谷歌 Capex 上调到 2050 亿:Token 交易市场的投资回报率能否消化 AI 军备竞赛

On July 23, Alphabet delivered a quarterly report that triggered both excitement and anxiety across the market.


The Token Marketplace May Be the Missing Liquidity Layer in Alphabet’s $205 Billion AI Bet

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 23, Alphabet delivered a quarterly report that triggered both excitement and anxiety across the market.

Total revenue reached $119.796 billion, up 24% year over year. Google Cloud revenue surged 82% to $24.768 billion, while contracted backlog climbed to $514 billion.

By conventional standards, these were remarkable results.

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, roughly double the previous year;

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

  • Free cash flow turned negative for the first time in decades, falling to negative $5.855 billion;

  • Management also indicated that spending could rise substantially again in 2027.

The market was no longer debating whether AI demand existed.

It was asking a harder question:

Can the economic value created by tokens generate returns quickly enough to support a $205 billion infrastructure bet?

The answer depends not only on how many tokens the market consumes.

It also depends on how efficiently every unit of compute can be converted into tokens, distributed to users, priced by the market, and settled as revenue.

That is why the token marketplace is becoming more than a developer convenience.

It may become the liquidity layer that determines whether the AI infrastructure cycle can sustain itself.


1. Where Is the Money Going?

Most of Alphabet’s second-quarter capital expenditure was directed toward AI infrastructure.

According to the figures cited in the source material, approximately 60% went toward servers, while the remaining 40% supported data centers, networking equipment, power capacity, and related facilities.

This spending covers far more than buying GPUs.

To operate an AI system at global scale, Alphabet must continuously invest in:

  • AI accelerators and servers;

  • Data-center construction;

  • High-speed networking;

  • Storage infrastructure;

  • Electricity procurement;

  • Cooling systems;

  • Security and reliability;

  • Model training and inference capacity.

And this is not a one-time construction cycle.

As more users adopt AI search, enterprise assistants, coding agents, document analysis, and multimodal services, the infrastructure must continue expanding.

Alphabet has reportedly raised additional financing, issued new debt, and explored further equity financing options to support this investment cycle.

The company is effectively transforming its balance sheet around one expectation:

Future token demand will be large enough to justify today’s infrastructure spending.

Management’s message has been consistent:

Demand is still exceeding investment.

That means the immediate constraint is not the number of customers willing to buy AI capacity.

The constraint is how quickly enough capacity can be built and brought online.

In some cases, even a company with Alphabet’s scale may need to rely on third-party infrastructure to supplement internal capacity.

That is a powerful signal.

AI demand is no longer waiting for supply.

Supply is trying to catch up with demand.


2. Where Is the Return?

Alphabet CEO Sundar Pichai described AI returns as being in an early stage.

At first, that may sound like a familiar explanation for heavy spending.

But the operating data suggests that returns are already beginning to appear.

Google Cloud generated $8.814 billion in operating profit during the quarter, while its operating margin expanded from 20.7% to 35.6%.

Its $514 billion backlog is approximately five times the segment’s current annualized revenue.

Meanwhile, Gemini APIs are processing approximately 22 billion tokens per minute, compared with around 16 billion in the previous quarter.

That is nearly 40% growth in one quarter.

Every minute, billions of tokens are being consumed across:

  • Developer APIs;

  • Enterprise AI systems;

  • Search experiences;

  • Coding tools;

  • Customer-service agents;

  • Document workflows;

  • Internal automation.

Token demand is clearly real.

The more difficult question is whether that demand can be monetized fast enough to absorb the depreciation, energy costs, financing expenses, and operational burden created by a capital program approaching $205 billion.

In other words, the issue is no longer whether tokens will be consumed.

It is whether token consumption can be converted into sufficiently efficient cash flow.


3. Token Demand Is Growing Nonlinearly

The source material cites data showing that 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.

This expansion is not driven only by more people opening AI chat applications.

It is driven by the increasing number of tasks performed inside each interaction.

A conventional search query may require one retrieval event.

An AI-native workflow can involve:

  1. Interpreting the request;

  2. Retrieving information;

  3. Reading long context;

  4. Calling tools;

  5. Generating an answer;

  6. Checking the result;

  7. Retrying failed steps;

  8. Executing follow-up actions.

One user request may trigger dozens of model calls.

One Agent may operate continuously for hours.

One enterprise workflow may process thousands of files every day.

As AI moves from occasional assistance into continuous production, token demand grows much faster than user counts.

