自由现金流转负 59 亿,Token 交易市场如何回答 Alphabet 的 AI 投入回报之问

Total revenue reached $119.796 billion, up 24% year over year and marking the company’s 12th consecutive quarter of double-digit growth.


Negative $5.9 Billion in Free Cash Flow: Can a Token Marketplace Answer Alphabet’s AI ROI Question?

Bilingual Article|English first, followed by the complete Chinese edition.
双语文章|英文全文在前,完整中文版请继续向下阅读。

On July 22, Alphabet released a quarterly report that appeared impressive on the surface but left the market with a much harder question underneath.

Total revenue reached $119.796 billion, up 24% year over year and marking the company’s 12th consecutive quarter of double-digit growth.

Google Cloud was even more striking. Revenue surged 82% to $24.768 billion, while operating profit rose from $2.826 billion to $8.814 billion.

Gemini reached approximately 950 million monthly active users. Its APIs processed around 22 billion tokens per minute, and Google Cloud’s contracted backlog climbed above $500 billion, reaching approximately $514 billion.

Yet Alphabet’s shares fell nearly 5% in after-hours trading.

The reason was another set of numbers:

  • Second-quarter capital expenditure reached $44.924 billion, nearly double the level of a year earlier;

  • Full-year capital expenditure guidance increased from $180–190 billion to $195–205 billion;

  • Free cash flow fell to negative $5.855 billion;

  • Management indicated that AI-related spending could rise substantially again in 2027.

According to the figures cited in the source material, this was Alphabet’s first negative free-cash-flow quarter since becoming a public company.

The market was no longer asking whether AI demand existed.

The question had changed:

Can AI-generated revenue return quickly enough to justify the enormous amount of capital being invested in compute infrastructure?

At the center of that question is one increasingly important economic unit:

The token.

Tokens connect data centers, GPUs, models, applications, developers, and enterprise customers. But to turn rapidly growing token consumption into sustainable cash flow, the industry needs more than powerful models.

It needs a more efficient mechanism for pricing, routing, distributing, and settling token capacity.


1. Where Is the Money Going—and Where Is the Return?

Most of Alphabet’s $44.924 billion in quarterly capital expenditure was directed toward AI infrastructure.

That includes far more than simply purchasing GPUs. A global AI platform requires continuous investment in:

  • AI accelerators and servers;

  • Data-center construction;

  • High-speed networking;

  • Storage systems;

  • Power capacity;

  • Cooling infrastructure;

  • Model training;

  • Real-time inference;

  • Security and service reliability.

According to the financial information cited in the original material, Alphabet has also expanded its financing capacity through debt issuance and other capital-market arrangements.

The company is effectively making one of the largest infrastructure bets in corporate history.

The assumption behind that bet is clear:

Future AI demand will be large and persistent enough to absorb the infrastructure being built today.

Alphabet CFO Anat Ashkenazi reportedly summarized the supply-demand relationship directly:

Market demand continues to exceed our investment.

In other words, the immediate problem is not a lack of customers.

The problem is that infrastructure construction cannot yet fully keep pace with demand.

But heavy demand does not automatically guarantee attractive returns.

Data centers must be built before they can process requests. GPUs must be purchased before they can generate tokens. Electricity, networking, maintenance, and financing costs begin long before all contracted revenue is recognized.

This creates a timing mismatch:

Capital goes out immediately, while token-generated revenue returns gradually.

That mismatch is one of the main reasons free cash flow can turn negative even while revenue and operating profit continue to rise.


2. The Return Is Early, but It Is Already Visible

Alphabet CEO Sundar Pichai has said that returns from AI remain in an early stage.

That statement may sound cautious, but the operating figures show that monetization has already begun.

Google Cloud generated $8.814 billion in quarterly operating profit, and its operating margin expanded from 20.7% to 35.6%.

Its contracted backlog reached approximately $514 billion, equivalent to roughly five times the segment’s current annualized revenue.

Gemini API throughput rose from around 16 billion tokens per minute in the previous quarter to 22 billion per minute, an increase of nearly 40% in three months.

Nearly 90% of Fortune 100 companies were reported to be using Gemini Enterprise.

These indicators tell a consistent story:

  • Enterprise adoption is growing;

  • Token consumption is increasing;

  • Cloud margins are improving;

  • Customers are committing future budgets;

  • AI is entering recurring production workflows.

The demand side is no longer theoretical.

The unresolved issue is speed:

Can token consumption be converted into recognized revenue and cash flow quickly enough to absorb annual capital expenditure approaching $205 billion?

