BitTorrent ventures into AI computing power: BTTInferGrid builds a decentralized AI inference computing power network

With AI Agent being applied in various complex scenarios such as enterprise workflows, automated production, and autonomous execution, the global AI industry has officially transitioned from "passive response" to a new stage of "autonomous execution." The core of industry competition has long shifted from merely comparing large model parameters to competing in execution capabilities, with strong logical reasoning abilities being the fundamental basis supporting this transformation.

The paradigm shift in application scenarios has also driven a fundamental change in the demand for upstream computing power infrastructure: the focus of computing power consumption is continuously shifting from model training to business reasoning, and this trend is irreversible. However, the current mainstream centralized computing power system has exposed issues such as high operational costs, weak elastic scaling, and insufficient service stability when faced with massive, high-frequency, and volatile reasoning requests, leading the entire AI industry to encounter developmental bottlenecks in computing power supply.

On June 17, the established decentralized transmission ecosystem BitTorrent launched a strategic product—BTTInferGrid, targeting the AI reasoning track and building a decentralized computing power network. This platform leverages a decentralized distributed architecture to efficiently aggregate scattered idle GPU computing resources from around the world, breaking down the connection barriers between the resource supply side and AI developers, providing open, easily accessible, verifiable on-chain computation results, and flexible pay-as-you-go AI reasoning computing power services.

Relying on the advantages of decentralized technology, BTTInferGrid not only fills the gaps of traditional centralized computing power in high concurrency and load fluctuation scenarios but also achieves a leap in computing power supply, reconstructing the resource allocation and circulation logic of the entire computing power ecosystem.

At the same time, BTTInferGrid is a strategic product upgraded from BitTorrent's existing BTFS service, marking BitTorrent's key extension of its long-standing decentralized resource scheduling capabilities from the storage track to the computing power field, and is a crucial move in its layout of the decentralized AI track.

The structure of computing power demand is shifting from "training" to "reasoning": BTTInferGrid aims to reconstruct the AI reasoning computing power supply in a decentralized manner

BTTInferGrid hopes to leverage a decentralized model to reconstruct the computing power supply system, addressing issues such as high costs and supply shortages in AI reasoning computing power, while enhancing the efficiency of large model reasoning, thus providing the industry with high-performance, high-resilience, and cost-effective computing power infrastructure.

If 2024 to 2025 is the "thousand model war" and the parameter arms race dominated by tens of thousands of clusters in the AI industry, then in 2026, with the large-scale implementation of AI Agents, AI will officially enter the "reasoning era" of explosive large-scale applications. AI reasoning is the key link in realizing the value of models, transforming "trained models" into practical applications, commercial value, and daily services. In simple terms, training is "teaching AI to learn," while reasoning is "enabling AI to be used"—for example, an autonomous vehicle recognizing a stop sign on an unfamiliar road is a typical reasoning behavior. Reasoning ability directly influences the user experience, operational costs, and commercial value of AI products.

There is a general consensus in the industry that more than 70% of computing power resources will be used for reasoning scenarios in the future. Oracle has predicted that the market size for reasoning computing power will eventually surpass that of training computing power. Zheng Weimin, an academician of the Chinese Academy of Engineering, also pointed out that the vast majority of computing power is currently consumed in daily interactions between users and large models. From a cost composition perspective, human labor accounts for only 3% of the expenses in large model reasoning, data accounts for 2%, while computing power accounts for as much as 95%; the computing costs of leading applications are quite substantial, with ChatGPT's daily reasoning costs around $700,000, and DeepSeek V3 reaching $87,000.

As the demand for AI computing power shifts from centralized training by a few tech giants to commercial reasoning scenarios for millions of developers across various industries, the evaluation criteria for underlying infrastructure have also changed. In the training era, developers primarily focused on the scale and efficiency of centralized computing power; entering the reasoning era, AI services are directly facing massive end-users, with daily interactions in the hundreds of billions generating enormous computing power consumption, shifting developers' focus to the cost per call, response speed, and service stability. Today, computing power supply, call costs, and service availability have become the core criteria for evaluating AI infrastructure and are key to determining whether AI applications can be successfully implemented.

