OpenAI 以模型供应商身份签下 AI 算力交易中心史上最长租约:20 年 8 吉瓦意味着什么

算力正在从一种按需购买的服务,转变为需要提前锁定的战略资产。


8 GW, 20 Years: The Lease Agreement That Turned Computing Power Into a Futures Contract

AI Infrastructure Has Entered the Era of Long-Term Computing Contracts

On August 17, 2026, OpenAI announced a landmark agreement with SB Energy, a subsidiary of SoftBank, securing approximately 8 gigawatts (GW) of IT capacity through a 20-year lease at the PORTS-Pike technology park in Ohio.

NVIDIA will provide up to $105 billion in financing guarantees and invest $1.5 billion directly into SB Energy. The project begins with 4.25 GW of capacity and has room to expand by another 3.75 GW in the future.

Built on the former site of a Cold War-era uranium enrichment facility, the data center is expected to become the world’s largest single AI data center.

So, what does 8 GW actually mean?

A typical nuclear power plant generates around 1 GW of electricity. San Francisco’s total annual electricity consumption in 2024 was roughly 5 GW. OpenAI’s newly secured computing capacity is therefore larger than the power consumption scale of an entire major city.

This agreement is not simply about renting server space.

It represents a fundamental transformation:

Computing power is moving from an on-demand service into a strategic asset that must be reserved years in advance.


The Ultimate Solution to Compute Shortages: Locking Capacity 20 Years Ahead

Why did OpenAI sign a 20-year agreement?

The answer lies in the imbalance between AI demand and infrastructure supply.

The computing requirements of frontier AI companies are growing faster than their own balance sheets can support.

In other words:

OpenAI has massive demand, but building billions of dollars worth of infrastructure independently requires long-term financing capacity, energy agreements, construction capability, and asset management expertise.

NVIDIA CEO Jensen Huang described the deal as a way to secure future capacity based on long-term customer demand — similar to supply chain planning.

NVIDIA is increasingly viewing:

  • GPUs

  • Advanced packaging

  • Memory

  • Networking

  • Land

  • Electricity

  • Data center capacity

as interconnected strategic resources.

The deeper meaning of this transaction is that computing power is moving from the spot market into the futures market.

OpenAI locks in capacity and pricing for two decades.

SB Energy receives financing certainty and a guaranteed anchor customer.

NVIDIA secures future GPU deployment opportunities.

Every participant gets strategic value.

But the consequence is clear:

Computing power is no longer just a commodity purchased whenever needed.

It is becoming a long-term asset requiring advance reservation.


Residual Value Guarantees: A New Financing Model for AI Infrastructure

The most interesting part of the agreement is not the lease itself, but the financing structure behind it.

NVIDIA’s $105 billion guarantee is not simply paying OpenAI’s rent.

It is a residual value guarantee mechanism.

If OpenAI stops leasing:

  1. SB Energy first searches for another tenant.

  2. If no replacement tenant is found, the facility can be sold.

  3. If the selling price falls below the project’s expected value, NVIDIA covers the difference.

The guarantee applies only to completed infrastructure and activates gradually as construction progresses.

The innovation here is important:

Traditional financing depends on trusting the tenant’s future repayment ability.

This model depends on trusting the long-term value of the asset itself.

The question changes from:

“Will OpenAI still exist 20 years from now?”

to:

“Will this AI infrastructure still have value even with a different tenant?”

NVIDIA is effectively using its own balance sheet to provide credit support for the emerging category of AI factories.


The Mirror of Compute Trading Centers: Centralized vs Decentralized Markets

Looking at the broader AI ecosystem, two different models are emerging.

One is the centralized infrastructure model represented by OpenAI.

One company.

One industrial park.

8 GW.

20 years.

Instead of purchasing computing power through a market, OpenAI directly locks future capacity through a strategic contract.

The other model is the emerging compute trading center.

These platforms aggregate fragmented AI supply and demand, allowing users to access multiple models through unified interfaces, settlement systems, and pricing mechanisms.