This is not merely a question of model intelligence.

It is a question of infrastructure utilization and token liquidity.


3. Negative Free Cash Flow Reflects a Monetization Timing Gap

Alphabet’s negative free cash flow does not necessarily mean AI is failing to generate revenue.

It more accurately reflects a mismatch between two cycles.

The infrastructure-investment cycle

Alphabet must first finance and build:

  • Data centers;

  • Compute clusters;

  • Networking capacity;

  • Storage systems;

  • Energy infrastructure;

  • Model-serving systems.

These costs occur before the capacity reaches full utilization.

The token-monetization cycle

After the infrastructure becomes available, customers must:

  • Integrate models;

  • Launch applications;

  • Migrate workloads;

  • Increase usage;

  • Consume tokens;

  • Pay invoices;

  • Allow contracted revenue to be recognized.

The second cycle naturally lags behind the first.

Google Cloud’s $514 billion backlog illustrates this gap.

The contracts exist, but much of the revenue has not yet been recognized because the related services will be delivered over time.

At the economic level, customers have reserved future AI capacity.

At the financial level, the associated cash and revenue are still moving through the delivery cycle.

That is why the market is so focused on liquidity.

Demand may be locked in, but the investment cannot earn its return until compute is converted into actual token consumption.


4. Token Demand Is Growing at a Nonlinear Rate

According to the industry data cited in the source text, average daily token usage in China rose from approximately 100 billion at the beginning of 2024 to around 140 trillion by March 2026.

That represents growth of more than one thousand times in roughly two years.

The increase is not driven only by more people opening AI applications.

It is also driven by the growing number of model calls inside each task.

A traditional software request may trigger one operation.

An AI-native workflow may require the system to:

  1. Interpret the objective;

  2. Retrieve information;

  3. Read long context;

  4. Call external tools;

  5. Generate a result;

  6. Validate the output;

  7. Retry failed steps;

  8. Continue with follow-up actions.

A single enterprise task may involve dozens of model calls.

An Agent may run for hours.

A customer-service system may process thousands of conversations every day.

As AI moves from occasional assistance into continuous business operations, token consumption can grow much faster than the number of users.

This is why tokens are increasingly viewed as a new production input.

Compute produces them.
Models shape them.
Applications consume them.
Businesses pay for the outcomes they enable.


5. From “Dead Compute” to “Live Tokens”

AI infrastructure has one important economic characteristic:

Unused capacity is perishable.

If a GPU cluster remains idle for an hour, that hour of capacity cannot be stored and sold the following week.

Electricity has already been consumed. Hardware continues to depreciate. Data-center and staffing costs remain.

Unused compute resembles an empty airline seat or unsold electricity generation: once the time window passes, the revenue opportunity disappears.

This is where a Token marketplace can improve capital efficiency.

During off-peak periods, spare capacity can be offered at more competitive prices.

Developers can move flexible workloads—such as testing, batch processing, content generation, and document analysis—into lower-cost windows.

Suppliers receive income from infrastructure that might otherwise remain idle.

During peak periods, scarce capacity can be allocated to higher-priority workloads, allowing suppliers to earn stronger returns.

The objective is not simply to reduce API prices.

It is to increase utilization.

The more continuously compute infrastructure generates billable tokens, the faster heavy capital expenditure can be converted into revenue-producing assets.


6. The Token Economy Still Lacks Mature Market Infrastructure

Token demand is expanding, but the surrounding market remains fragmented.

Three fundamental questions have yet to be answered efficiently.

Who sets the price?

Most model providers still publish fixed price lists.

But fixed prices do not always reflect real-time capacity, demand, latency, cache utilization, regional supply, or service guarantees.

Who matches supply with demand?

Large model providers can build their own global distribution networks.

Smaller AI companies, regional compute centers, and independent suppliers may have valuable capacity but lack direct access to enterprise customers.

Who settles the transaction?

Developers consume tokens immediately, while suppliers may wait weeks or months to receive payment.

During that time, suppliers must continue covering:

  • Electricity;

  • Hardware depreciation;

  • Bandwidth;

  • Cloud infrastructure;

  • Operations;

  • Technical staffing.

The market has supply and demand, but the connection between them remains inefficient.

Some capacity sits idle while other customers face shortages.

Some developers pay premium rates while lower-cost supply remains difficult to discover.

Some suppliers generate usage but wait too long for cash to return.

The missing element is liquidity.


7. The Rise of the Token Distribution Layer

A new intermediary layer is emerging between upstream model and compute suppliers and downstream developers and enterprises.