This creates what some analysts describe as “token inflation.”

The phrase does not necessarily mean that token prices are rising.

It means the total volume of tokens circulating through the economy is expanding rapidly.

That growth is repricing the entire AI value chain:

  • GPUs become more valuable because they produce inference;

  • Data centers become more valuable because they host continuous token production;

  • Models become more valuable when they attract sustained usage;

  • Applications become more valuable when they generate recurring demand;

  • Routing and settlement platforms become more valuable when they improve liquidity.


4. The Token Economy Still Lacks Mature Market Infrastructure

Token consumption is growing rapidly, but the surrounding market remains fragmented.

Three fundamental questions are still unresolved:

Who sets the price?

Most model providers publish fixed price lists.

But a fixed rate does not always reflect real-time supply, demand, regional capacity, response latency, cache usage, or service guarantees.

Who matches supply with demand?

Large AI companies can distribute their own models globally.

Smaller model providers, regional compute centers, and independent suppliers may have high-quality or underused capacity but lack direct access to enterprise buyers.

Who settles the transaction?

Developers consume tokens immediately.

Suppliers may wait weeks or months to receive payment.

During that time, they must continue paying for electricity, servers, bandwidth, cloud resources, and technical operations.

The market has demand.

It has supply.

It has multiple price lists.

What it still lacks is sufficient liquidity.

Without a more efficient market structure, some infrastructure remains idle while other customers face shortages and high prices.

Some suppliers struggle to monetize capacity, while some developers pay more than necessary.

This mismatch is precisely where a token marketplace can create value.


5. Turning “Dead Compute” Into “Live Tokens”

Compute capacity is perishable.

If a GPU cluster sits idle for an hour, that unused capacity cannot be stored and sold tomorrow.

The electricity, equipment depreciation, and operating expenses have already occurred.

In this sense, unused compute resembles an empty airline seat or unsold electricity generation.

Once the moment passes, the economic opportunity disappears.

A token marketplace can help convert this “dead compute” into “live tokens.”

During periods of low demand, available capacity can be offered at more competitive prices.

Developers can move flexible workloads—such as batch processing, testing, document analysis, or content generation—to lower-cost periods.

Suppliers gain revenue from infrastructure that might otherwise remain idle.

During peak demand, scarce capacity can be directed toward higher-priority workloads, with suppliers receiving stronger returns.

The objective is not simply to lower prices.

It is to increase utilization.

The higher the utilization rate of AI infrastructure, the faster capital expenditure can be converted into revenue-producing token output.

That makes liquidity directly relevant to Alphabet’s $205 billion investment question.


6. Unexhub: Building a Market Layer for Token Liquidity

Unexhub positions itself as a token aggregation and distribution compute marketplace.

Its purpose is not merely to place another API gateway between model providers and developers.

Its broader goal is to create a market layer where fragmented AI capacity can be:

  • Aggregated;

  • Compared;

  • Priced;

  • Routed;

  • Purchased;

  • Settled.

On the supply side, Unexhub brings together providers of different sizes, including services associated with mainstream domestic and international models such as Gemini, ChatGPT, Claude, DeepSeek, and GLM, as well as models and compute capacity provided by smaller companies and independent suppliers.

On the demand side, enterprises and developers can access multiple model resources through one interface instead of maintaining:

  • Multiple API formats;

  • Separate authentication systems;

  • Fragmented balances;

  • Different billing rules;

  • Repeated operational integrations.

The platform attempts to solve the Token economy’s three central problems:

Price discovery, supply-demand matching, and settlement.

Its model is built around real-time settlement, early-stage zero commission, market-based API bidding, and millisecond-level routing.


7. Real-Time Settlement: Converting Usage Into Cash Flow

For small and medium-sized AI suppliers, long settlement cycles can create severe financial pressure.

A supplier may deliver inference capacity today but wait one or two months before receiving payment.

During that period, it must continue covering:

  • Electricity;

  • Hardware depreciation;

  • Network bandwidth;

  • Cloud-resource costs;

  • Operations;

  • Technical staff.

Usage may increase while available cash continues to shrink.

Unexhub’s real-time settlement mechanism is designed to shorten the distance between token delivery and supplier revenue.

Its principle is straightforward:

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

This turns delayed accounts receivable into reusable working capital.

The practical benefits include:

  • Faster recovery of operating costs;

  • Lower financing pressure;

  • More predictable cash flow;

  • Greater ability to expand capacity;

  • Less dependence on long billing cycles.

