Looking at NVIDIA's Long-Term Logic from CCTV's Naming of "Token"

Recently, I saw that China Central Television (CCTV) has officially translated "token" as "词元" (word element).
On the surface, this seems like just a translation issue, but I think its underlying significance is substantial. Because once a technical concept is consistently used by official media, it often indicates that it is no longer just a niche term within geek circles, model circles, or investment circles, but is beginning to enter broader public discourse.
More importantly, a token, or "词元，" is essentially the fundamental unit of measurement in the AI era.
In the past, when we looked at the internet, the core metrics were DAU, MAU, time spent, and traffic;
when we looked at cloud computing, the core metrics were computing power, storage, bandwidth, and cloud revenue;
in the AI era, a very important new metric might be token usage volume.
Because every Q&A session, every line of code written, every document summarized, every task automatically executed by an Agent consumes tokens in the background. The greater the token consumption, the more the AI is being called upon; the more the AI is called upon, the stronger the inference demand; the stronger the inference demand, the greater the underlying need for computing power, networks, storage, data centers, and electricity.
Even more staggering is that, according to public reports, China's average daily token usage was about 100 billion in early 2024, reached 100 trillion by the end of 2025, and surpassed 140 trillion by March 2026.
This number is astonishing.
If DAU and MAU were user activity metrics for the mobile internet era, then token usage volume might be the real usage intensity metric for the AI era.
China's daily token consumption at the 140 trillion level indicates that AI is no longer just in PowerPoint presentations, nor is it just a novelty being tried by a few people in chats. It has begun to enter a phase of real-world usage on both the consumer (C-end) and business (B-end) sides.
Whether it's C-end search, writing, coding, office work, or B-end customer service, risk control, knowledge bases, healthcare, education, or government affairs, all will continuously consume tokens in the background.
This is also why I believe AI infrastructure development is not slowing down.
Many people looking at AI focus on one question: Will large model training peak? Will cloud providers' capital expenditure slow down? Will too many GPUs be bought?
These questions are certainly important, but focusing solely on training easily underestimates the subsequent inference demand.
Training is phased; inference is continuous.
Training is like building a factory; inference is like the factory operating daily.
As long as AI applications truly proliferate, inference demand may be more persistent, more dispersed, and harder to satisfy in one go than training demand.
So who benefits most from this in the end?
I think the answer is clear: the shovel sellers.
AI application companies are still battling over business models, model companies are still battling over prices, and many C-end products are still subsidizing users. But as long as the entire industry continues to train, infer, deploy, compete for speed, and compete for user experience, the underlying computing power infrastructure remains unavoidable.
And NVIDIA's greatest strength has long been more than just selling GPUs.
It sells GPUs, the CUDA ecosystem, networking, complete rack systems, data center solutions, and a whole set of infrastructure surrounding the AI factory. In other words, NVIDIA has shifted from "selling cards" to "selling AI infrastructure."
This is also the core reason I am long-term bullish on NVIDIA.
Not because of how much its stock has risen in the short term, nor because the market is hyping AI again, but because token/词元 usage itself is becoming the "electricity meter" for AI demand expansion. As long as this meter keeps spinning at high speed, it's hard to say the underlying demand for computing power is over.
Of course, this doesn't mean NVIDIA's stock price will only go up and never down.
The market will still trade on several issues in the short term:
Will AI capital expenditure overheat?
Can cloud providers' investments generate sufficient returns?
Will ASICs and in-house chips divert some demand?
Will geopolitical restrictions affect the Chinese market?
Will high valuations be punished if growth slows?
These risks are real and cannot be ignored.
But I think we need to distinguish two things:
Will AI stocks correct because they've risen too much? Of course they will.
Has the AI industry trend slowed down? It doesn't seem like it at the moment.
There's a saying in investing that I strongly agree with:
Short term, look at sentiment; medium term, look at performance; long term, look at the industry.
The same applies to NVIDIA.
In the short term, NVIDIA will be affected by market sentiment, valuation fluctuations, geopolitical risks, and interest rate expectations. It will correct after rising too much and rebound after falling too much.
In the medium term, the market will watch its quarterly earnings reports, seeing if data center revenue, gross margins, orders, Blackwell/Rubin deliveries, and cloud provider capital expenditures can continue to be delivered.
But in the long term, what truly determines NVIDIA's value is whether the AI industry itself continues to expand.
If token usage continues to skyrocket, if AI applications move from novelty to daily productivity, if inference demand keeps growing, then NVIDIA, as a core supplier of AI infrastructure, has its long-term thesis intact.
So for me, CCTV naming "词元" actually reinforces one judgment:
AI is shifting from a conceptual narrative to an infrastructure narrative.
And on this main line of infrastructure, NVIDIA remains one of the most core shovel sellers.
I can't predict how the stock price will fluctuate in the short term.
But as long as there are no red flags in the fundamentals, as long as AI token consumption continues to grow, NVIDIA's long-term logic remains intact.
For such a company, I prefer to take a long-term perspective rather than easily abandon a core position due to short-term price movements.