isn’t associated with one or even a few aspects of making—and selling and using—AI. At every step of the process, you must put in more money. It’s a vicious cycle that doesn’t stop, not unless AI models become super-efficient, or companies figure out a way to monetise in a way that benefits all parties.

A sizeable chunk of investment is going into AI data centres which hold the brains—literally—to put two and two together every time you talk to ChatGPT, Claude, or Gemini on your phone or PC. These data centres are engineering marvels but mostly they are made of two key components: compute (GPUs/TPUs) and memory (HBM/DRAM). Your smartphone uses a minuscule version of that which is also mind-blowing. Unlike phones, however, AI data centres need gobs of RAM to work—in tech parlance it is called processing, inferencing, so on and so forth—and there is only so much RAM they can get.

Only a handful of companies are making RAM. The likes of Samsung, SK Hynix, and Micron come to mind. Using money and influence, top AI companies have been able to bulk-order chips in advance, so much so that some manufacturers like Micron have completely exited the consumer space—for the time being—meaning that they are not making any more RAM for other gadgets like smartphones. Those still making RAM for your phone and PC have fewer supplies causing a bidding war for who gets access and at what cost.

Even Apple, with all its might, is not immune to the RAMpocalypse or RAMageddon or whatever industry veterans call it. Tim Cook recently compared the ongoing situation of RAM demand-supply mismatch to a “hundred-year flood” before Apple hiked prices of MacBooks and iPads by up to 85 per cent. Current iPhone prices haven’t gone up yet, but speculation is rife that the next iPhone—iPhone 18—could start at a higher asking price than before. Apple has reportedly been courting the White House to allow business with Chinese memory makers like CXMT to offset some of the blow. But Apple is Apple. And Apple is not alone. Sky-rocketing prices of RAM and storage are putting many companies at risk. Latest to join the chorus of raising the red flag is Zoho.

Zoho’s co-founder and chief scientist Sridhar Vembu often takes to social media to share tidbits and updates that inspire both hope and confidence. At times, they give a reality check. One of his recent posts has done just that: inform and confirm what we here at India Today Tech have been consistently reporting about for months now.

The TL;DR version is this:

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 doesn’t say if Zoho is on the verge of raising prices. But it is a sign, and it is not a good sign. For a company that wants to do “boring and predictable” business, raising prices would be a last resort type of thing, one can safely assume. But the writing was on the wall.

Speaking to India Today Tech earlier this year, Ramprakash Ramamoorthy, Director of AI Research at Zoho, had pulled back the curtain on just how fast enterprise hardware costs are spiralling.

“Generic servers have seen their market prices surge by four times compared to December 2025,” Ramamoorthy had revealed. “Securing these components, too, is an uphill task as demand is far exceeding the supply. Procurement delays can stretch into multiple quarters.”

The conversation went above and beyond product and tech to how companies are changing to face the seemingly new normal. In the past, the standard software financial playbook for companies like Zoho was simple: payroll came first, operational overhead came second, and AI inference was a distant, somewhat manageable third. That hierarchy is now flipping upside down.

“Inference cost is now a line item and not a rounding error,” Ramamoorthy had said. “Typically, if you look at a software company's balance sheet, the bigger expense is going to be people. The second thing is going to be operations like travel, office, and all of that. The third thing is your server cost which is your inference cost. Not just AI inference, all inference. But now it is moving up, and eventually, I think next year it will become the first big line item for a software company's expenses.”

Vembu’s post is ample proof that for businesses trying to hold the line on customer pricing, the pressure on gross margins is becoming unsustainable now. His reference point was of course one industry report—collated by Tom’s Hardware—claiming high-density server memory such as 128GB enterprise DDR5 modules has climbed up to 500 per cent over the last twelve months, reaching nearly ten times its historical lows. Internet today is chock-full of data all of which corroborate the same thing which is that memory prices have increased by unprecedented amount. Moreover, things are not expected to cool down anytime soon with reports suggesting the crisis could stretch well into next year and the year after that. Which begs to ask, what is the solution?

For Zoho, at least, one solution is building its own tech and hardware like the Nathu La server. You can read more about it here. But there is more.

“A technical note: for a long time, programming languages were designed with the assumption that memory is 'free.' That era has now ended,” Vembu says. “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.”

In an era where infrastructure expenses threaten to surpass payroll, the winners of the next software cycle won’t necessarily be those who deploy the biggest models or write code the fastest. They will be the teams that master efficiency—building on smarter compilers, memory-conscious languages, and bare-metal architectures designed to squeeze every drop of performance out of every byte. Be that as it may, the era of cheap, careless computing is over. The era of lean engineering has begun.

- Ends