The industry is moving toward a layered computing economy:

  • Large AI laboratories secure strategic capacity through long-term contracts.

  • Developers and enterprises acquire computing resources through open markets.

The future AI economy will likely contain both:

Compute futures markets for strategic players.

Compute spot markets for everyone else.


The Rise of AI Infrastructure Financialization

One overlooked detail is that chip companies are becoming infrastructure financiers.

NVIDIA is no longer only selling GPUs.

It is becoming a participant and organizer of AI infrastructure development.

The company is helping finance data centers, securing future deployment space, and building a broader AI industrial ecosystem.

This reflects a larger trend:

AI infrastructure is no longer just hardware.

It is:

  • Energy

  • Capital

  • Computing capacity

  • Data centers

  • Distribution networks

all combined.

At the same time, platforms like Unexhub are solving another side of the market equation.

If NVIDIA’s model manages long-term “compute futures,” Unexhub focuses on optimizing the “compute spot market.”

Its API dynamic bidding mechanism allows token prices to adjust according to real-time supply and demand.

High-demand periods create higher supplier returns.

Low-demand periods create lower-cost opportunities for developers.

One manages long-term capacity planning.

The other improves short-term market efficiency.

Together, they represent two different layers of the future AI economy.


Unexhub: Building the Global Token Aggregation and Compute Trading Infrastructure

As AI moves toward a token-based economy, the biggest challenge is no longer simply creating better models.

The challenge is making AI resources easier to access, price, trade, and distribute.

Unexhub is building a global Token aggregation and compute trading center, connecting AI suppliers with developers and enterprise users through a unified marketplace.

The platform has already attracted AI model providers across different scales, including leading models such as:

  • Gemini

  • ChatGPT

  • Claude

  • DeepSeek

  • GLM

as well as high-quality AI services from startups and independent developers.

Through supplier aggregation and competitive pricing mechanisms, Unexhub enables users to access AI capabilities at significantly lower costs.

Key advantages include:

For Developers and Enterprises

  • Up to 70% cost reduction compared with standard market pricing

  • Pay only for actual token usage

  • Millisecond-level intelligent routing

  • Access multiple AI models through one unified interface

Developers no longer need to manage multiple accounts, APIs, billing systems, and payment channels.

For AI Suppliers

  • One-click integration into global distribution channels

  • Real-time settlement of usage revenue

  • More opportunities to monetize idle computing resources

Unexhub is turning fragmented AI capacity into a liquid marketplace.


The Next Stage of AI: Computing Power Becomes a Strategic Asset

The OpenAI–SB Energy agreement reveals a fundamental shift:

AI computing power is moving from:

“available whenever needed”

to:

“strategically reserved before demand arrives.”

For engineers and companies, two conclusions are becoming clear:

First:

API pricing will become increasingly market-driven.

Idle capacity will become cheaper.

Peak demand will become more expensive.

Long-term commitments will receive preferential pricing.

Second:

Most companies will not sign 20-year, multi-gigawatt compute contracts like OpenAI.

They will rely on compute trading centers to access flexible, affordable AI resources.

That is exactly where platforms like Unexhub become valuable.

The next AI competition will not only be about who builds the strongest models.

It will be about:

Who can make AI tokens flow faster, cheaper, and more transparently.

8 吉瓦,20 年:一份把算力锁成“期货合约”的租约

AI 基础设施正在进入长期算力合约时代

2026 年 8 月 17 日,OpenAI 宣布与软银旗下 SB Energy 达成战略协议,在美国俄亥俄州 PORTS-Pike 科技园区通过一份 20 年长期租约锁定约 8 吉瓦(GW)的 IT 设备容量。

根据协议,英伟达将提供最高 1050 亿美元融资担保,并向 SB Energy 投资 15 亿美元。项目初始规划容量为 4.25 吉瓦,未来还可扩展约 3.75 吉瓦。

这座建于冷战时期前铀浓缩厂旧址上的超级数据中心,未来有望成为全球最大的单体 AI 数据中心之一。

那么,8 吉瓦到底意味着什么?

