Note: This is an experimental and opinionated piece.

---

It's been a few years since AI craze took place in this world, and the current timeline seems unforgiving with ongoing wars, housing crisis, most of human population getting locked in social media echo chambers of brainrot, things becoming more expensive, and with the cost of living demanding higher salaries, well, the math is not mathing anymore. There are as many positive news, but they are not necessarily related to resolving the issues that will be discussed in this article because what history has shown us time and again throughout human history is that social depression is significantly influenced by economic instability (which itself is usually caused by a mix of the aforementioned events). That is to say, would you really care about the advancements in clean energy when you can't afford to pay your Spotify Premium subscription?

That is a rhetorical questioin, so without bringing your Steam Wishlist to attention, let's get straight to the topic. The consumer market is in shambles--DRAM chip shortages casue the prices for hardware to go up: Valve revealed the prices for their new Steam Machine, which is going to start at $1,049 for its cheapest configuration; XBOX increased the prices of its consoles; Apple increased the prices of its devices; and smartphones and PCs in general are going to be more expensive as time goes. It is now very hard for people to build their dream PC as they are forced to save up until they can afford to buy them. The prices, however, keep going up.

As I said, this is all because DRAM became more expensive. But is the reason why it became more expensive--because the produced amount of chips don't meet the demand (a.k.a. "chip shortage")? Since last year, there have been major deals between AI companies--particularly, Anthropic and OpenAI--and memory chip manufacturers (Micron, SK Hynix, and Samsung) to secure massive amounts of DRAM from these three companies (that are the only producers of DRAM chips on the global market). Few things happened, so let's start with OpenAI.

Last year it was reported that OpenAI has made deals with two memory chip manufacturers--Samsung and SK Hynix--as part of their "Stargate initiative" (a joint venture by OpenAI, SoftBank, Oracle, and MGX), for an estimated 40% of the world's DRAM supply. The deals, per OpenAI's official report, are as follows:

What they mean by "wafer starts per month" (WSPM; it's not a typo--it's a term used in semiconductor manufacturing) is that they (DRAM manufacturers) begin processing 900,000 new wafers per month in their DRAM production lines. And they made it clear through their official statements that those 900,000 wafers are directed towards Project Stargate.

Let's do some napkin math (yes, we will also take into account that those numbers above are projections and estimates): each wafer has approximately 50-100 chips (it depends), but let's say it's approximately 50 chips per wafer on average. Multiplying 50 by 900,000 would be ~45,000,000 chips per month. We will subtract 10,000,000 because of possible defects per wafer, even though our approximation already counts the average. That is still a huge amount of chips. Pew Research Center states that a typical data center has around 2,000 to 5,000 servers. The number of data centers in United States is estimated by Data Center Map to be 4,438. An average data center takes 1 to 2 years to build, while large hyperscale data centers (AWS scale) takes 2 to 3 years, and more massive ones (for AI)--from 3 to 7 years.

Continuing with our napkin math, we will assume that a high-end AI training server has around 10 GPUs per server (note that server =/= server rack; NVIDIA's Vera Rubin NVL72 supercomputer rack comes with 72 GPUs; and a typical server has 2-8 GPUs). We will also assume that a single data center OpenAI is drooling over would have around 100,000 servers (another big number), and that they are planning to build 6 of those (one number that is actually official). Multiplying 10 GPUs by 100,000 servers gives us 1,000,000 GPUs in total. Multiplying that by 6 data centers gives us 6,000,000 GPUs.

That is a big number, however, those two (not three, we're not counting Micron!) manufacturers are targeting ~35,000,000 chips *per month* and *just for Stargate* (per Samsung's official report). So, assuming it would take 2 years (24 months) of hard and impressive work to complete all 6 of their data centers, by the time they finish the data centers, the chip manufacturers would produce 24 months * 35,000,000 chips = 840,000,000 chips (almost one billion). 6,000,000 is only 17.14% of approximate monthly supply.

All these three paragraphs are there to say that the amount of wafers they receive that causes "chip shortages" is too big and absurd. The "40%" comes from several news reports, for example:

Even $70B per month for an estimated two years results in an absurd number way above planned $500B (which is 1,680,000,000,000 = $1.7T). This just doesn't sit well with me, all these numbers are too high for a single company, and it makes me believe there is something more that is going on behind the scenes. If the 40% part is true, that would mean that the total amount of DRAM would be more than twice the above approximations, and is the type of number that is too high to allege that it somehow causes so-called chip shortages.