Its role is not merely to resell API access.

It improves the efficiency of the Token supply chain by providing:

  • Unified integration across multiple models;

  • Centralized billing;

  • Model comparison;

  • Intelligent routing;

  • Bulk purchasing;

  • Cost optimization;

  • Supplier distribution;

  • Usage settlement.

For developers, this layer reduces the need to repeatedly integrate different providers.

For enterprises, it simplifies account management, billing, model selection, and service governance.

For suppliers, it creates access to demand without requiring each provider to build an independent global sales and settlement network.

Unexhub is built around this logic.

Its ambition is not to become merely another API gateway, but to build a Token aggregation and distribution compute marketplace—a market layer where AI capacity can be discovered, compared, routed, priced, and settled.


8. Unexhub: Connecting Fragmented Token Supply and Demand

Unexhub aggregates AI resources from suppliers of different sizes.

These include services associated with mainstream domestic and international models such as Gemini, ChatGPT, Claude, DeepSeek, and GLM, as well as model and compute resources supplied by smaller companies and independent providers.

Developers and enterprises can access multiple models through a unified interface instead of separately maintaining:

  • Different API formats;

  • Multiple authentication systems;

  • Fragmented account balances;

  • Separate pricing rules;

  • Repeated technical integrations;

  • Multiple invoices.

The platform can route requests according to:

  • Model capability;

  • Current price;

  • Available capacity;

  • Response latency;

  • Task type;

  • Geographic region;

  • Service stability.

This turns multiple isolated model endpoints into a more usable market.

The commercial value lies not only in providing access.

It lies in making token supply easier to compare, purchase, and redirect as conditions change.


9. Real-Time Settlement: Shortening the Path From Compute to Cash

For small and medium-sized suppliers, settlement speed can determine whether growth is sustainable.

A supplier may deliver inference capacity today but wait one or two months before receiving its share of the revenue.

Meanwhile, operating expenses continue.

Unexhub’s real-time settlement mechanism is designed to reduce this delay.

Its principle is straightforward:

Once token capacity has been delivered and consumed, the corresponding supplier revenue should enter settlement promptly.

This does not eliminate every form of working-capital pressure, but it can significantly shorten the cash-conversion cycle.

For suppliers, the practical benefits include:

  • Faster recovery of operating costs;

  • Lower cash-flow pressure;

  • More predictable income;

  • Greater ability to maintain and expand capacity;

  • Less dependence on lengthy accounts-receivable cycles.

The idea of “settling what has already been supplied and consumed” turns delayed revenue into reusable working capital.

That creates a tighter connection between Token production and infrastructure reinvestment.


10. Dynamic API Pricing: Letting Supply and Demand Discover Value

Fixed API prices cannot fully reflect constantly changing compute conditions.

During periods of low demand, large amounts of inference capacity may remain unused.

During peak periods, the same resources can become scarce.

Unexhub’s API bidding model allows eligible suppliers to adjust pricing according to current supply and demand.

When capacity is abundant

  • Suppliers can offer lower prices;

  • Developers can run flexible workloads more economically;

  • Idle infrastructure can generate revenue;

  • Batch tasks can move to off-peak periods.

When demand is strong

  • High-priority workloads can secure capacity;

  • Suppliers can earn stronger returns;

  • Scarce resources can be allocated more efficiently.

This allows the market to distinguish between:

  • Peak and off-peak use;

  • Real-time and delay-tolerant workloads;

  • General-purpose and specialized models;

  • Standard and guaranteed service levels;

  • Low-cost and premium-performance endpoints.

Developers gain more control over costs.

Suppliers gain more flexibility in monetizing both idle and scarce capacity.

Tokens begin to move from provider-controlled billing units toward market-priced digital resources.


11. Early-Supplier Zero Commission: Reducing Market-Entry Friction

During the marketplace’s early development stage, qualifying suppliers may receive a zero-commission participation benefit.

This lowers the cost of testing a new distribution channel.

A smaller model provider or independent compute supplier can connect capacity, observe real demand, and begin generating usage revenue without immediately giving up a share of every transaction.

The value includes:

  • Lower entry costs;

  • Better protection for thin margins;

  • More freedom to test pricing;

  • Faster access to developers and enterprise customers;

  • A shorter path from idle capacity to revenue.

This is a standard but important strategy for launching a two-sided market.

More suppliers give buyers greater choice and stronger price competition.

More buyers make the platform more valuable to suppliers.

Early zero-commission access helps accelerate this network effect.