一座大型核电站通常约提供 1 吉瓦电力;而旧金山整个城市 2024 年全年用电规模约为 5 吉瓦。

换句话说:

OpenAI 此次锁定的算力容量,已经超过了一座大型城市的能源规模。

但这笔交易真正重要的地方,并不是“租了一座机房”。

它代表着一个更深层的变化:

算力正在从一种按需购买的服务,转变为需要提前锁定的战略资产。


算力短缺时代的终极解法:提前 20 年锁定产能

为什么 OpenAI 要签一份长达 20 年的合同?

答案来自 AI 行业正在加剧的供需矛盾。

前沿 AI 模型训练、推理服务和智能体应用,对算力的需求增长速度已经远远超过传统基础设施建设速度。

对于 OpenAI 这样的 AI 巨头来说:

不是没有需求。

而是需求增长太快。

建设如此规模的数据中心,需要:

  • 巨额资本投入;

  • 长周期能源规划;

  • 芯片供应保障;

  • 数据中心建设能力;

  • 长期信用支持。

因此,OpenAI 选择了一种新的路径:

提前锁定未来 20 年的算力资源。

英伟达 CEO 黄仁勋曾表示,这类合作本质上类似供应链战略管理——不是简单投资,而是根据未来客户需求提前布局关键产能。

如今,英伟达关注的不再只是 GPU。

它正在重新定义 AI 基础设施:

过去:

芯片 → 服务器 → 数据中心

现在:

芯片 + 电力 + 土地 + 网络 + 数据中心容量 = AI 工厂

这笔交易的本质,是把算力从:

现货市场(Spot Market)

推向:

期货市场(Compute Futures Market)。

OpenAI 提前锁定容量。

SB Energy 获得长期稳定客户。

英伟达获得未来 GPU 部署空间。

三方形成新的 AI 基础设施生态。


残值担保:AI 基础设施融资的新模式

这笔交易最值得关注的,并不是 8 吉瓦本身,而是背后的金融结构。

英伟达提供的 1050 亿美元担保,并不是简单替 OpenAI 支付租金。

它采用的是一种:

残值担保(Residual Value Guarantee)模式。

如果未来 OpenAI 不再继续租赁:

第一步:

SB Energy 会寻找新的租户。

第二步:

如果无法找到替代客户,则出售数据中心资产。

第三步:

如果出售价格低于项目价值,英伟达补足差额。

这种模式最大的创新在于:

它改变了传统融资逻辑。

过去:

金融机构关注:

“租户未来有没有能力付款?”

现在:

金融机构关注:

“这座 AI 基础设施本身有没有长期价值?”

也就是说,信用从企业主体转移到了资产价值。

英伟达正在利用自身产业地位,为“AI 工厂”这种新型基础设施资产提供信用背书。


算力交易中心的另一面:集中式与市场化并存

如果把视角扩大,会发现未来算力市场正在形成两种模式。

模式一:战略锁定型算力

代表:

OpenAI + SB Energy

特点:

  • 单一企业;

  • 超大规模;

  • 长周期合同;

  • 提前锁定未来容量。

8 吉瓦、20 年租约,本质上就是一份:

算力期货合约。

它服务的是全球顶级 AI 实验室。


模式二:市场交易型算力

另一条路线,则是正在兴起的算力交易中心。

中国近年来正在探索:

  • 词元交易平台;

  • 算力超市;

  • 算力银行;

  • AI 模型聚合市场。

例如:

内蒙古词元交易平台于 2026 年上线,首批接入三大运营商及头部科技企业模型资源,覆盖通用大模型、推理模型、多模态模型、代码模型等领域。

中国算力平台也不断汇聚华为、阿里云、百度、腾讯云等企业算力资源。

这些平台解决的是另一类问题:

如何让分散的 AI 算力资源形成统一市场。

未来 AI 算力生态可能会形成:

大型 AI 公司:

通过长期协议锁定战略算力。

普通企业和开发者:

通过算力交易中心按需购买。

因此:

期货市场服务顶级玩家,现货市场服务广大开发者。

两者共同组成未来 AI 基础设施体系。


AI 基础设施正在金融化

这笔交易还有一个容易被忽略的变化:

芯片公司正在从硬件供应商,变成 AI 基础设施组织者。

英伟达未来提供的不只是 GPU。

它正在参与:

  • 数据中心建设;

  • 能源规划;

  • 资本融资;

  • 算力部署。

AI 基础设施正在从单纯硬件竞争,进入:

资本 + 能源 + 算力 + 模型 + 分发网络的综合竞争。

与此同时,Unexhub 正在解决 AI 市场另一端的问题:

如何让分散的模型资源和算力资源更加高效流通。

如果英伟达解决的是:

“未来 20 年的大规模算力储备。”

那么 Unexhub 解决的是:

“今天每一次 Token 调用如何更高效交易。”


Unexhub:构建全球 Token 聚合与算力交易中心

随着 AI 应用进入规模化阶段,行业竞争已经不只是模型能力竞争。

未来更重要的是:

Token 如何流通。

谁能够让 AI 能力:

  • 更容易接入;

  • 更透明定价;

  • 更低成本使用;

  • 更快速结算;

谁就掌握下一阶段 AI 基础设施入口。

Unexhub 正在打造全球 Token 聚合与分发的算力交易中心。

平台已经接入不同规模的 AI 服务供应商,包括:

  • Gemini;

  • ChatGPT;

  • Claude;

  • DeepSeek;

  • GLM;

同时也汇聚大量中小 AI 开发者提供的特色模型能力。

通过供应商聚合、动态竞价和智能路由机制,Unexhub 让企业和开发者能够更加灵活地调用全球 AI 模型资源。


开发者权益:更低成本、更简单接入、更灵活调用

对于企业和开发者而言,Unexhub 的核心价值包括:

1. Token 实付,降低 AI 使用成本

无需购买多个平台套餐。

开发者可以按照实际 Token 消耗付费。

通过供应商竞争和动态价格机制,部分模型调用成本最低可达到市场价格约 3 折水平,帮助企业降低高频 AI 使用成本。


2. 毫秒级智能路由

面对:

  • Gemini;

  • Claude;

  • ChatGPT;

  • DeepSeek;

  • GLM;

等不同模型体系。

Unexhub 通过智能路由帮助用户自动匹配:

  • 更适合的模型;

  • 更优价格;

  • 更快响应路径。

减少开发者维护多个 API 接口的复杂度。


3. AI 供应商一键接入,实时结算收益

对于模型供应商:

无需单独建设全球销售渠道。

接入 Unexhub 后,可以:

  • 快速连接企业客户;

  • 获取更多调用需求;

  • 按调用量实时结算分成。

让闲置算力和模型能力真正进入市场流通。


未来 AI 竞争:不只是模型之争,而是流通效率之争

OpenAI 与 SB Energy 的 20 年算力租约揭示了一个趋势:

未来算力不会只是:

“需要时购买”。

而会变成:

“提前规划、市场交易、动态调度。”

对于普通企业而言,并不是每家公司都需要,也无法承担一份 20 年、8 吉瓦级别的算力合同。

更多企业需要的是:

一个开放、高效、透明的算力市场。

这正是算力交易中心存在的价值。

未来 AI 竞争,不只是:

谁拥有最大的模型参数。

而是:

谁能够让 Token 流通得更快、更便宜、更透明。

从长期算力期货,到实时 Token 交易,AI 基础设施正在进入一个新的时代。

而 Unexhub 正在成为连接:

模型供应商 → 算力资源 → 企业用户 → 开发者生态

之间的重要基础设施。

AI 的下一阶段,比拼的不只是智能。

更是:

算力如何被交易、分配和释放。