The phrase “supply what is available and settle what has been consumed” is not merely a payment feature.

It creates a faster cash-flow loop between compute production and infrastructure reinvestment.

For smaller suppliers, that loop can determine whether growth remains financially sustainable.


8. Zero Commission for Early Suppliers

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

This reduces the cost of testing a new distribution channel.

A smaller model company or independent compute provider can connect its resources, observe real market demand, and begin generating usage revenue without immediately surrendering a portion of every transaction to platform fees.

The benefit includes:

  • Lower market-entry costs;

  • Better protection for thin margins;

  • More flexibility when testing prices;

  • Faster access to developer and enterprise demand;

  • A shorter route from idle capacity to revenue.

This is a classic two-sided marketplace strategy.

The platform first needs sufficient supply to offer buyers choice, stability, and competitive pricing.

More suppliers attract more demand.

More demand makes the platform more valuable to suppliers.

Early zero-commission access helps accelerate that network effect.


9. API Bidding: Allowing the Market to Discover the Price

Most model APIs still rely on fixed price lists.

A provider publishes a rate, and the developer either accepts it or chooses another service.

But compute capacity changes constantly.

During off-peak periods, large amounts of inference capacity may remain unused.

During peak periods, the same capacity may become scarce.

A fixed price cannot accurately reflect both situations.

Unexhub’s API bidding model allows pricing to respond dynamically to supply and demand.

When capacity is abundant

  • Suppliers can lower prices to attract workloads;

  • Developers can run flexible tasks more cheaply;

  • Idle infrastructure can generate revenue;

  • Batch tasks can move to lower-cost periods.

When demand rises

  • High-priority workloads can secure capacity;

  • Suppliers can earn stronger returns;

  • Scarce resources can be allocated more efficiently.

The market can distinguish between:

  • Peak and off-peak usage;

  • Real-time and delay-tolerant tasks;

  • General-purpose and specialized models;

  • Standard and guaranteed service;

  • Low-cost and premium-capacity endpoints.

For developers, this provides access to lower prices when timing is flexible.

For suppliers, it improves utilization during quiet periods and increases potential returns when demand is strong.

This is how tokens begin moving from provider-controlled billing units toward market-priced digital resources.


10. Millisecond-Level Routing: Matching Every Task With Suitable Capacity

A token marketplace cannot operate efficiently if every model remains isolated behind a separate API.

Unexhub therefore uses a unified access layer and millisecond-level routing to connect requests with suitable model capacity.

Routing can consider:

  • Model capability;

  • Current token price;

  • Available capacity;

  • Response latency;

  • Task type;

  • Regional availability;

  • Service stability.

A simple classification request does not need to use the most expensive frontier model.

A complex coding or reasoning workflow may require a stronger endpoint.

A flexible batch task can be routed toward lower-cost capacity, while a real-time enterprise request may prioritize latency and availability.

When a provider becomes expensive, congested, or temporarily unavailable, the routing layer can redirect suitable workloads behind the same interface.

Developers do not need to rebuild the entire integration.

This reduces:

  • Repeated API development;

  • Manual model switching;

  • Account fragmentation;

  • Operational maintenance;

  • Billing complexity.

For selected models and usage plans, users may obtain prices as low as approximately 30% of the corresponding official or prevailing market rate.

The value is especially visible in high-volume workloads.

A small discount on ten calls is limited.

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

  • AI search;

  • Coding Agents;

  • Enterprise knowledge bases;

  • Customer-service automation;

  • Document processing;

  • Batch content generation;

  • Multi-Agent workflows.

The objective is not merely to make one request cheaper.

It is to make sustained, high-intensity AI usage economically manageable.


11. Why Liquidity Is the Real Answer to the $205 Billion Bet

Alphabet’s $205 billion capital expenditure plan is ultimately based on one expectation:

Token demand will continue growing faster than the infrastructure depreciates.

Whether the bet succeeds depends on two variables.

Variable One: Will Token demand continue growing?

The evidence is already substantial:

  • Gemini serves hundreds of millions of users;

  • More than nine million developers access its models through APIs;

  • API throughput has reached 22 billion tokens per minute;

  • Google Cloud backlog has expanded to $514 billion;

  • Enterprise adoption continues to rise.

Demand appears real and increasingly recurring.

Variable Two: Can Token circulation become efficient enough?

This is the less discussed variable.

