In the investment frenzy of computing power in 2026, two seemingly unrelated pieces of news emerged almost simultaneously: on one side, the AI computing power service provider Nscale announced a long-term strategic partnership with the humanoid robot startup Figure, with the cooperation scale reportedly exceeding 6 billion dollars, potentially locking in and deploying up to 100,000 NVIDIA Vera Rubin platform GPUs in the future; on the other side, the AI company Humain, supported by Saudi Arabia's sovereign wealth fund PIF, launched its first fundraising round of about 2.5 billion dollars to prepare a fund focused on investing in data centers, financing its data center project in Saudi Arabia in collaboration with Al Moammar Information Systems—this project initially plans for a capacity of about 250 megawatts, with a long-term goal of expanding to about 1 gigawatt. This is a capital race around AI computing power: Nscale-Figure bets on locking GPU resources in advance, while Humain attempts to create a new computing hub by combining electricity and data center space with long-term funds; they have no direct business relationship yet are competing for future computing capacity and infrastructure pricing power on the same track. As demand for computing power continues to swell on both the training and inference fronts, "where does the money come from, and where do the chips go" is no longer a financial technical detail but a core variable that determines the computing power landscape: whoever can gain the upper hand in the game of capital source structure and allocation of computing resources will have a better chance of seizing the initiative in the next round of AI infrastructure restructuring.
6 Billion Contract Secures Figure's Humanoid Computing Power
For Figure, which specializes in humanoid robots, computing power is no longer merely "backend resources," but a core component akin to motors and joints. To enable robots to perform complex tasks like grasping, walking, and interacting in real environments, it requires repeated iterations of large-scale training models and then compressing these models into a form that can perform stable inference on devices; each new dataset collected and each new scenario means a new round of training and inference consumptions. Such a business rhythm is difficult to establish on an uncertain and potentially short-supplied cloud GPU supply, thus Figure needs a solid computing power backing that can extend over many years and continuously expand.
Nscale, in this context of demand, has "locked in" future GPU resources through a long-term strategic cooperation. Public information indicates that this collaboration involves over 6 billion dollars, potentially involving up to 100,000 of NVIDIA's next-generation Vera Rubin platform GPUs (according to a single source). Essentially, Figure is betting with real capital on the humanoid robot training and inference needs for the next few years, and it is also an early realization of Nscale's voice in the computing power service market. Currently, there is no authoritative public data on the specific performance parameters and production timeline of Vera Rubin, nor have the contract's duration, payment terms, and delivery schedules been disclosed. However, from the amount and scale alone, it is evident this is no longer the traditional "pay-as-you-go cloud computing power," but has evolved into a large contract model that locks in long-term capacity: computing power service providers securing supply channels for the next-generation GPU platform in advance, while cutting-edge application providers lock in their training and inference lifelines with long-term, large orders. As similar contracts grow, the computing power industry is transitioning from the past model of elastic on-demand supply to a phase dominated by capitalized long-term capacity reservations.
Saudi PIF Bets on 1 Gigawatt Data Center
If Nscale and Figure’s long-term computing power contract is about "locking orders," then the Saudi sovereign wealth fund PIF has chosen to "lock underlying assets." The AI company Humain, supported by PIF, is positioned at the forefront of this strategic game: on one hand, it undertakes the strategic intentions of sovereign capital in the AI field, positioning itself as the operational and integration platform for the computing power ecosystem in the region; on the other hand, it functions as a market-based company to handle fundraising and project deployment, bridging the gap between policy vision and concrete engineering. This structure allows PIF to avoid personally managing each data center while still maintaining decision-making authority at key nodes through controlling and supportive funding.
In terms of specific projects, Humain has chosen to collaborate with the local Saudi IT service provider Al Moammar Information Systems to advance data center construction: the initial planned capacity is about 250 megawatts, while the long-term overall scale is drawn up to approximately 1 gigawatt, directly parallel to the scale of globally ultra-large computing power clusters. To provide "ammunition" for this chain, according to Bloomberg reports, Humain is raising about 2.5 billion dollars to establish a fund focused on investing in data centers, providing financing for cooperative data center assets. The key terms of the fund's investor structure, management fee rate, and duration have not yet been disclosed, but from the setting of allocating funds to a single asset class, it appears closer to a project financing channel dominated by sovereign capital: first locking the money into a fund dedicated to electromechanical, land, and rack investments, and then the fund providing ongoing leverage for phased expansions to push from 250 megawatts to 1 gigawatt, evolving a single fund into a long-term possession of computing power infrastructure assets.
Sovereign Capital and Startups Compete for Computing Power Assets
If we place Humain’s data center fund and Nscale-Figure's long-term contract on the same balance sheet, they almost stand at opposite ends of the computing power capital chain. On the Humain side, sovereign wealth funds like PIF and local IT companies such as Al Moammar Information Systems collaborate to invest, pouring money directly into data centers, racks, and power capacity, in exchange for hard assets such as data centers starting from 250 megawatts, planning long-term expansions to about 1 gigawatt, as well as infrastructure cash flow templates covering multi-year construction periods. The Nscale-Figure collaboration is led by computing power service providers and humanoid robot startups, with downstream application demand driving upstream resource locks; the core assets are not a piece of land or a building but are the schedulable NVIDIA Vera Rubin platform GPU computing power anticipated for the coming years, representing a prior reservation and possession of "available capacity." The two sides also diverge in terms of funding sources and risk preferences: the former is closer to relying on sovereign capital credit, pursuing ultra-long-cycle infrastructure returns; the latter bets on the time axis of product and model iterations, willing to take on over 6 billion dollars in contracts to ensure Figure's future training and inference won't be hindered by resource shortages.