12. Millisecond-Level Routing and More Flexible Token Costs

A mature Token marketplace must be able to match each task with an appropriate model.

A basic classification request does not need the most expensive frontier model.

A complex coding or reasoning workflow may require a more capable endpoint.

A flexible batch task may prioritize price, while a customer-facing application may prioritize latency and availability.

Unexhub uses a unified interface and millisecond-level routing to allocate requests according to such conditions.

If one provider becomes expensive, congested, or temporarily unavailable, suitable workloads can be redirected without requiring the developer to rebuild the application.

This helps reduce:

  • Repeated integration work;

  • Manual model switching;

  • Account fragmentation;

  • Maintenance costs;

  • Billing complexity.

For selected models and usage plans, developers may access resources at prices as low as approximately 30% of the corresponding official or prevailing market rate.

The benefit becomes more meaningful as usage grows.

A discount on a few API calls may be minor.

Across millions or billions of tokens, it can materially reduce the cost of:

  • AI search;

  • Coding Agents;

  • Enterprise knowledge bases;

  • Customer-service automation;

  • Document analysis;

  • Batch content generation;

  • Multi-Agent workflows.

The objective is not merely to provide a one-time low price.

It is to make sustained, high-volume AI use more economically manageable.


13. Turning Backlog Into Consumption, and Consumption Into Cash

Google Cloud’s $514 billion backlog is a monetary expression of future demand.

Customers have signed contracts, but the associated revenue will be recognized as services are delivered.

This means the capacity may be commercially reserved before all related cash and accounting revenue have returned.

A more efficient Token market can help shorten several parts of that cycle:

  1. Available compute is exposed to more buyers;

  2. Routing improves infrastructure utilization;

  3. Flexible pricing accelerates demand matching;

  4. Developers consume the resulting tokens;

  5. Usage is measured and billed;

  6. Suppliers receive revenue more quickly;

  7. Funds can be reinvested into capacity.

The shorter this cycle becomes, the more efficiently infrastructure investment can produce cash flow.

A Token marketplace does not remove the need for data centers or GPUs.

It helps those assets work harder.


14. What Unexhub Is Really Monetizing

Unexhub is not simply monetizing API access.

Its deeper product is market efficiency.

Supply aggregation

Fragmented model and compute resources become easier to discover.

Demand distribution

Suppliers gain access to enterprise and developer demand without independently building a global sales channel.

Price discovery

Dynamic bidding allows prices to respond to capacity and demand.

Intelligent routing

Requests can be assigned according to cost, capability, latency, and availability.

Transaction settlement

Usage can be measured and supplier revenue settled more quickly.

Enterprise services

Organizations can receive customized support for:

  • Model selection;

  • API adaptation;

  • Agent integration;

  • Concurrency requirements;

  • Unified billing;

  • Cost monitoring;

  • Service stability;

  • Ongoing technical operations.

The platform’s value is therefore broader than “cheap tokens.”

It is an attempt to make global AI capacity flow more efficiently between suppliers and users.


15. My View: The Real Question Is Whether Revenue Can Catch Up With Spending

Alphabet’s negative $5.855 billion in free cash flow reflects a temporary but consequential mismatch:

The company is spending on AI infrastructure faster than that infrastructure is returning cash.

That does not mean Token demand is absent.

The $514 billion backlog, 950 million Gemini users, 22 billion tokens processed per minute, and widespread enterprise adoption all point to substantial demand.

The question is not whether AI can generate revenue.

It is whether the speed of monetization can catch up with the speed of investment.

A Token marketplace addresses this second question.

By increasing utilization, improving price discovery, connecting fragmented supply with demand, and shortening settlement cycles, it can help turn compute capacity into billable Token output more efficiently.

When tokens are consumed like electricity, routed like network traffic, and traded according to market conditions, AI infrastructure gains a more measurable path toward return on investment.

Pichai may be correct that AI returns are still early.

But “early” does not mean nonexistent.

A functioning Token marketplace can help convert early infrastructure investment into a faster, more continuous cash-flow cycle.

Alphabet has already pushed its capital onto the table.

The industry’s next challenge is ensuring that every available unit of compute can find demand, every consumed token can be priced efficiently, and every completed transaction can return cash to the supply side.

Only then can negative free cash flow begin to look less like a warning—and more like a transitional cost of rebuilding the world’s AI production system.


自由现金流负 59 亿美元:Token 交易市场如何回答 Alphabet 的 AI 投资回报之问

中文完整版|上方为英文全文,以下为完整中文内容。
Chinese Edition|The complete English version appears above.

