Understand the business trends of AI through 6 questions

Note-taking expert says:

Recently, the AI circle has been quite bustling.

Anthropic has become the fastest-growing company in human history. Its annualized revenue soared from $1 billion at the end of 2024 to $47 billion in May 2026. Just yesterday, it completed its Series H financing of $65 billion, with a post-investment valuation of $965 billion, surpassing OpenAI to become the world's most highly valued AI company. It is expected to launch an IPO this autumn.

DeepSeek's valuation has reached $45 billion. The National Major Industry Investment Fund led the first-round financing of 70 billion yuan (approximately $10 billion), and the list of investors has been basically finalized.

Kimi completed a $2 billion financing, with a post-investment valuation exceeding $20 billion. It has raised over $3.9 billion in total in half a year, becoming the champion in financing among domestic large model startups.

Jieyue Xingchen completed nearly $2.5 billion in financing and is removing its red-chip structure to sprint for a Hong Kong stock IPO.

ByteDance has increased its AI infrastructure investment in 2026 from 160 billion yuan to over 200 billion yuan. Bloomberg further reported that the total capital expenditure cap may reach $70 billion (approximately 500 billion yuan).

Doubao launched three tiers of paid subscriptions on May 4th, firing the first shot to end the era of free domestic AI.

If we compare the AI economy to the four seasons of spring, summer, autumn, and winter, what season is it in today? Is it spring, summer, or the transition between spring and summer? Or is it approaching autumn, as the outside world speculates that it is in a bubble period?

The answer is actually hidden in the cycle.

Today, we will use the six-dimensional scoring framework for cycle judgment proposed by Professor Su Dechao from the School of Philosophy at Wuhan University, the founding consultant of the PPE Academy of Notesman, and the teaching tutor of the Western Philosophy course to thoroughly clarify this matter.

I. Quantitative scoring of the six dimensions for cycle judgment

Many people like to use the elimination method to judge the industry cycle: it's not winter, spring has long passed, and autumn doesn't seem to have arrived yet. Finally, they come up with a correct but useless answer: summer. The elimination method only gives you an answer that seems right but doesn't tell you why it's right.

A truly useful judgment should be quantified from specific dimensions.

The "six-dimensional scoring framework for cycle judgment" scores an industry based on six dimensions: "narrative vs. delivery, system connectivity, delivery ability, ROI (return on investment, the same below) rationalization, industry common phenomena, and capital environment". Each dimension is scored from 0 to 2 points. The higher the total score, the closer it is to autumn.

Let's score each dimension one by one.

1. Narrative vs. delivery: From storytelling to checking the accounts

This is the first dimension and also the one where changes are most easily perceived.

In 2022, when ChatGPT first emerged, everyone was saying that "AI will change everything". But no one asked what specific things it could do or how much cost it could save.

Just telling stories was enough. That was spring.

The situation is different today. Doubao has launched paid subscriptions, and the three tiers of prices are clearly written on the page - 68 yuan, 200 yuan, and 500 yuan. The paid functions focus on PPT generation, data analysis, and film and television production. It's not just about storytelling; it's about delivering specific capabilities and then charging based on those capabilities.

OpenAI's advertising platform is even more straightforward: advertisers buy advertising space in ChatGPT and pay per click. On May 5th, the self-service advertising management tool was launched for testing, and the minimum investment threshold of $50,000 was removed, allowing even small and medium-sized enterprises to directly invest.

"AI will change the advertising industry in the future" is a narrative, while "providing you with an advertising channel right now" is delivery. They are two different things.

The narrative still exists, and people are still saying that "AI will change the world", but delivery has already occupied a considerable proportion.

For this dimension, score 1 point.

2. System connectivity: From isolated islands to protocols

In spring, each AI product was an isolated island. If you wanted to integrate ChatGPT into your company's system, you had to write a set of adaptation codes yourself. If you changed to another model, you had to rewrite them all over again.

In April 2026, Google released the Gemini Enterprise Agent Platform, integrating agent management into the enterprise's existing work processes.

