Should countries have a upper limit on the ratio of server:client memory supply chain capacity? If no one can buy client hardware to access the cloud, how would cloud providers survive after driving their customers to extinction?
It shouldn't be possible for one holding company (OpenAI) to silently buy all available memory wafer capacity from Samsung and SK Hynix, before the rest of civilization even has the opportunity to make a counteroffer.
What if we realize that 8 GB of memory is actually a tremendous amount, and experience a resurgence in desktop operating systems as people begin to prioritize memory for productive computation again instead of using up a gigabyte for a chat client?
> Should countries have an upper limit on the ratio of server:client memory supply chain capacity? If no one can buy client hardware to access the cloud, how would cloud providers survive after driving their customers to extinction?
You mean a central planning, command and control economy? There is a lot of history of countries trying these things and they don’t have the outcome you want.
DRAM manufacturing is a global business. If one country starts imposing purchase limits for whatever reason, the DRAM consumers are going to laugh as they move their data centers and operations to another country that doesn’t try to impose their laws on a global market.
> It shouldn't be possible for one holding company (OpenAI) to silently buy all available memory wafer capacity from Samsung and SK Hynix, before the rest of civilization even has the opportunity to make a counteroffer.
Good news: That’s not how markets work. DRAM manufacturers don’t list a price and then let OpenAI buy it all up. Contracts are negotiated. Market prices fluctuate.
No supplier of anything is going to let all of their inventory disappear to one buyer without letting the market have a chance to bid the price higher.
Had Samsung known SK Hynix was about to commit a similar chunk of supply — or vice-versa — the pricing and terms would have likely been different. It’s entirely conceivable they wouldn’t have both agreed to supply such a substantial part of global supply if they had known more...but at the end of the day - OpenAI did succeed in keeping the circles tight, locking down the NDAs, and leveraging the fact that these companies assumed the other wasn’t giving up this much wafer volume simultaneously…in order to make a surgical strike on the global RAM supply chain..
The Chinese government directed CXMT to convert production from DDR4 to DDR5 as soon as the company was able. The order was said to have been given in the 4th quarter of 2024, and the price transition changed from a decrease to an increase in the middle of March 2025.. A wholesale conversion from DDR4 to DDR5 would probably be very expensive to perform, and would thus be unusual for a company that was focused on profitability. As a government-owned company, CXMT does not need to consistently turn a profit, and this was a factor in the government’s decision to suddenly switch from DDR4 to DDR5.
Any market where the prices are negotiated is a bad market.
That means that it is a market where there is an asymmetry of power between vendors and buyers, caused by the fact that the vendors know more than the buyers, i.e. only the vendors know the right price for their products.
Therefore in such a market there are winners and losers among the buyers. Those buyers who buy quantities great enough to have negotiating power and who have knowledge about the right prices can buy at those prices, while the other buyers are fooled by the vendors into paying excessive prices.
The fact that the big-volume buyers deserve discounts has nothing to do with price negotiation. In a good market, where there is enough competition, the volume discounts can be public and available for anyone.
Also, a public auction for a product where the demand exceeds the offer has nothing to do with a secret price negotiation.
Any vendor who promotes price negotiation is a vendor who desires to steal money from its customers, instead of performing a mutually advantageous exchange.
Antitrust laws don't work because they're subjective and are enforced by political appointees.
The simpler solution is a tax on scale -- a graduated corporate revenue tax, aggregated across any group of entities which meet the common control [1] criteria. Then it's just a tax, and you simply have to collect it. Very little wiggle room.
If splitting your company in half wouldn't impair any of its lines of business, the CEO has a powerful financial incentive (lower tax rates on the two halves) to do so.
I don't think that works in the situation of fabs. They are big and expensive pieces of tech with the latest fabs being the most expensive to construct.
You can't exactly break up a chip manufacturer when they have just 1 or 2 plants tooled for the latest memory.
IMO, recognizing chip fabrication as a national security asset and turning it into a public corporation would be the better way to go. Let the likes of intel/amd/or micron continue developing chips. But also, take control of the most expensive and risky part of chip manufacturing to make sure we don't fall behind due to corporate budget cuts. You also keep and continue to build expertise in a vital part of modern society.
Is that a joke? Socializing important industries has always been a disaster. Government run organizations have never been able to innovate on a sustained basis.