7 月 22 日,Alphabet 交出了一份表面亮眼、底层却让市场更加焦虑的季度财报。

公司总营收达到 1197.96 亿美元,同比增长 24%,连续第 12 个季度实现双位数增长。

Google Cloud 的表现更为突出:营收同比大涨 82%,达到 247.68 亿美元;运营利润从 28.26 亿美元跃升至 88.14 亿美元。

Gemini 应用月活跃用户达到约 9.5 亿,API 每分钟处理约 220 亿 Token;Google Cloud 已签约但尚待未来履行的积压订单,也首次突破 5000 亿美元,达到约 5140 亿美元。

但 Alphabet 股价却在盘后下跌近 5%。

原因来自另一组数字:

  • 第二季度资本开支达到 449.24 亿美元,同比接近翻倍;

  • 全年资本开支指引从 1800 亿至 1900 亿美元,上调至 1950 亿至 2050 亿美元;

  • 自由现金流降至 负 58.55 亿美元;

  • 管理层还表示,2027 年相关开支可能继续大幅增长。

按照原文引用的数据,这是 Alphabet 上市以来首次出现季度自由现金流转负。

市场已经不再追问 AI 是否存在需求。

问题变成了:

AI 创造收入的速度,能不能赶上巨额算力投资花钱的速度?

而连接数据中心、GPU、模型、应用、开发者与企业客户的核心计量单位,正是 Token。

要把不断增长的 Token 消耗转化为持续现金流,行业需要的不只是更强的模型。

还需要更加高效的定价、路由、分发与结算机制。


一、钱烧到哪里去了,回报又在哪里?

Alphabet 第二季度 449.24 亿美元的资本开支,大部分投向了 AI 基础设施。

这并不只是购买 GPU。

一套面向全球运行的 AI 系统,需要持续投入:

  • AI 加速器与服务器;

  • 数据中心建设;

  • 高速网络;

  • 存储系统;

  • 电力容量;

  • 冷却设施;

  • 模型训练;

  • 实时推理;

  • 安全与服务稳定性。

按照原文所引财务信息,为了支撑这轮基础设施建设,Alphabet 也在通过债券及其他资本市场工具扩大融资能力。

这家公司正在进行企业史上规模最大的一轮基础设施投入之一。

这场投资建立在一个清晰判断之上:

未来持续增长的 AI 需求,足以消化今天建设的算力基础设施。

Alphabet CFO 阿什肯纳齐在财报沟通中释放的信息也很明确:

市场需求仍然超过公司的投资能力。

换句话说,当前的问题并不是缺少愿意购买 AI 服务的客户。

而是基础设施建设速度,暂时还跟不上需求增长。

但需求旺盛,并不意味着资本回报会自动出现。

数据中心必须先建,GPU 必须先买,电力、网络、运维和融资成本会提前发生;客户调用量、账单收入和合同确认,则要在服务交付后逐渐形成。

这便产生了一个时间差:

资本开支即时流出,Token 收入逐步回流。

这也是为什么营收和运营利润仍在增长,自由现金流却可能转负。


二、AI 回报处于早期,但并不等于没有回报

Alphabet CEO 桑达尔·皮查伊表示,AI 的投资回报仍处于早期阶段。

这句话听起来相对谨慎,但经营数据说明,商业回报已经开始出现。

Google Cloud 本季度运营利润达到 88.14 亿美元,运营利润率从 20.7% 上升至 35.6%。

其积压订单达到约 5140 亿美元,相当于当前云业务年化收入的五倍左右。

Gemini API 每分钟处理的 Token 数量,则从上一季度约 160 亿增长至 220 亿,单季度增幅接近 40%。

接近 90% 的《财富》100 强企业,据称已经开始使用 Gemini Enterprise。

这些数据共同说明:

  • 企业采用正在加速;

  • Token 调用持续增加;

  • 云业务利润率正在改善;

  • 客户正在提前锁定预算;

  • AI 正从试验工具进入长期生产流程。

需求侧已经不再停留在理论层面。

真正尚未解决的问题是速度:

Token 消耗能否足够快地转化为收入和现金流,从而覆盖最高接近 2050 亿美元的年度资本开支?