However, in the face of exponentially rising reasoning demands, the shortcomings of mainstream centralized computing power systems are becoming increasingly prominent: GPU rental prices continue to rise, platform services frequently experience outages, and many AI applications are forced to shut down due to computing power costs. These issues are concentrated in the following three aspects:

• First, insufficient elasticity in computing power scheduling cannot cope with traffic peaks and valleys, leading to a dilemma of cost and stability imbalance: Although leading AI companies and cloud vendors continue to increase their investments in computing power facilities, the growth of reasoning demand is rapid and exhibits significant peak and valley characteristics—request volumes can surge dozens of times during daytime office or marketing peaks; at night, they can plummet sharply. Centralized data centers lack the elastic scheduling capabilities to adapt to these dynamic changes: if configured for peak demand, depreciation costs during low periods are high; if configured for average demand, services may be interrupted during peak periods, falling into the dilemma of "high cost" and "low stability." Meanwhile, centralized computing power must also layer costs such as data center construction, electricity, operations, and commercial profits, ultimately leading to high computing power costs that severely compress the trial-and-error space for small and medium-sized innovative teams, creating an urgent need for new solutions that combine cost advantages with elastic scheduling capabilities.

• Second, rising GPU rental prices hinder innovation for small and medium-sized enterprises and developers: While open-source large models (such as Qwen, DeepSeek, etc.) have lowered the entry barriers in the AI field, the deployment and operation of models still rely on stable, inexpensive, and easily accessible reasoning computing power. However, the reality is that GPU rental costs are continuously rising; for example, the hourly rental price of the mainstream H100 graphics card has increased from $1.70 in October 2025 to $2.35 in March 2026, a nearly 40% increase in half a year. The high costs deter many individual developers and small enterprises with quality solutions, trapping them in a "having models but no computing power" dilemma, severely stifling innovation vitality and scalable development in the AI industry.

• Third, a large amount of idle GPU resources globally are not being effectively utilized, leading to a serious mismatch between supply and demand: In stark contrast to the market's "computing power shortage," there is a vast amount of idle high-performance GPU computing resources globally, scattered across personal devices, university laboratories, small data centers, and facilities left over from cryptocurrency transitions. Due to a lack of standardized access channels and efficient scheduling engines, this computing power cannot enter the mainstream reasoning market, creating a contradictory situation where demand side experiences "difficulty in obtaining cards" while the supply side has "sleeping computing power," with significant room for improvement in resource utilization and an urgent need to resolve the mismatch between supply and demand.

In summary, the current AI reasoning computing power market is facing three structural dilemmas: on one hand, centralized supply cannot balance cost and elasticity; on the other hand, rising computing power rental prices suppress AI innovation; and on the third hand, there are vast amounts of idle GPU resources that have long remained dormant. In the face of this series of industry challenges, BTTInferGrid, relying on decentralized technology, brings a new solution to address the mismatch between computing power supply and demand.

BTTInferGrid aims to connect globally dispersed idle GPU resources with a massive number of AI developers efficiently through a decentralized approach, fundamentally breaking the monopoly and bottlenecks of centralized computing power. On one hand, the platform integrates scattered idle GPU computing power to build an open and shared computing power infrastructure; on the other hand, it opens up connection channels between the supply side and the demand side, eliminating the entry barriers and pricing black boxes of traditional centralized models. At the same time, relying on the incentives and collaborative mechanisms of DePIN, BTTInferGrid can continuously provide high-cost performance reasoning computing power, fundamentally resolving the core pain points of high computing power costs and supply shortages, truly unleashing the reasoning efficiency and commercial value of large models.

BTTInferGrid: Building a decentralized computing power network for AI reasoning scenarios, with three major advantages redefining the computing power allocation mechanism

BTTInferGrid has a clear and defined positioning, focusing on building a decentralized computing power network for AI reasoning scenarios, connecting global idle GPU computing power supply with AI reasoning market demand, and providing an open access, verifiable results, and pay-as-you-go global AI computing power service system.

Specifically, BTTInferGrid relies on the underlying network mechanism of DePIN to accurately match computing power supply with the explosive growth of AI reasoning demand, achieving bidirectional value empowerment on both supply and demand sides:

• Supply Side: Efficiently aggregates fragmented idle GPU resources globally, building an open and shared computing power base. At the same time, leveraging DePIN's incentives and intelligent scheduling mechanisms, it opens up low-threshold, sustainable monetization channels for computing power holders, turning the globally idle "sleeping GPUs" into "liquid assets"; on the other hand, it ensures stable and elastic scaling of computing power, creating a high-cost performance, highly scalable, and secure global reasoning service capability.

