Everyone is talking about AI as if it has already arrived. Board decks assume it. Vendors sell it as if it is shipping today. Founders are rebuilding their plans around it. I understand the excitement. But I have spent my working life since 1978 as a corporate firefighter, walking into big systems after something broke, and I learned early to ignore the slide deck and go look at the plumbing.
The plumbing of AI is not software. It is chips, memory and factories. And when you look at those, the story changes. The physical supply chain says AI at full scale is not arriving now. On current reporting, it arrives in rations, and the full wave looks more like 2028 or 2029.
Strip AI down to its physical parts and it rests on two places most business leaders rarely think about: South Korea and Taiwan. This article is about why, and what a founder should do about it.
Every serious AI processor needs a special kind of memory sitting right beside it. It is called high bandwidth memory, or HBM.
Here is the plain version. An AI processor is like a very fast kitchen. It can cook at incredible speed, but only if ingredients arrive just as fast. Ordinary memory is a delivery van stuck in traffic. HBM is a set of memory chips stacked on top of each other and placed right next to the processor, so data moves a short distance through many lanes at once. Without it, the most expensive processor in the world sits idle, waiting.
That is why HBM matters. It is not a nice extra. It is a gate. If you cannot get the memory, you cannot ship the AI system, no matter how many processors you have.
Only three companies make HBM at scale. SK Hynix and Samsung are both South Korean. Micron is an American company, with its memory production mostly in Japan and Taiwan.
That is the entire list. Three suppliers for one of the most important components in the AI economy, and two of them are headquartered in one country.
Their capacity is reported to be sold out into 2027. The memory makers and the press covering them have described their HBM output as committed to customers well in advance. I would treat the exact timeframe as reported rather than settled, because these commitments shift. But the direction is clear. The memory that AI depends on has largely been spoken for before most businesses have even placed an order.
Memory is one side. The other side is the logic chip, the processor that does the actual thinking.
The most advanced AI chips in the world are manufactured by one company, TSMC, in Taiwan. The famous names in AI design their own chips, but most of them do not build them. They send the designs to Taiwan, and TSMC turns them into silicon. Then those chips are packaged together with HBM to make the systems that power the AI tools everyone is talking about.
Put the two together and you get a simple picture. The physical foundation of AI runs through two small corners of East Asia: memory on one side, in South Korea, and advanced logic on the other, in Taiwan.
When demand outruns supply, the normal answer is competition. New players see the profit and rush in. That is how markets usually fix shortages.
It does not work quickly here. A leading edge chip fab is one of the most complex things humans build. It takes several years to plan, construct, equip and bring up to good yields. It costs enormous sums. It needs equipment that only a handful of suppliers make, and skilled people who take years to train. And HBM and advanced logic are among the hardest chips on earth to make well. Stacking memory chips and packaging them alongside a processor is its own specialised craft.
There are serious efforts underway. The United States is backing new fabs on home soil. Japan is investing to rebuild its chip industry. The European Union has its own programme to bring more manufacturing to Europe. These are real, and they matter for the long run. But they are coming slowly, and none of them closes the gap in the next couple of years. Building a fab is not like opening a new office. You cannot hurry concrete, clean rooms and yield.
So what happens when supply is fixed and demand is huge? The supply gets rationed, and the biggest buyers go first.
The hyperscalers, the giant cloud and AI companies, have the money and the long term contracts to lock in capacity early. They are at the front of the queue. Everyone else gets AI through those companies, through their clouds, platforms and tools, on their schedule and at their price.
That is why I say the rollout will be strung out. Not a flood, but a drip. The tools will keep improving, and you will see impressive launches. But AI at the scale the headlines imply, running cheaply inside every business, depends on memory and chips that are already committed. On current reporting, the full wave looks closer to 2028 or 2029 than to next quarter.
I look at this from the Philippines, Taiwan's close neighbor just to the south. For years I have tried to understand why Taiwan matters so much to so many powerful countries. My conclusion is simple: Taiwan holds the key to AI, and nothing moves without it. From here, that is not an abstract idea on a slide. It is a supply risk sitting right next door, and it shapes every plan I make.
This brings me to the question people raise whenever Taiwan comes up: will China invade?
I want to be careful here. This is analysis, not a prediction, and nobody outside a few rooms in a few capitals knows the answer. But the supply chain gives us a useful lens.
Taiwan holds the key to advanced AI chips for the next several years. Those fabs are extraordinarily delicate. A war would very likely damage or shut them down, and the skilled people who run them could not simply be replaced. In other words, an invasion would risk destroying the very prize that makes Taiwan so valuable. And that prize is something everyone depends on, China included, since China's own economy is tied into the same global electronics supply chain.
This is often called the silicon shield argument. The idea is that Taiwan's chip industry raises the cost of conflict so high for every party that it acts as a deterrent.
There are real counterarguments, and they deserve a fair hearing. Political goals can outweigh economic ones; leaders do not always act on a spreadsheet. A shield can weaken as other countries build their own fabs, which is exactly what the United States, Japan and Europe are trying to do. Pressure does not have to mean invasion. A blockade or a long squeeze could disrupt supply without a single fab being destroyed. And deterrence only works until it does not.
So my view is not that conflict is impossible. It is that, for the next several years, the chips raise the price of conflict for everyone, and that is a meaningful weight on the scale. For a founder, the practical lesson is the same either way: the region is a supply risk you should plan around, not ignore.
Here is something I have noticed. This conversation is everywhere on YouTube. Long videos break down HBM, TSMC and the geopolitics of chips, and they draw big audiences. But on LinkedIn, where business decisions actually get discussed, I almost never see it framed this way.
My guess is that it feels too technical for business feeds and too geopolitical for marketing feeds, so it falls between the cracks. That is a gap. The leaders making AI budgets for the next three years should understand the physical constraints behind those budgets. If the people who sign the cheques only see the hype and never see the plumbing, they will plan for the wrong timeline.
You cannot change the chip supply chain. You can change how you plan around it. Here is where I would focus.
None of this means waiting. It means working with what is available, being realistic about cost and timing, and preparing your business so you can scale when the supply finally catches up.
The headlines say AI is here. The supply chain says it is arriving in rations, through two countries most business leaders never think about. Plan for the real timeline, not the hype cycle, and you will be ready when the floodgates open.
Charles K Davis has worked as a corporate firefighter since 1978 and is the founder of SERIO Design FX, home of M.A.D. (Maverick Advantage Design) and M.A.P. (Maverick Advantage Platform).
M.A.D. Designs Your Brand. M.A.P. Makes You Known For It.
Stop Reading. Start Seeing.