这已经不只是模型能力问题。

它也是基础设施利用率和 Token 流动性问题。


三、负自由现金流反映的是变现周期错配

Alphabet 自由现金流转负,并不必然代表 AI 无法创造收入。

更准确地说,它反映了两个周期之间的错配。

算力投资周期

Alphabet 必须先投入资金建设:

  • 数据中心;

  • GPU 集群;

  • 网络容量;

  • 存储系统;

  • 能源设施;

  • 模型服务系统。

这些成本会在基础设施充分投入使用前发生。

Token 变现周期

基础设施上线后,客户还需要:

  • 接入模型;

  • 上线应用;

  • 迁移业务;

  • 扩大使用量;

  • 实际消耗 Token;

  • 支付服务费用;

  • 等待合同收入逐步确认。

第二个周期天然落后于第一个周期。

Google Cloud 的 5140 亿美元积压订单,正好说明了这种差距。

合同已经签署,但因为服务需要在未来逐步交付,大量收入尚未确认。

从经济层面看,客户已经预订未来算力。

从财务层面看,对应现金流仍在交付周期中。

所以,市场真正关心的是流动性。

需求已经被锁定,但只有算力被转化为实际 Token 消耗,这笔投资才能形成回报。


四、Token 需求正在非线性增长

根据原文援引的行业数据,国内日均 Token 调用量从 2024 年初约 1000 亿,增长至 2026 年 3 月约 140 万亿。

两年左右,增长超过一千倍。

推动增长的,并不只是更多用户开始打开 AI 应用。

更重要的是,每项任务内部的模型调用次数不断增加。

一条传统软件请求,可能只对应一次操作。

一条 AI 原生工作流,却可能需要系统:

  1. 理解目标;

  2. 检索资料;

  3. 阅读长上下文;

  4. 调用外部工具;

  5. 生成结果;

  6. 验证输出;

  7. 重试失败步骤;

  8. 继续完成后续任务。

一项企业任务可能触发数十次模型调用。

一个 Agent 可能持续运行数小时。

一套智能客服系统每天可能处理数千次对话。

当 AI 从偶尔辅助进入持续生产,Token 消耗的增长速度便会超过用户增长速度。

这也是 Token 开始成为新型生产资料的原因。

算力生产 Token;
模型塑造 Token;
应用消耗 Token;
企业为 Token 创造的业务结果付费。


五、把“死算力”转化为“活 Token”

AI 算力有一个非常重要的经济特征:

闲置容量具有时效性。

如果一组 GPU 在某个小时没有工作,这个小时的推理能力无法保存到下周再出售。

电费已经发生,硬件仍在折旧,数据中心与人员成本也不会暂停。

闲置算力很像飞机起飞后的空座位,也像没有及时被消耗的即时电力:使用窗口一旦过去,收入机会便随之消失。

Token 交易市场可以帮助把这类“死算力”转化为“活 Token”。

在低谷时段,剩余容量可以用更具竞争力的价格释放。

开发者可以将测试、批处理、内容生成和文档分析等具有时间弹性的任务,安排到低成本窗口。

供应商则可以从原本可能闲置的基础设施中获得收入。

进入高峰时段后,稀缺容量可以优先分配给高价值、高时效任务,供应商也能获得更高回报。

其核心目标并不只是降低 API 价格。

更重要的是提高算力利用率。

AI 基础设施越能持续生产可计费的 Token,巨额资本开支就越快转化为能够产生收入的资产。


六、Token 经济仍然缺少成熟的市场基础设施

Token 需求快速增长,但围绕它的市场仍然高度分散。

三个基础问题尚未得到高效解决。

谁来定价?

多数模型供应商仍然使用固定价目表。

但固定价格无法始终反映实时容量、市场需求、延迟水平、缓存使用、区域供给和服务保障。

谁来撮合供需?

大型模型企业可以建设全球分发网络。

中小 AI 公司、区域算力中心和独立供应商,可能拥有优质算力,却缺少直接触达企业客户的渠道。

谁来完成结算?

开发者可以即时消耗 Token,供应商却可能需要等待数周甚至数月才能收到款项。

在此期间,电费、硬件折旧、带宽、云资源和技术运维仍然需要持续投入。

市场并不缺少供给和需求。

它真正缺少的是高效率的连接方式。

一部分算力可能处于闲置状态,另一部分客户却在承受容量短缺和高价格。

一些开发者支付较高费用,较低成本的供给却难以被发现。

部分供应商已经产生调用收入,现金却迟迟无法回流。

缺失的核心要素,就是流动性。


七、Token 分销中间层正在形成

在上游模型与算力供应商和下游开发者、企业客户之间,一个新的中间层正在出现。

它的作用并不只是转售 API,而是改善 Token 供应链效率,包括:

  • 统一接入多个模型;

  • 集中管理账单;