Anyways, continuing the timeline, two months later Micron announced its decision to exit the Crucial consumer memory business (which affects both RAM sticks and SSDs).

This statement alone put everyone into a panic mode, and people started buying Micron's Crucial products before they would go out of stock. The situation is disrupting supply and demand, causing rapid cost spikes and market panic. This, in turn, affects all other hardware dependent on DRAM, such as storage devices (SSDs), graphics cards (VRAM), and more.

Anthropic, the developer of Claude AI, is also signing deals with memory suppliers (all three of them), as well as raising funding at almost $1T post-money valuation with over a dozen investors.

There haven't been any publicly announced plans by Anthropic similar to Project Stargate, and what is known from those reports is that they are also partnering with other tech companies to secure enough computing infrastructure to train and run much larger AI models.

Let's say 1 gigawatt is around 1,000,000 GPUs, roughly speaking, that would be the best-case upper bound. Combining both deals, that would be approximately 10,000,000 GPUs put into action. But this is a rough estimation, ignoring CPUs; it's just that, if we assume that those are AI-training setups, the vast majority of power usage would come from running GPU clusters. And those are major companies we're talking about, with their beast setups. They invest *billions* and, surely, they should generate revenue that is higher than their spendings. And this is the direct reason why people say that this sounds like a bubble.

According to Jon Peddie Research, "the global PC GPU market exceeded 251 million units in 2024, with a 6% year-over-year increase":

Those are tens of millions of GPUs being sold for PCs alone. There is a huge demand in the consumer market. And even if you argue that there is a huge demand among AI companies, then they are getting direct partnerships, they are about to produce an unimaginable number of WSPM, building lots of large-scale data centers like elonia's terafab. Instead, chip manufacturers choose to sign deals with AI companies. But there's more to that: just few days ago, a class action lawsuit was filed by seventeen plaintiffs against Samsung, SK Hynix, and Micron Technology alleging that the three tech manufacturers are colluding to artificially increase the price of RAM. A good summary of this news is covered by The Destin Channel in the video titled "RAM Makers SUED For Price Fixing".

If this is true--and so far we've reasoned that something does not really add up from all these deals--then this could be a good explanation to why chip shortages actually took place: artificial manipulation with the market, which is nothing really new. But what I am personally hoping for is that this gets more attention and makes enough people angry that it could somehow reverse the damage caused by the insane price increases for DRAM chips. Manufacturers enjoy it, AI companies as well--they all benefit from this chaos.

When building a new PC, about half of the total cost goes toward memory sticks and SSDs. This shouldn't be a thing. And it doesn't even make sense, because if those companies were regulated, there would be certain consumer market protection laws. Companies should not be equaled to people, to regular consumers, instead there should be equity--consumers must be prioritized over companies' needs. There should be market equity. I don't want to go down into a slippery slope saying this, but even though there are separate markets for businesses and individual consumers, there should be a case when society becomes dependent on something, it should operate in the interests of society rather than corporate entities, if that makes sense. It's like in budgeting: a certain percentage of produced chips must be secured for regular consumers.

We can only wait and see for what comes next from this mess that major tech companies have made. Gamers are hoping to afford to buy their next beast PC that can run GTA VI and next-generation games built on the ray-tracing tech. NVIDIA is betting on its cloud gaming platform, which if ever becomes a serious choice for gamers would most likely end the personal computing industry. In other words, you won't own your PC anymore--you will outsource the compute power, the services, and everything else from tech monopolies, leaving you with just an interface that would demand you to pay monthly subscriptions. Pay to see the current time. This is the dystopia we're heading towards, unless we start taking things seriously and start fighting against corrupt governments, monopolies and companies that hurt the consumer market.

Hey, I don't want to be all negative about this whole thing, but we are heading towards technocratic dystopia, and major tech companies spend billions of dollars to lobby governments in order to get away from any sort of regulation that would make them pay their fair share of taxes and not abuse their market position in order to get rid of competition.