Even if demand is large, returns can be weakened by:

  • Idle infrastructure;

  • Fragmented supply;

  • Slow settlement;

  • Inflexible pricing;

  • Poor model routing;

  • Unpredictable availability;

  • Repeated integration costs.

Liquidity improves the connection between capital expenditure and revenue.

It helps every available unit of compute find a buyer more quickly.

It helps buyers find more suitable capacity at a reasonable price.

It shortens the time between inference delivery and supplier payment.

That is why a token marketplace matters.

It does not create demand by itself.

It makes existing demand easier to monetize.


12. From Early AI Returns to Real-Time Cash-Flow Loops

Pichai is correct that AI returns are still in an early stage.

But “early” does not mean “absent.”

The cloud business is already generating stronger revenue, higher margins, and larger contracted commitments.

The next step is improving the speed and efficiency with which infrastructure investment becomes cash flow.

A functioning token marketplace can create a tighter loop:

  1. Suppliers provide available compute;

  2. The platform converts compute into accessible model capacity;

  3. Routing matches demand with suitable resources;

  4. Developers consume tokens;

  5. Usage is measured and priced;

  6. Supplier revenue is settled;

  7. Capital can be reinvested into additional capacity.

The shorter and more efficient this loop becomes, the easier it is for the AI infrastructure economy to sustain continuous expansion.

Token liquidity therefore affects more than API pricing.

It affects capital efficiency.


13. Unexhub’s Value Is Broader Than Low-Cost Tokens

Unexhub’s value proposition is not limited to offering cheaper API access.

Lower prices may attract users, but infrastructure quality determines whether they can continue operating at scale.

For developers, the platform can provide:

  • Unified access to multiple mainstream models;

  • Usage-based Token billing;

  • Millisecond-level routing;

  • More flexible model selection;

  • Lower integration and maintenance costs;

  • Eligible resources priced as low as roughly 30% of prevailing rates.

For suppliers, it can provide:

  • Access to a broader demand channel;

  • Real-time usage settlement;

  • Early-stage zero-commission incentives;

  • Dynamic pricing opportunities;

  • A route for monetizing idle capacity.

For enterprise customers, it can support:

  • Model selection;

  • API adaptation;

  • Agent-workflow integration;

  • Concurrency management;

  • Unified billing;

  • Cost monitoring;

  • Stability guarantees;

  • Ongoing technical services.

The deeper product is not simply an API.

It is the infrastructure that allows global model capacity to circulate more efficiently.


14. My View: Liquidity Will Decide the Outcome

Alphabet’s major AI indicators now form a coherent picture:

  • 950 million Gemini users;

  • More than nine million API developers;

  • 22 billion tokens processed per minute;

  • $24.768 billion in quarterly cloud revenue;

  • $514 billion in contracted backlog;

  • Up to $205 billion in planned annual capital expenditure.

These numbers do not describe a small experiment.

They describe the reconstruction of a production system.

Tokens are becoming a new industrial input for the AI era.

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

But a production input needs liquidity.

Without efficient routing, pricing, distribution, and settlement, even enormous demand can coexist with idle capacity, high costs, and slow cash recovery.

Unexhub is attempting to provide that liquidity by:

  • Aggregating AI suppliers of different sizes;

  • Connecting resources associated with Gemini, ChatGPT, Claude, DeepSeek, GLM, and other models;

  • Offering unified API access;

  • Routing requests in milliseconds;

  • Supporting market-based bidding;

  • Offering qualifying capacity at prices as low as approximately 30% of prevailing rates;

  • Settling supplier revenue in real time;

  • Supporting customized enterprise integration.

The $205 billion AI infrastructure bet will not be won by capital expenditure alone.

It will be won when every unit of compute can be converted into tokens, every token can find suitable demand, and every transaction can return cash to the supply side efficiently.

Alphabet has pushed its chips to the center of the table.

The next question is whether Token liquidity can keep pace.


Token 交易市场,或许是 2050 亿美元算力赌局缺少的流动性解药

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 23 日,Alphabet 交出了一份让市场同时感到兴奋与焦虑的季度财报。

公司总营收达到 1197.96 亿美元,同比增长 24%;Google Cloud 营收同比大涨 82%,达到 247.68 亿美元;已签约但尚待未来履行的积压订单,则进一步增长至 5140 亿美元。

按照传统指标衡量,这是一份相当亮眼的成绩单。

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;

  • 自由现金流数十年来首次转负,降至 负 58.55 亿美元;

  • 管理层还表示,2027 年的相关支出可能继续大幅增长。

市场现在已经不再怀疑 AI 是否存在需求。

它在追问一个更困难的问题:

Token 所创造的经济价值,能否足够快地形成回报,支撑这场最高达 2050 亿美元的基础设施投资?