Lots have even in America. 2 famous ones are the NIH and the National Laboratories.
In fact, the largest, most advanced, and best known semiconductor manufacturer is primarily government owned: TSMC.
The only thing that gets in the way of their ability to sustain innovation is administrations hostile to publicly funded research.
Outside of innovative industries, there are plenty of examples of important government ran organizations aren't "disasters". Some of which can only be effectively ran via government. For example, healthcare.
What's been a disaster is relying on privatization and capitalism to solve all problems. That's the system of government we had in the dark ages.
Not only that, there was government-led chip research in Taiwan before TSMC (ITRI). And it was going nowhere. If Morris had stayed in ITRI, Taiwan would probably look like a developing country whose primary value is to host the US military bases today.
TSMC's largest shareholder remains the Taiwanese government. And it would not have been a thing without the direct intervention of the government through ITRI.
It would not exist without the government's direct intervention.
I mean, to take that one step further, if the underlying process-node technology (e.g. EUV) were nationalized, then you an entire nation-state's budget (and ability to get cheap loans) could be thrown at the problem of rapid horizontal buildout of fab capacity. Economics similar to nuclear power generation.
Exactly. And even if it ultimately doesn't turn a profit (which, who knows, it probably will turn a profit) you've still created a pretty favorable circumstance for chip manufacturers.
There's a reason why basically only Intel does inhouse fabrication and even they have had to rely out outsourcing it.
What do you think antitrust laws are going to do? Force Micron to continue producing consumer DIMMs in a heavily commoditized market? Stop server builders from building servers? Stop companies from building data centers around the globe?
Monopolies never last in the tech industry. IBM had a monopoly on mainframes. And they even kind of still do, but now no one cares because disruptive innovations in other computing platforms have made mainframes nearly irrelevant. Now startups like OpenAI and Perplexity are using disruptive innovations to rapidly make Google's traditional web search business irrelevant.
OpenAI is having to pull insane amounts of funding to even stay alive.
Google is about to lay waste to everyone.
Google is using their nation state wealth to once again dominate a new sector. Wealth gained from unfair monopolization of search and web and mobile.
Google changed the notion of the URL bar to search. They control every ingress. Now, if you want to access a name brand registered trademark, it flows through Google search. Brands have to pay extortion money to Google to keep others from sniping their rightful name brand.
Google gets even more money because there's a bidding war.
Google pays to put themselves as middle men in front of nearly every web access.
Google doesn't just have a monopoly on this, it's downright unethical and should be tried in court or have laws written to make this illegal.
I have no love for OpenAI, but how do they even compete with the hundreds of billions of dollars this nets?
> Monopolies never last in the tech industry
Google is an invasive species in the ecosystem. They're killing viable competition by engorging themselves and taxing non-productively.
Capitalism should be hard. It should be live or die regardless of size or scale. Google is barely breaking a sweat.
These suggestions seem to me part of an absurd struggle against basic market economics.
This isn’t antitrust because the companies aren’t reselling it to you at a much higher price after cornering the market (cough cough Ticketmaster & scalpers).
> These suggestions seem to me part of an absurd struggle against basic market economics
Perhaps the limited competition caused by 25+ memory manufacturers consolidating down to a 3-member cartel is a sign of market failure. I use "cartel" deservedly, as the RAM manufacturers were found guilty of price-fixing in multiple times in the past.
And they're effectively saying they've had enough of running call centers, tracing lost parcels, weirdo customers who show up at the factory, running marketing campaigns etc.
A consumer facing business is a lot of overhead, and since more and more hardware now has soldered ram, it is a shrinking business too.
Shrinking businesses are super hard to run - it's far easier to grow a business than shrink it whilst maintaining the same margins.
> And they're effectively saying they've had enough of running call centers, tracing lost parcels, weirdo customers who show up at the factory, running marketing campaigns etc.
When this is a company's core complaint, then the usual strategy for getting out of the D2C business (without losing D2C revenue) is finding a channel partner willing to absorb the dealflow. I.e. turning your B2C channel into a single B2B(2C) enterprise customer.
It shouldn't be possible for one holding company (OpenAI) to silently buy all available memory wafer capacity from Samsung and SK Hynix, before the rest of civilization even has the opportunity to make a counteroffer.