  • 比较不同模型;

  • 智能路由请求;

  • 通过规模采购优化成本;

  • 为供应商分发需求;

  • 按实际调用完成结算。

对开发者而言,这一层减少了重复接入不同供应商的工作。

对企业而言,它简化了账户、账单、模型选择和服务治理。

对供应商而言,它提供了新的需求渠道,不必每家都独立建设全球销售与结算系统。

Unexhub 正是建立在这套逻辑之上。

其目标并不是成为另一道普通 API 网关,而是建设一个 Token 聚合与分发算力交易中心,让 AI 产能能够被发现、比较、路由、定价与结算。


八、Unexhub:连接分散的 Token 供需

Unexhub 聚合不同规模的 AI 供应商。

其中既包括与 Gemini、ChatGPT、Claude、DeepSeek、GLM 等海内外主流模型相关的服务,也包括中小企业与独立供应商提供的模型和算力资源。

企业与开发者可以通过统一接口使用多个模型,不必分别维护:

  • 不同的 API 格式;

  • 多套鉴权体系;

  • 分散的账户余额;

  • 各异的计费标准;

  • 重复的技术集成;

  • 多张独立账单。

平台还可以根据以下条件分配请求:

  • 模型能力;

  • 当前价格;

  • 可用容量;

  • 响应延迟;

  • 任务类型;

  • 服务区域;

  • 稳定性要求。

这能够将彼此孤立的模型端点,组合成一个更容易使用的市场。

平台的商业价值并不只来自“能够调用模型”。

更在于让 Token 供给变得更容易比较、购买,并能随着条件变化灵活切换。


九、即时结算:缩短从算力到现金的路径

对中小供应商而言,结算速度可能直接决定业务能否持续扩张。

供应商今天已经交付推理能力,却可能需要等待一两个月才能获得调用分成。

但运营费用仍在持续产生。

Unexhub 的实时结算机制,希望缩短这一时间差。

其核心原则是:

Token 产能一旦交付并被实际消耗,相应的供应商收入就应尽快进入结算。

它无法消除所有资金压力,却可以显著缩短现金转换周期。

对供应商而言,具体价值包括:

  • 更快回收运营投入;

  • 降低现金流压力;

  • 提高收入可预测性;

  • 增强维护与扩容能力;

  • 减少对漫长应收账期的依赖。

“已经供给、已经消耗的部分及时结算”,能够把延迟收入更快转化为可周转资金。

这也会在 Token 生产与基础设施再投资之间,建立更紧密的循环。


十、动态 API 定价:让供需共同发现价值

固定 API 价格很难完整反映不断变化的算力市场。

低谷时段,大量推理容量可能处于闲置状态。

高峰时段,同样的资源又可能迅速变得稀缺。

Unexhub 的 API 竞价模式,让符合条件的供应商可以根据供需情况动态调整价格。

当容量充足时

  • 供应商可以降低价格;

  • 开发者可以更经济地执行弹性任务;

  • 闲置基础设施能够产生收入;

  • 批处理任务可以转移至低谷时段。

当需求旺盛时

  • 高优先级任务可以锁定容量;

  • 供应商可以获得更高收益;

  • 稀缺算力得到更合理的配置。

市场还能进一步区分:

  • 高峰与低谷调用;

  • 实时与可延迟任务;

  • 通用与专业模型;

  • 普通与保障型服务;

  • 低成本与高性能端点。

开发者因此获得更强的成本控制能力。

供应商也能更灵活地变现闲置容量与稀缺容量。

Token 开始从模型厂商单方面定价的计费单位,走向由市场供需发现价格的数字资源。


十一、早期供应商零佣金:降低入场摩擦

在平台建设初期,符合条件的早期供应商可享受 零佣金入驻权益。

这可以降低测试新分发渠道的成本。

中小模型供应商和独立算力提供者,可以先接入产能、观察真实需求并获得调用收入,而不必从第一笔交易开始就向平台让出部分收益。

其具体价值包括:

  • 降低市场进入成本;

  • 保护较薄的利润空间;

  • 提供更灵活的价格测试条件;

  • 更快接触企业和开发者需求;