答案并不只取决于市场最终会消耗多少 Token。

还取决于每一份算力能否被快速转化为 Token,被分配给用户,被市场合理定价,并最终结算为现金收入。

因此,Token 交易市场已经不只是一项方便开发者接入模型的服务。

它可能成为决定这一轮 AI 基础设施周期能否持续运转的流动性层。


一、钱究竟烧到哪里去了?

Alphabet 第二季度的大部分资本开支,都投向了 AI 技术基础设施。

按照原文援引的数据,大约 60% 用于服务器,其余 40% 则投向数据中心、网络设备、电力容量及相关设施。

这些投入远不只是购买 GPU。

要在全球范围内运行 AI 服务,Alphabet 还需要持续建设:

  • AI 加速器和服务器;

  • 数据中心;

  • 高速网络;

  • 存储基础设施;

  • 电力供应;

  • 冷却系统;

  • 安全与可靠性体系;

  • 模型训练与推理容量。

而且,这不是一次性的建设周期。

随着更多用户开始使用 AI 搜索、企业助手、编程 Agent、文档分析和多模态服务,底层基础设施必须持续扩张。

Alphabet 据称已经通过额外融资、发行债券,以及筹备新的股权融资方式,为这一轮投资提供支持。

这相当于把资产负债表押在一个核心判断上:

未来 Token 需求创造的价值,足以覆盖今天的大规模基础设施投入。

管理层释放出的信息也非常明确:

市场需求仍然超过公司的投资能力。

这意味着,眼下的限制因素并不是没有客户愿意购买 AI 算力。

而是基础设施能否足够快地建设并投入运行。

在部分情况下,即使是 Alphabet 这样规模的公司,也可能需要通过第三方基础设施补充自有容量。

这个信号非常清楚:

AI 需求并没有在等待供给。

供给正在努力追赶需求。


二、回报到底在哪里?

Alphabet CEO 桑达尔·皮查伊表示,AI 回报仍处于早期阶段。

这句话听起来或许像是对巨额投入的常见解释,但经营数据确实表明,回报已经开始出现。

Google Cloud 本季度运营利润达到 88.14 亿美元,运营利润率由 20.7% 提高至 35.6%。

其 5140 亿美元积压订单,约为当前云业务年化收入的五倍。

Gemini API 每分钟处理约 220 亿 Token,而上一季度还是约 160 亿,单季度增长近 40%。

每一分钟,数百亿 Token 都在以下场景中被真实消耗:

  • 开发者 API;

  • 企业 AI 系统;

  • AI 搜索;

  • 编程工具;

  • 智能客服;

  • 文档工作流;

  • 内部自动化。

Token 需求显然是真实的。

更困难的问题是,这些需求能否足够快地完成商业变现,覆盖接近 2050 亿美元资本开支所带来的折旧、电力、融资和运营成本。

换句话说,问题已经不再是 Token 会不会被消耗。

而是 Token 的消耗,能否被高效转化为现金流。


三、Token 需求正在非线性增长

原文援引的数据显示,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,调用规模增长超过一千倍。

推动这轮增长的,并不只是越来越多人打开 AI 聊天应用。

更重要的是,每次交互内部包含的任务数量越来越多。

一次传统搜索,可能只需要完成一次检索。

一条 AI 原生工作流,却可能需要:

  1. 理解问题;

  2. 检索资料;

  3. 读取长上下文;

  4. 调用工具;

  5. 生成答案;

  6. 检查结果;

  7. 重试失败步骤;

  8. 执行后续动作。

一次请求,可能触发数十次模型调用。

一个 Agent,可能连续运行数小时。

一条企业工作流,每天可能处理数千份文件。

当 AI 从偶尔辅助转向持续生产,Token 需求的增长速度就会远远超过用户数量的增长。

一些分析机构将这种现象称为“Token 通胀”。

这里的通胀未必是价格上涨,而是 Token 流通总量快速膨胀。

这种规模增长正在重新评估整条 AI 产业链:

  • GPU 因为能够持续生产推理能力而被重估;

  • 数据中心因为承载 Token 生产而被重估;

  • 模型因为真实调用量而被重估;

  • 应用因为能够创造持续 Token 需求而被重估;

  • 路由、交易与结算平台则因为改善流动性而被重估。


四、Token 经济还缺少成熟的市场基础设施

Token 消耗快速增长,但围绕它的市场仍然高度分散。

三个基础问题尚未得到充分解决。

谁来定价?