• Demand Side: For global AI developers, it provides easy access, verifiable on-chain results, and pay-as-you-go global reasoning services. Compared to the high premium pricing of centralized cloud vendors, BTTInferGrid possesses extreme cost advantages and elastic scaling capabilities, helping small and medium-sized innovative teams and independent developers reduce business trial-and-error costs, efficiently complete product validation and business iteration, while also empowering the upstream computing power supply ecosystem.

Thus, BTTInferGrid not only effectively addresses the urgent demand for low-cost, high-elasticity computing power from AI developers during the "application battle" phase but also opens up a sustainable value realization channel for the vast amounts of idle hardware resources globally.

More importantly, the BTTInferGrid platform will successfully build a self-sustaining positive growth flywheel: idle GPU nodes continuously expand, reasoning computing power costs continuously decrease, attracting more developers to join; market demand continues to rise, further incentivizing global computing power suppliers to join the ecosystem. BTTInferGrid reconstructs computing power supply through a decentralized model, transforming scarce, high-priced dedicated AI computing power into inclusive, on-demand AI public infrastructure.

In terms of product performance advantages, most decentralized GPU platforms currently on the market generally face issues such as high access thresholds for computing power, insufficient service credibility, and economic models that are difficult to operate sustainably. BTTInferGrid optimizes from the underlying architecture, achieving comprehensive breakthroughs in three dimensions: computing power aggregation, service verification, and economic system sustainability, forming a unique core competitiveness, with specific advantages as follows:

1. Open access computing power supply network, rapidly aggregating global idle GPU resources: Traditional cloud computing access thresholds are high (requiring compliant data centers, fixed public IPs, expensive switches, etc.), while BTTInferGrid has built a truly open access computing power supply network, allowing any entity or individual with idle GPU or other computing resources to seamlessly connect as long as they meet basic performance parameters (such as memory capacity, computing power benchmarks) and network stability requirements. This design significantly lowers the participation threshold for the supply side of computing power resources, enabling global idle GPU computing power to achieve rapid networking and matrix aggregation.

2. Verifiable service quality and node behavior, solving the trust issues of decentralization: The biggest pain point of decentralized computing is credibility—how to prevent miners from using low-end graphics cards to impersonate high-performance cards? How to ensure that reasoning results are authentic and credible? BTTInferGrid constructs a cross-verified closed loop through task scheduling (intelligent distribution), challenge verification (cryptographic sampling), consensus scoring (dynamic reputation scores), and on-chain coordination (smart contract rewards and penalties), effectively enhancing the credibility of reasoning services.

3. Demand-driven economic model, creating a sustainable ecosystem: Early DePIN projects often fall into a "high token issuance to attract nodes blindly mining, but due to a lack of real demand, lead to token inflation, price plummeting, and node exit" death spiral. BTTInferGrid established from the outset the goal of creating an economically sustainable ecosystem driven by real demand—using actual reasoning calls and node performance as the core incentive basis. Only when AI developers truly pay to call models can computing power providers receive core revenue sharing and reputation bonuses. This design will strongly promote the healthy growth of supply scale and market demand, ensuring the long-term healthy and sustainable development of the network ecosystem.

In summary, from breaking traditional access thresholds, allowing any globally compliant performance-standard idle GPU to seamlessly connect to the open supply grid, to building a fully verifiable trust defense line through task scheduling, challenge verification, consensus scoring, and on-chain rewards and penalties, and finally bidding farewell to speculative bubbles by anchoring incentives to the real demand-driven economic model of AI reasoning calls—BTTInferGrid is redefining the allocation mechanism of computing power resources from the dimensions of resource aggregation, service credibility, and value distribution.

BTTInferGrid will phase in the creation of a new computing ecosystem driven by real demand

BTTInferGrid is not simply "computing power aggregation," but a sophisticated decentralized computing power network that integrates AI reasoning task scheduling and execution, intelligent matching and connection of computing power supply and demand, and on-chain resource coordination and settlement.