  • 缩短闲置容量转化为收入的路径。

这是双边市场常见但重要的启动机制。

供应商越多,买方获得的选择与价格竞争越充分。

买方越多,平台对供应商的渠道价值也越高。

零佣金权益有助于加快这一网络效应。


十二、毫秒级路由与更灵活的 Token 成本

成熟的 Token 市场,需要为不同任务匹配适合的模型。

简单分类任务没有必要调用价格最高的旗舰模型。

复杂编程和推理任务,则可能需要更强的端点。

具有时间弹性的批处理任务可以优先考虑价格;面向客户的实时应用,则更重视延迟与可用性。

Unexhub 通过统一接口和毫秒级路由,根据这些条件分配请求。

当某个供应商价格上升、容量拥堵或暂时不可用时,合适的任务可以被切换至其他资源,而不需要开发者重建应用。

这能够减少:

  • 重复集成;

  • 手动切换模型;

  • 账户分散;

  • 运维成本;

  • 账单复杂度。

在部分符合条件的模型与调用方案中,开发者最低可以获得约为相应官方标准价或市场价格 30% 的综合调用成本。

这一权益会随着使用规模扩大而更加明显。

少量调用节省的费用可能有限。

但当业务持续消耗数百万、数亿甚至数十亿 Token 时,更低的单位成本可以显著优化以下场景的预算:

  • AI 搜索;

  • 编程 Agent;

  • 企业知识库;

  • 智能客服;

  • 文档分析;

  • 批量内容生产;

  • 多智能体工作流。

平台追求的不是一次性的最低价。

而是让长期、高频 AI 使用拥有更加可持续的成本结构。


十三、把积压订单转化为调用,再把调用转化为现金

Google Cloud 的 5140 亿美元积压订单,是未来需求的货币化表达。

客户已经签署合同,但相关收入仍需随着服务交付逐步确认。

这意味着,算力可能已经被商业合同锁定,但现金和收入还没有全部回流。

更加高效的 Token 市场,可以帮助缩短其中多个环节:

  1. 可用算力被更多买方发现;

  2. 路由提高基础设施利用率;

  3. 灵活价格加快供需匹配;

  4. 开发者实际消耗 Token;

  5. 使用量被计量并形成账单;

  6. 供应商更快获得收入;

  7. 资金继续投入新产能。

这个循环越短,基础设施投资转化为现金流的效率就越高。

Token 交易市场不会取代数据中心和 GPU。

它的作用,是让这些重资产运行得更加充分。


十四、Unexhub 真正在变现什么?

Unexhub 所变现的并不只是 API 接入能力。

更深层的产品,是市场效率。

聚合供给

让分散的模型与算力资源更容易被发现。

分发需求

让供应商不必独立建立全球销售渠道,也能触达企业和开发者。

发现价格

通过动态竞价,让价格响应容量与需求。

智能路由

根据成本、能力、延迟和可用性分配请求。

交易结算

准确计量用量,并更及时地向供应商结算收入。

企业服务

企业还可以获得围绕以下需求的定制支持:

  • 模型选择;

  • API 接口适配;

  • Agent 工作流接入;

  • 并发要求;

  • 统一账单;

  • 成本监控;

  • 稳定性保障;

  • 持续技术运维。

因此,平台的价值远不止“更便宜的 Token”。

它试图让全球 AI 产能在供应商与用户之间流动得更加高效。


十五、我的判断:真正的问题,是赚钱速度能否赶上花钱速度

Alphabet 负 58.55 亿美元的自由现金流,反映了一个暂时却十分关键的错配:

公司投资 AI 基础设施的速度,暂时快于基础设施回收现金的速度。

这并不意味着 Token 没有需求。

5140 亿美元积压订单、9.5 亿 Gemini 用户、每分钟 220 亿 Token 的处理量,以及企业采用速度,都证明了需求的规模。

问题不是 AI 能不能赚钱。

而是赚钱速度能否赶上投资速度。

Token 交易市场所回答的,正是第二个问题。

通过提高算力利用率、改善价格发现、连接分散供需并缩短结算周期,它可以帮助算力更高效地转化为可计费的 Token。

当 Token 像电力一样被消耗,像网络流量一样被路由,又根据市场条件进行交易时,AI 基础设施便拥有了一条更加可计算的投资回报路径。

皮查伊所说的“回报仍处于早期阶段”,并不等于回报不存在。

一套成熟的 Token 交易市场,可以帮助把早期重资产投入,转化为更快、更连续的现金流循环。

Alphabet 已经把巨额资本推到了桌面中央。

接下来,整个行业必须保证:

每一份可用算力都能找到需求;
每一个被消耗的 Token 都能得到合理定价;
每一笔完成的交易都能高效地把现金送回供给侧。

只有到了那时,负 58.55 亿美元的自由现金流,才可能不再被视为一声警报,而被重新理解为重建全球 AI 生产体系过程中必须经历的过渡成本。