Microsoft Copilot is embedded in the Office suite, and Amazon's AI shopping assistant opens sponsored answers to brands. Their boundaries are loosening.

Only when the connectivity reaches a certain level can such a synchronous reaction occur.

Partial protocolization has been achieved, but standardized protocols have not yet become the mainstream.

For this dimension, score 1 point.

3. Delivery ability: From occasional help to stable work

In spring, AI was like an intern, able to help occasionally, but most of the time you had to spend a long time correcting its work.

Doubao has over 300 million users. As of March 2026, its daily average Token call volume has exceeded 120 trillion, a growth of over 1,000 times compared to May 2024, and it has doubled again in the past three months.

OpenAI has 900 million weekly active users, 50 million individual subscription users, and over 9 million enterprise users.

What are these enterprises using AI for? Writing code, reviewing contracts, automatically generating marketing copy, and handling customer service tickets. All these are scenarios where people are replaced from repetitive labor. This is no longer a trial scale; it's using AI to do work.

The delivery has been realized. But has delivery ability become the core competitiveness? It's hard to say. The stories are still being told, but the enterprises that can stand firm based on stable delivery do not yet dominate.

For this dimension, score 1 point.

4. ROI rationalization: From being unable to calculate to starting to calculate

In spring, no one calculated ROI. How much computing power was invested and how much output was obtained? It was impossible to calculate.

Now, some people are starting to calculate: Tencent has significantly increased the price of its Hunyuan API (according to industry news, the price of some models has increased by over 400%), which is due to the hard pressure of computing power costs.

The computing power cost of a single GPT-level model inference request is about $0.01 - $0.03. When the call volume reaches hundreds of millions or billions, the cost will balloon to an astronomical figure. The high-frequency calls of Doubao's 345 million monthly active users are forcing ByteDance to face this problem.

Zhipu has raised its prices three times this year, and Alibaba Cloud has cancelled the basic package of its Bailian platform. Behind these decisions is the same logic: Products with unclear ROI can no longer hold on.

Is ROI clearly calculable? A small number of closed-loop scenarios have started to show ROI, but the calculation methods are vague. Some people calculate the Token cost, some calculate the improvement in human efficiency, and some calculate the customer acquisition conversion rate. The standards are not unified.

For this dimension, score 1 point.

5. Industry common phenomena: From no one talking about profitability to some people starting to doubt

In spring, everyone was expanding, and no one talked about profitability. Burning money to achieve growth was the default mode.

Now, Doubao has launched a paid model, and large model manufacturers such as Zhipu and Yuezhianmian have raised their prices. This in itself is an admission that they can no longer bear the costs and must charge.

But has the model of burning money to achieve growth been widely rejected? Not yet. Capital is still investing, and leading manufacturers are still expanding, but the pace has slowed down, and people are starting to calculate.

For this dimension, score 1 point.

6. Capital environment: From randomly claiming valuations to starting to face pressure

In spring, financing was extremely easy, and valuations could be randomly claimed. A single PPT could secure tens of millions of dollars in investment.

OpenAI has launched an advertising platform, with an advertising revenue target of $2.5 billion this year. In 2025, the company's total revenue was $13 billion, but it had a loss of $8 billion.

Although OpenAI has a valuation of $852 billion, Anthropic, which is catching up, had an annualized revenue of over $30 billion as of April, exceeding OpenAI's $25 billion for the first time. Therefore, its valuation in the private secondary market has caught up to nearly $1 trillion, also surpassing OpenAI.

OpenAI is facing huge cost pressure, and the story of expanding and then financing no longer holds. It must make money.

Is it a financing winter? Not yet. Leading manufacturers can still raise funds, but the valuation logic has changed: in the past, it was based on the imagination space, but now it is based on the revenue ability.

For this dimension, score 1 point.

Each of the six dimensions scores 1 point, with a total score of 6 points. According to the framework, a score of 0 - 4 points indicates spring, 5 - 7 points indicates summer, 8 - 10 points indicates autumn, and 11 - 12 points indicates winter.