Looking from a broader perspective, sovereign wealth funds and industrial capital are delineating their respective spheres of influence around AI computing power infrastructure. In recent years, PIF has continuously increased its investments in global technology and AI infrastructure, providing projects like Humain with a "foundation" to solidify local electricity and data center capabilities, while the contract models like Nscale-Figure represent global industrial capital and startups competing for upper-tier computing power benefits by locking in GPU production capacity and capping future usage rights. Currently, there is no public evidence showing a direct business or equity correlation between these two transactions; they are pushing forward at the same time independently but point to the same structural change: computing power is no longer merely an operational cost on corporate profit and loss statements; it is being split into tradable, financeable, and reservable long-term infrastructure assets and financial instruments, and being pre-positioned and sliced by different types of capital.
10,000 GPUs and 1 Gigawatt Data Center Supply and Demand Dynamics
When Nscale and Figure write potential orders for up to 100,000 NVIDIA Vera Rubin platform GPUs in multi-year contracts, the gauge of supply-side expansion is suddenly stretched to "massive scale." Almost simultaneously, the data center planned by Humain and its partners begins at approximately 250 megawatts while aiming for about 1 gigawatt in the long term, while another line plunges from data centers and land down into the power grid and infrastructure. The demand for 100,000 top-end GPUs corresponds to an entire upstream demand for training clusters, while a 1 gigawatt-level data center means locking in substantial electricity loads, cooling systems, and fiber optic networks at a specific geographic coordinate for the long term—these two expansion curves apply pressure to the entire computing power base at different links but the same magnitude.
The issue is that this pressure will not linger just on the designers' blueprints. A 1 gigawatt-level data center requires the local power grid to reserve a considerable amount of capacity for it, water-cooling or air-cooling systems need to continuously dissipate heat for high-density GPU racks, and backbone networks must leave redundant bandwidth for massive model training and inference traffic—any delays at any point could create a time gap between "signed orders" and "data centers that can be opened." The swift supply-side expansion is expected to lead to downward pressure on computing power prices in the short term, alleviating congested queues and high rental costs, but simultaneously pushing the resource allocation dynamics from "who has the money to buy computing power" to "who is qualified to occupy the best positions in electricity, geography, and networks," raising the entry threshold for leading collaborations and ultra-large scale projects further. The timing gaps and uncertainties between supply and demand will ultimately determine who gains greater initiative in this round of computing power capital race in terms of prices, resources, and thresholds.
The Next Chapter of AI Infrastructure After the Escalation of the Computing Power War
When comparing Nscale-Figure and Humain on the same timeline, a clear outline emerges: on one end is a multi-year collaboration worth over 6 billion dollars that secures NVIDIA Vera Rubin platform GPUs for cutting-edge applications like humanoid robots; on the other end is a data center fund of about 2.5 billion dollars driven by sovereign capital like PIF, promoting a project starting from 250 megawatts and planning for around 1 gigawatt in Saudi Arabia. These are not two isolated transactions but rather a script that is in rapid operation—computing power service providers, equipment suppliers, and sovereign capital are jointly transforming AI infrastructure into a core battleground for high-intensity capital expenditure and long-term games. The next chapter may unfold along several paths: first, computing power supply may further concentrate in the hands of a few comprehensive players who can simultaneously secure GPU quotas, control electricity and site, and sign multi-year large contracts, leaving smaller teams to compete for leftover resources on the "track" they build; second, regional clusters formed around electricity prices, geography, and policies may gradually take shape, with nodes in Saudi Arabia and other countries competing in a shifted manner, and supply and demand no longer being a unified global pool but a movement and arbitrage between multiple regional pools; third, models like Humain that finance data centers in the form of specialized funds may be imitated by more capital, with long-term computing power contracts, infrastructure funds, and other tools further financializing GPU and electromechanical assets, making "who can raise money" and "how long is the money willing to be locked" equally important competitive dimensions alongside technical capabilities. For readers, what truly warrants continuous attention is whether the GPUs from the Nscale-Figure collaboration can be delivered as scheduled according to the contract, how the contract duration and exclusivity are delineated, whether Humain's fund can reach its target scale and how its investor structure shapes up, and how the construction rollout of the Saudi data center evolves from 250 megawatts to a larger scale, as these variables' actual trajectories will directly shape the concentration of AI computing power supply and demand, regional patterns, and the boundaries of capital power.
Join our community to discuss and become stronger together!
AiCoin exclusive Hyperliquid benefits: https://app.hyperliquid.xyz/join/AICOIN88
AiCoin exclusive Aster benefits: https://www.asterdex.com/zh-CN/referral/9C50e2
On-chain Telegram community: https://t.me/AiCoinWhaleData
On-chain community: https://www.aicoin.com/link/chat?cid=N6OVMor5g
AiCoin on-chain Twitter: https://x.com/aicoinwhaledata
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
