In the past few days, a new term has quietly become popular in the AI ​​community—"token economy."

01. The Token Economy: Rewriting the Underlying Logic of Data Centers

What is a token?

From the most basic definition, a token is simply the smallest unit of information processed by a large model. For example, humans read in units of words; when AI understands the world, it breaks down text, images, and even speech into individual tokens, then calculates and generates data one by one.

But the significance of tokens goes far beyond the technical level. At this year's China Development Forum, Liu Liehong, Director of the National Data Administration, gave a crucial statement: "Tokens are not only the value anchor of the intelligent era, but also the settlement unit connecting technology supply and commercial demand." In other words, future AI services will no longer have vague pricing, but can be precisely measured and directly traded.

At the 2026 GTC conference, NVIDIA founder Jensen Huang bluntly stated, "Tokens are the new commodity."

To illustrate this point, he gave a highly impactful formula: Revenue = Tokens per watt × Available gigawatts. He explained that data centers have now become 24/7 "word factories," taking in electricity and data and outputting words. The revenue of a "factory" depends on the product of the efficiency and scale of word production.

Once this logic holds true, the identity of the data center changes. It is no longer just a computing power hub, but more like a highly energy-dependent manufacturing system.

The impact of this change is clearly felt in the numbers.

For example, a foreign research report shows that ChatGPT processes approximately 200 million requests daily, corresponding to a power consumption of over 500,000 kilowatt-hours. This figure is almost equivalent to the daily electricity consumption of 17,000 American households.

Data from the International Energy Agency shows that in 2024, global data center electricity consumption reached 415 terawatt-hours, approaching the scale of a country's electricity consumption; by 2030, this figure is expected to further climb to 945 terawatt-hours, approaching Japan's total annual electricity consumption.

Even more significant is yet to come: a single large model training session often consumes the equivalent of the annual electricity consumption of a small to medium-sized city. As AI moves from training to large-scale application, the demand for inference is exploding, and computing power usage is growing exponentially.

The latest data from the National Bureau of Data Science confirms this: as of March this year, my country's daily average word-based usage has exceeded 140 trillion, more than a thousand-fold increase compared to the 100 billion level at the beginning of 2024.

This is essentially a very straightforward equation: every AI interaction consumes words; and behind every word-based usage lies the simultaneous consumption of computing power and electricity.

02. When "Word-Based Factories" Start Doing the Calculations, Why Has Photovoltaics Become the Optimal Solution?

If you understand the logic of the word-based economy, you'll find that the competition between data centers, superficially about computing power, is actually about electricity costs.

In a word-based pricing system, every kilowatt-hour of electricity directly impacts profits.

It is against this backdrop that the role of photovoltaics has begun to change. It is no longer just one energy option, but has gradually become part of the cost model.

Let's look at the most intuitive point: price. Over the past few years, the cost of photovoltaic (PV) power generation has continued to decline, falling below that of traditional thermal power in many regions. This change has been amplified in the context of the data economy.

For data centers that need to operate 24/7, electricity is no longer an auxiliary cost but a core expense. Whoever can lock in lower and more stable electricity prices can obtain higher profit margins in data production. In other words, PV offers low-cost electricity that can be locked in long-term.

However, cheapness alone is not enough. With the advancement of global "dual-carbon" goals, the computing power industry is also being incorporated into the framework of green transformation. This year's Government Work Report included "computing and electricity synergy" in the new infrastructure system for the first time, sending a very clear signal that future data centers will not only need to compute quickly but also use green power.

Moreover, policy has already set hard constraints. Previously, the "Special Action Plan for Green and Low-Carbon Development of Data Centers" issued by the National Development and Reform Commission and other departments clearly requires that by the end of 2025, the proportion of green electricity in newly built data centers at national hub nodes must reach more than 80%. This means that the zero-carbon attribute of PV is no longer a bonus but is becoming an entry barrier.

If cost and policy address usability, the next challenge is how to use it more efficiently. This is precisely another advantage of photovoltaics: flexibility.

Compared to traditional energy sources, photovoltaics possesses inherent distributed characteristics, allowing for deployment near data centers to achieve "power generation equals power consumption." This model not only reduces transmission losses but, more importantly, alleviates grid pressure and improves the certainty of electricity usage.

Of course, a real problem exists: photovoltaics is not stable. However, this has given rise to another key combination: photovoltaics + energy storage. Through the regulation of energy storage systems, excess electricity generated during the day can be stored and released at night or during peak demand, supporting the 24/7 operation of data centers. This combination essentially transforms fluctuating energy into a dispatchable and stable supply, ensuring the continuity of word production.

These elements combined are forming a new closed loop: green electricity generation → grid transmission → computing power consumption → word monetization. Within this loop, photovoltaics has gradually moved from the energy supply side to the production side.

At a deeper level, this change is not a random market choice but a result of the resonance between policy and industry. As the integration of computing and power enters the implementation phase, the two previously independent systems of electricity and computing power are being reconnected.

03. More Than Just an Opportunity: Photovoltaics Face a High-Difficulty Upgrade

However, it would be too simplistic to only consider the benefits to photovoltaics. While the meta-economy has indeed opened a new door for photovoltaics, the other side is not without its challenges.

First and foremost, the real contradictions are at play. Data centers require electricity around the clock, at high loads, and cannot be shut down; photovoltaics, on the other hand, have similarly distinct characteristics: generating electricity during the day and resetting at night, with significant fluctuations due to weather conditions.

When word generators become rigidly dependent on electricity, this mismatch is amplified. The power grid must implement more precise scheduling between peak power generation and peak power consumption; otherwise, either the generated electricity cannot be used, or there will be insufficient supply when needed. This means that if photovoltaics wants to truly integrate into the computing power system, it must overcome a hurdle: moving from simply generating electricity to providing stable power.

Once stability is involved, the problem extends from the technical level to the economic level. To match the power demands of data centers, photovoltaic projects often need to be combined with energy storage, intelligent scheduling systems, and even participate in the coordinated optimization of the regional power grid. These capabilities can indeed solve the problem, but they mean higher initial investment, more complex system structures, and longer payback periods.

In other words, in the word generator economy, photovoltaic companies are no longer just calculating the cost of power generation, but a comprehensive calculation of the entire power supply capacity.

Further down the line, there is increased pressure on the technical level. When electricity begins to directly affect word generator output, the requirements for electricity in data centers also increase, demanding not only stability but also fast response, low fluctuation, and high quality. This presents new challenges for photovoltaics itself: higher conversion efficiency, lower degradation, and stronger system synergy.

Meanwhile, old problems persist. As "photovoltaics + computing power" becomes the new narrative, capital will quickly flood in. However, the photovoltaic industry itself has not yet fully emerged from the shadow of overcapacity; once supply expands again, price wars are likely to return.

Even more complex variables come from the rules themselves. The meta-economy is still an evolving system; many rules, whether for green electricity trading mechanisms or the specific paths of computing-electricity synergy, are still being adjusted. For companies, this means simultaneously planning, adapting, and bearing the risks brought by uncertainty.

Considering all these factors, the meta-economy brings to photovoltaics not just new growth, but a comprehensive upgrade.

Opportunities certainly exist, but whether one can seize them depends not on production capacity, but on the ability to become an irreplaceable part of that new system.