2025-12-25 06:27:43

The underlying logic of this AI boom isn't that complicated: big internet companies spend money to buy NVDA graphics cards, which provide computing power, then break down that computing power into tokens to sell externally. Basically, it's a retail business of selling computing power.

Following this line, a few very sobering conclusions can be drawn.

**The stronger the computing power, the cheaper the tokens, but the demand actually increases**

Chip performance improves year after year, and the cost per token will definitely decrease. But there's a catch: pricing power is not in your hands at all; it’s in the hands of the model developers.

As long as programmers still need to write code, analysts need to perform data analysis, product managers want to improve efficiency, and enterprises pursue automation, token consumption will not stop. Instead, it will become more and more intense.

There are also a bunch of other players now: API wrapper startups, AI middlemen, micro-innovation SaaS, various "AI+" tools... The brokerage software you use, office tools, video apps, design platforms—all are adding AI features. The token consumption of upper-layer applications will ultimately impact the lower layers—the demand for hardware computing power.

**The US's computing power shortage is a systemic issue, not a short-term fluctuation**

Many underestimate the difficulty of building data centers in the US. Land approval, power infrastructure, rack installation, cooling systems, union costs... each of these steps is a trap in the US.

View Original

This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.

10 Likes

Reward
10
4
Repost
Share

Comment

0/400

YieldFarmRefugee

· 2h ago

This is the essence of cutting leeks. The pricing power is all in the hands of OpenAI, and we can only passively get cut. --- So true, the token usage never stops, and in the end, we still have to pay NVDA. --- Building data centers in the US alone can ruin you with just a labor union, no wonder computing power has always been tight. --- Now even a broken app has to incorporate AI features, and token consumption is really outrageous. --- So the ones who always make money are the sellers of shovels, not the gold diggers. This principle has been told by our ancestors long ago. --- If the pricing power is in the hands of model manufacturers, how can we entrepreneurs play? We are being tightly restricted. --- It feels like a new round of cloud computing bubble, just with a different disguise called AI. --- Chip performance upgrades, cheaper tokens, and even greater demand—this closed-loop design is brilliant.

View OriginalReply0

CoffeeNFTs

· 2h ago

Basically, it's just a graphics card business, but wrapped in an AI shell. The fact that pricing power is locked in is incredible; the retail investors will always be retail investors. Wait, the model of middlemen making a profit margin seems to never die. Infinite growth in computing power but costs actually decrease? Why does this script feel so familiar? The many issues with building data centers in the US are much more complicated than I imagined. It's really just a gamble on who can survive until the final round of funding. Token consumption is the same as current gas fees, only going up and never down. This logic also applies to ETH mining—same old story, different packaging. So, at its core, it's still a hardware business. Don't be fooled by the flashy stuff.

View OriginalReply0

YieldWhisperer

· 2h ago

actually let me do the math on this... tokens get cheaper but demand explodes? that's just classic unsustainable tokenomics dressed up as innovation. seen this exact death spiral pattern before

Reply0

ColdWalletGuardian

· 2h ago

Basically, big companies are just harvesting profits, and the token depreciation rate can't keep up with demand growth. In the end, hardware vendors still make money. The pricing power is really tight; model manufacturers have it firmly in their grasp. The cost of building data centers in the US, those union folks really know how to get it done. Now there are all kinds of "AI+" tools flooding the market, and the consumption is indeed terrifying. Can NVDA's stock price not skyrocket? The entire industry chain's big players all have to pay up. The logic here is really just capital exploiting profits; whoever falls behind will die. The cheaper the token, the more it gets used? Sounds like a vicious cycle. It's really just about who can survive the longest in this gold rush.

View OriginalReply0