What is the state of summer?

Narrative and delivery coexist. There is still room for imagination, but the accounts are already on the table. Capital can still be invested, but returns are starting to be questioned. The user base is still growing, but it is starting to be segmented. Some people are willing to pay, while others only use the free version.

There are signs of autumn in some areas. If the signals such as Doubao's charging and OpenAI's advertising continue to strengthen and another 2 points are added, for example, the ROI becomes clear and capital comprehensively requires delivery, then autumn will really arrive.

Summer is a stage where narrative and delivery coexist, but delivery is becoming increasingly important. Everyone knows that the story still needs to be told, but after the story is told, the acceptance criteria are quietly changing to: what exactly have you delivered?

II. Why is it now:

3 signals, 2 drivers

The six-dimensional scoring for cycle judgment is a static analysis. From a dynamic perspective, what has happened recently?

On May 6th local time, Dario Amodei, the founder of Anthropic, said, "Our growth rate has exceeded the exponential level. In the first quarter of this year, our revenue and usage volume increased by 80 times annually, achieving explosive growth. We are providing more computing power as fast as possible at an unprecedented speed."

This is the company with the fastest revenue growth in human history. The day when AI truly becomes a productivity infrastructure may have arrived.

In the same week, two signals appeared simultaneously: Doubao started charging, and OpenAI started selling advertising.

On the surface, it seems like a coincidence, but there are two underlying lines: cost pressure and commercialization opportunities.

Signal 1: Doubao starts charging

Why does Doubao start charging? The high-frequency calls of over 300 million users have made the computing power cost a problem that must be faced.

As of March 2026, the daily average Token call volume has exceeded 120 trillion, a growth of over a thousand times compared to May 2025, and it has doubled again in the past three months.

According to the public calculation by Zheshang Securities quoted by multiple media outlets, ByteDance's capital expenditure in 2025 was approximately 160 billion yuan, of which about 90 billion yuan was used for AI computing power procurement, and the rest was used for infrastructure and network equipment.

The free model can really no longer hold on. A cost estimate circulated in the technology community shows that the hardware depreciation accounts for about 58% of a single inference, and the electricity cost accounts for about 29%. The larger the user base, the higher the cost.

Converting 120 trillion Tokens at the public API discount price, it is equivalent to having an income of 300 - 500 million yuan per day.

But what about now? The C-end income is zero.

Such a thousand-fold growth resulting in zero income is unprecedented among Chinese Internet companies in the past 15 years.

Meanwhile, the price of domestic Tokens has entered a continuous upward trend. Zhipu has raised the price of its API. The price of the GLM5.1 API has increased by 10%, and the price of the overseas version has more than doubled. Alibaba Cloud has cancelled the basic package of its Bailian platform. The free public testing of GLM5.0, MiniMax2.5, and Kimi2.5 has ended.

But there is also a downward trend: DeepSeek V4-Pro has been discounted to 25%, at $0.25 per million tokens. The price of Alibaba Cloud's Tongyi Qianwen visual understanding model has been reduced by over 80%. The input price of Doubao 2.0Lite for one million tokens is only $0.6.

The large model manufacturers are being segmented: some are raising prices, while others are lowering prices.

There are two drivers: pressure and opportunity.

Signal 2: OpenAI starts selling advertising

Why does OpenAI start selling advertising? It's half due to pressure and half due to opportunity.

On the pressure side: In 2025, the company's revenue was $13 billion, but it had a cash loss of $8 billion. With 50 million individual subscription users and 9 million enterprise users, corresponding to annual revenues of billions of dollars, but the computing power costs, R & D costs, and operating costs combined exceed this figure. The subscription revenue is not enough to cover the costs, and the accounts don't add up.

On the opportunity side: According to industry observations and estimates, the ARR (annual recurring revenue) of the advertising pilot has reached $100 million in less than two months after its launch. The advertising monetization potential of its 900 million weekly active users is huge.

According to the prediction of authoritative institutions, Meta's annual advertising revenue will