多数模型供应商仍然使用固定价目表。

但固定价格很难实时反映市场需求、区域容量、响应速度、缓存情况和服务保障差异。

谁来撮合供需?

大型 AI 企业可以建设自己的全球销售与分发体系。

中小模型厂商、区域算力中心和独立供应商可能拥有优质或闲置容量,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,但供应商可能需要等待数周甚至数月才能收到款项。

在此期间,电费、服务器、带宽、云资源和技术运维成本仍然需要持续支付。

市场并不缺需求。

也不缺算力供给。

更不缺不同厂商的价目表。

它真正缺少的是流动性。

如果没有更加高效的市场结构,一部分算力可能长期闲置,另一部分用户却面临容量紧张和价格过高。

一些供应商难以将容量变现,一些开发者则不得不承担更高的调用成本。

这正是 Token 交易市场可以创造价值的地方。


五、把“死算力”变成“活 Token”

算力具有明显的时效性。

如果一组 GPU 在某个小时处于闲置状态,这部分容量无法保存到明天再出售。

电费、设备折旧和运维成本已经发生。

从这个角度看,闲置算力很像起飞后的空座位,也像无法储存的即时电力。

使用窗口一旦过去,商业机会便随之消失。

Token 交易市场可以帮助把这类“死算力”转化为“活 Token”。

在需求较低的时段,供应商可以用更具竞争力的价格释放剩余容量。

开发者则可以将批处理、测试、文档分析和内容生产等具有时间弹性的任务,安排到成本更低的时段。

供应商原本可能闲置的基础设施,因此获得收入。

当需求进入高峰,稀缺容量则可以优先分配给价值更高、时效要求更强的任务,供应商也能获得更加合理的回报。

其核心目标并不只是降价。

而是提高算力利用率。

AI 基础设施的利用率越高,资本开支就越有可能更快转化为能够产生收入的 Token。

这让 Token 流动性与 2050 亿美元资本开支的回报问题,产生了直接联系。


六、Unexhub:为 Token 流动建设市场层

Unexhub 将自身定位为一个 Token 聚合与分发算力交易中心。

它的目标并不是简单地在模型供应商与开发者之间增加一道 API 网关。

更大的愿景,是建立一个市场层,让分散的 AI 产能可以被:

  • 聚合;

  • 比较;

  • 定价;

  • 路由;

  • 采购;

  • 结算。

在供给侧,Unexhub 汇聚不同规模的服务商,包括与 Gemini、ChatGPT、Claude、DeepSeek、GLM 等海内外主流模型相关的资源,也包括中小企业与独立供应商提供的模型和算力能力。

在需求侧,企业与开发者可以通过统一接口调用不同模型,而不必分别维护:

  • 多种 API 格式;

  • 多套鉴权方式;

  • 分散的账户余额;

  • 不同的计费标准;

  • 重复的运维体系。

Unexhub 希望解决 Token 经济的三个核心问题:

发现价格、匹配供需和完成结算。

其主要机制包括实时结算、早期零佣金、API 市场化竞价和毫秒级路由。


七、实时结算:将 Token 消耗转化为现金流

对中小 AI 供应商而言,漫长的结算周期可能造成明显的资金压力。

供应商今天已经交付推理算力,收入却可能需要等待一两个月才能到账。

在此期间,仍然需要支付:

  • 电费;

  • 硬件折旧;

  • 网络带宽;

  • 云资源成本;

  • 日常运维;

  • 技术人员费用。

调用量可能不断增加,可用现金却持续减少。

Unexhub 的实时结算机制,希望缩短 Token 交付与供应商收入到账之间的距离。

其逻辑非常直接:

Token 产能一旦交付并被真实消耗,相应的供应商分成就应尽快进入结算。

这能够将漫长账期中的应收收入,更快转化为可周转资金。

实际价值包括:

  • 更快回收运营成本;

  • 缓解融资和现金流压力;

  • 提高收入可预测性;

  • 获得更强的扩容能力;

  • 降低对长期账期的依赖。

“供多少,结算多少”不只是一项付款功能。

它能够在算力生产与基础设施再投资之间,形成更快的现金循环。

对于中小供应商而言,这种循环甚至可能决定高速增长能否持续。


八、早期供应商零佣金权益

在平台市场启动阶段,符合条件的早期供应商可以享受 零佣金入驻权益。

这能降低测试新分发渠道的成本。

中小模型企业或独立算力供应商可以先接入资源、观察真实市场需求并获得调用收入,不必从第一笔交易开始,就向平台让出部分收益。

其价值包括:

  • 降低进入市场的成本;

  • 为较薄的利润空间提供保护;

  • 获得更灵活的价格测试空间;

  • 更快触达开发者与企业需求;

  • 缩短闲置容量转化为收入的路径。

这是一种典型的双边市场启动策略。

平台首先需要吸引足够丰富的供给,才能为买方提供更多选择、更稳定的容量和更有竞争力的价格。

供应商越多,需求越容易被吸引。

需求越多,平台对供应商的渠道价值也越高。

早期零佣金有助于加速这一网络效应。


九、API 动态竞价:让市场发现 Token 价格

大多数模型 API 目前仍然使用固定价目表。

供应商公布价格,开发者选择是否接受。

但算力容量始终在发生变化。

低谷时段,大量推理资源可能处于闲置状态。

到了高峰时段,同样的容量又可能迅速变得紧张。

固定价格很难准确反映两种完全不同的状态。

Unexhub 的 API 竞价机制,让价格能够根据供需动态变化。

当算力充足时

  • 供应商可以降低价格吸引任务;

  • 开发者可以更低成本运行弹性工作;

  • 闲置基础设施能够产生收入;

  • 批处理任务可以转移至价格更低的时段。

当需求上升时

  • 高优先级任务可以锁定容量;

  • 供应商可以获得更高收益;

  • 稀缺资源得到更加高效的配置。

市场还可以进一步区分:

  • 高峰与低谷使用;

  • 实时任务与可延迟任务;

  • 通用模型与垂直模型;

  • 普通服务与保障型服务;

  • 低成本端点与高性能端点。

对于开发者来说,任务具有时间弹性时,可以获得更低价格。

对于供应商来说,则可以提高低谷期利用率,并在需求旺盛时获取更合理的回报。

Token 也由此开始从模型厂商单方面定价的计费单位,转向由市场动态发现价格的数字资源。


十、毫秒级路由:为每项任务匹配合适的容量

如果每个模型都被封闭在独立 API 后面,Token 交易市场就无法高效运行。

Unexhub 因此通过统一接入层和毫秒级路由,将请求与更加合适的模型容量连接起来。

路由过程可以综合考虑:

  • 模型能力;

  • 当前 Token 价格;

  • 可用容量;

  • 响应延迟;

  • 任务类型;

  • 区域可用性;

  • 服务稳定性。

一个简单分类任务,没有必要始终使用价格最高的旗舰模型。

复杂编程或推理工作流,则可能需要能力更强的端点。

具有时间弹性的批量任务,可以被分配给成本更低的资源;对响应速度要求更高的企业请求,则可以优先选择低延迟和高可用服务。

当某个供应商价格上涨、容量拥堵或暂时不可用时,路由层可以在统一接口后方完成资源切换。

开发者不必重新搭建整套系统。

这能够减少:

  • 重复 API 开发;

  • 手动切换模型;

  • 账户与余额分散;

  • 日常运维成本;

  • 账单管理复杂度。

在部分符合条件的模型和调用方案中,用户最低可获得约为相应官方标准价格或市场价格 30% 的综合调用成本。

这一权益在高调用量场景中尤其明显。

十次调用的折扣价值可能有限。

但当业务持续消耗数百万、数亿甚至数十亿 Token 时,更低的单价可以显著降低以下场景的运营成本:

  • AI 搜索;

  • 编程 Agent;

  • 企业知识库;

  • 智能客服;

  • 文档处理;

  • 批量内容生成;

  • 多智能体工作流。

平台的目标并不只是让某一次请求更便宜。

而是让长期、高强度的 AI 使用,在成本上真正可持续。


十一、为什么流动性才是 2050 亿赌局的核心答案?

Alphabet 最高达 2050 亿美元的资本开支计划,本质上押注的是一件事:

Token 需求的增长速度,会持续超过基础设施折旧和资金成本。

这场投资能否获得回报,取决于两个变量。

第一个变量:Token 需求能否继续增长?

目前已经出现了相当明确的证据:

  • Gemini 拥有数亿级用户;

  • 每月有超过 900 万开发者通过 API 使用模型;

  • API 每分钟处理 220 亿 Token;

  • Google Cloud 积压订单达到 5140 亿美元;

  • 企业采用速度持续上升。

Token 需求是真实的,而且越来越具有持续性。

第二个变量:Token 流通效率能否足够高?