In the decentralized computing ecosystem of BTTInferGrid, all participants form three core roles around the "supply, use, and verification" of computing power:

• Computing Power Suppliers (Miners): Provide idle GPU resources, undertake and execute AI reasoning tasks, with the system automatically allocating corresponding rewards based on verified actual workload, task completion quality, and dynamic performance scores.

• Computing Power Demanders (AI Developers): BTTInferGrid provides a standardized API service interface, supporting developers in accessing globally distributed GPU resources.

• Network Guardians (Validators): Participate in the decentralized verification and scoring system, auditing and randomly challenging the computational performance of miner nodes, identifying abnormal behaviors, and maintaining network service quality. Meanwhile, validators earn rewards by maintaining network integrity, jointly ensuring the fairness and credibility of the network.

In summary, for AI developers, BTTInferGrid brings a more cost-effective, highly scalable, and secure AI reasoning service, effectively alleviating issues of product interruptions and customer loss due to insufficient computing power. For GPU providers, it revitalizes global edge and idle hardware resources, establishing a sustainable revenue channel for GPU resource providers, allowing every unit of computing power to realize its value in the reasoning era.

In terms of specific product implementation, unlike the heavy asset model of traditional centralized cloud vendors that "first stack hardware, then wait for demand," DePIN inherently faces bidirectional coordination challenges during its initial construction—oversupply can lead to idle nodes and token economic collapse, while undersupply can harm developer experience and system efficiency. Therefore, BTTInferGrid has formulated a clear, robust, and demand-oriented phased launch strategy, abandoning disorderly and extensive growth, prioritizing resource utilization, economic sustainability, and steady expansion of technical architecture.

• Short-term goal (2026): Network cold start, complete the access of core underlying nodes and validation of distributed reasoning services, gradually expand the scale of GPU nodes.

• Mid-term goal (2027): Ecological diversification, improve the stability and privacy security of network services, while being compatible with more AI model formats and reasoning frameworks, gradually extending to application scenarios such as model fine-tuning.

• Long-term goal (2028 and beyond): Become the AI-native underlying infrastructure, building a computing layer preferred for AI Agents and automated applications, providing elastic computing support for large-scale AI applications, ultimately enabling computing power, distributed storage, and on-chain smart contracts to operate in a unified architecture.

In terms of implementation, BTTInferGrid also adopts a phased evolution strategy. In the initial launch phase, the network primarily uses professional graphics cards, with the supply side (miners) requiring approval for access, while demand side users can call reasoning services through the platform. In the future, it will evolve into a fully open supercomputing grid: supporting various types of GPUs, including consumer-grade, professional-grade, and data center-grade, with access and pricing based on performance; miners will have open access while introducing a staking mechanism to ensure service quality; the demand side will have an open standardized API interface, compatible with various AI model formats and reasoning frameworks, providing flexible deployment options.

Currently, BTTInferGrid has successfully integrated several mainstream AI open-source large models, including Alibaba Cloud's Qwen series Qwen3.6 27B and Qwen2.5 7B Instruct, as well as Meta's Llama 3.1 8B Instruct. AI developers can flexibly call based on actual business scenarios. In the future, the platform will continue to expand the model ecosystem, providing developers with more cutting-edge model support.

More importantly, BTTInferGrid has the long-term accumulation of BitTorrent and BTFS as a solid backing, possessing inherent development advantages. BitTorrent and its subsidiary BTFS have been deeply engaged in the decentralized storage field for many years, with BitTorrent having over 100 million active users and 2 billion installations, successfully validating the feasibility of the DePIN model and accumulating mature capabilities in resource access, token incentives, on-chain settlement, and community operations. As a strategic product for BitTorrent's layout in the AI track, BTTInferGrid is upgraded from the existing BTFS service, allowing these mature experiences to be seamlessly transferred to the AI reasoning computing power field, rapidly promoting ecosystem growth.

Relying on decentralized technology, BTTInferGrid precisely addresses the industry dilemma of "idle computing power" and "computing power shortage." Its concepts of open access, decentralized collaboration, verifiable contributions, and community co-construction not only provide a powerful breakthrough against the traditional centralized computing power monopoly but also, with a clear product positioning and solid technical foundation, outline an imaginative new blueprint for decentralized global computing power. Here, every unit of idle computing power will be activated, and every developer will be able to reach an intelligent future at inclusive costs.