这是更容易被忽视的变量。

即使市场需求庞大,以下因素仍可能拖累资本回报:

  • 算力处于闲置状态;

  • 供给过于分散;

  • 结算周期过长;

  • 定价缺少弹性;

  • 模型路由效率不足;

  • 服务可用性不可预测;

  • 企业重复承担接入成本。

流动性可以加强资本开支与收入之间的连接。

它让每一份可用算力更快找到买方。

也让买方以更合理的价格找到合适资源。

同时,它还可以缩短算力交付与供应商回款之间的时间。

这正是 Token 交易市场的重要性。

它本身并不凭空创造需求。

但它能让已经存在的需求更容易完成变现。


十二、从“早期回报”走向“实时现金流循环”

皮查伊所说的 AI 回报处于早期阶段,并不等于目前没有回报。

Google Cloud 已经实现了更高收入、更强利润率和更大的合同储备。

下一步,是提高基础设施投资转化为现金流的速度和效率。

一套成熟的 Token 交易市场,可以形成更加紧密的循环:

  1. 供应商提供可用算力;

  2. 平台将算力转化为可调用的模型资源;

  3. 路由系统为需求匹配合适容量;

  4. 开发者消耗 Token;

  5. 使用量被准确计量和定价;

  6. 供应商收入进入结算;

  7. 资金继续投入新的算力建设。

这一循环越短、越高效,AI 基础设施经济就越容易支撑持续扩张。

因此,Token 流动性影响的不只是 API 价格。

它影响的是整个行业的资本效率。


十三、Unexhub 的价值不只是低价 Token

Unexhub 的价值并不只是提供更便宜的 API 调用。

更低价格可以吸引用户,但基础设施质量决定用户能否持续扩大业务。

对开发者而言,平台可以提供:

  • 多个主流模型的统一接入;

  • 按实际 Token 用量付费;

  • 毫秒级路由;

  • 更灵活的模型选择;

  • 更低的接入与维护成本;

  • 符合条件的资源最低约为市场价格的 30%。

对供应商而言,平台可以提供:

  • 更广泛的需求渠道;

  • 实时调用分成结算;

  • 早期零佣金权益;

  • 动态定价机会;

  • 闲置容量的变现路径。

对企业客户而言,平台还可以支持:

  • 模型选择;

  • API 接口适配;

  • Agent 工作流集成;

  • 并发管理;

  • 统一账单;

  • 成本监控;

  • 稳定性保障;

  • 持续技术服务。

平台真正提供的,并不只是一个 API。

而是让全球模型与算力资源流动得更加高效的基础设施。


十四、我的判断:流动性将决定这场赌局的结果

Alphabet 当前几项关键 AI 数据,已经构成了一幅清晰的图景:

  • 9.5 亿 Gemini 用户;

  • 超过 900 万 API 开发者;

  • 每分钟处理 220 亿 Token;

  • Google Cloud 单季度收入 247.68 亿美元;

  • 积压订单达到 5140 亿美元;

  • 全年资本开支最高达到 2050 亿美元。

这些数字描述的并不是一场小规模试验。

而是一套生产体系的彻底重构。

Token 正在成为 AI 时代的新型生产资料。

算力负责生产 Token;
模型负责塑造 Token;
应用负责消耗 Token;
企业则为 Token 创造的结果付费。

但生产资料必须拥有流动性。

如果缺少高效的路由、定价、分发与结算,即使需求十分庞大,也可能同时出现算力闲置、成本过高和资金回收缓慢。

Unexhub 正在尝试提供这层流动性:

  • 聚合不同规模的 AI 供应商;

  • 连接 Gemini、ChatGPT、Claude、DeepSeek、GLM 等主流模型资源;

  • 提供统一 API 接入;

  • 以毫秒级速度完成请求路由;

  • 支持市场化竞价;

  • 让符合条件的容量最低达到约市场价格的 30%;

  • 实时结算供应商调用收入;

  • 为企业提供定制化系统接入服务。

2050 亿美元的 AI 基础设施赌局,不会仅靠资本开支取胜。

它需要每一份算力都能够被转化为 Token,每一个 Token 都能找到合适的需求,每一笔交易都能高效地把现金重新送回供给侧。

Alphabet 已经把筹码推到了桌面中央。

接下来,就要看 Token 的流动性能否跟上。