NVIDIA did not disclose a $26 billion budget reserved for open-weight AI models. Its October 2025 filing reported $26 billion in multi-year cloud-service commitments expected to support research and development and DGX Cloud. NVIDIA executives later told WIRED that most of the investment would support open-model development. The distinction matters: commitments are not the same as cash already spent, and the filing did not assign the entire amount to model training.
What NVIDIA’s filing actually disclosed
As of October 26, 2025, NVIDIA reported $26 billion in multi-year commitments for cloud services. The company said the agreements were expected to support research and development efforts and DGX Cloud offerings, its managed AI-training service. The filing did not describe the $26 billion as a dedicated open-model budget. NVIDIA’s quarterly filing also warned that some cloud capacity could be reduced, terminated, or sold to other parties, which could lower the commitments.
The scheduled payments in that filing were:
| Fiscal period | Scheduled payment |
|---|---|
| Fiscal 2026, fourth quarter | $1 billion |
| Fiscal 2027 | $6 billion |
| Fiscal 2028 | $6 billion |
| Fiscal 2029 | $5 billion |
| Fiscal 2030 | $4 billion |
| Fiscal 2031 and thereafter | $4 billion |
These are scheduled payments under multi-year cloud-service commitments, not evidence that NVIDIA had already paid $26 billion or that every dollar would go toward model training.
The figure also has a date attached to it. In its next annual filing, NVIDIA reported $27 billion in multi-year cloud-service commitments as of January 25, 2026, supporting R&D. That later disclosure did not make the earlier $26 billion figure a model-specific allocation either. The fiscal 2026 Form 10-K is the newer commitment figure.
#1 Best Overall
- NVIDIA Volta GV100 Architecture — 4,608 CUDA Cores, 640 1st-Gen Tensor Cores delivering 14 TFLOPS FP32 and 112 TFLOPS deep learning performance for AI training, inference, HPC, and scientific computing workloads
- 32GB HBM2 ECC Memory — 900 GB/s Bandwidth — High-bandwidth memory on a 4096-bit bus with ECC error correction provides the memory capacity and throughput required for the largest AI models, simulations, and datasets
- PCIe 3.0 x16 Interface — 250W TDP — Standard PCIe Gen3 connectivity with passive cooling designed for enterprise rack server deployment in HPE ProLiant, Dell PowerEdge, and Supermicro platforms with adequate chassis airflow
- NVLink — Scale to 96GB Unified Memory — Connect two V100 GPUs via NVLink at 300 GB/s bi-directional bandwidth to scale GPU memory from 32GB to 96GB for larger AI training and HPC workloads
- Multi-Precision Computing — Supports FP64 (7 TFLOPS), FP32 (14 TFLOPS), FP16 (112 TFLOPS) and INT8 precision modes for flexible deployment across training, inference, and scientific simulation workloads
Why the $26 billion figure was linked to open models
WIRED’s March 11, 2026 report brought together the filing and interviews with NVIDIA executives. Those executives told the publication that most of the investment would support open-model development. WIRED also reported that NVIDIA was expanding its Nemotron work and that the company uses models to test its chips, networking, storage, and broader data-center designs. The headline’s open-model framing is therefore a reported interpretation based on the filing and interviews, not wording found in the SEC disclosure.
This distinction is useful for understanding both the money and the business strategy: cloud capacity can support a range of research and infrastructure work, while executives’ comments indicate that open models are a major intended use. The public filings cited here do not specify what share of the commitments will go to open-weight models.
What “open-weight” means—and what it does not
An open-weight model makes its trained parameters available for download, usually subject to a license. That can let developers run a model themselves, adapt it, or deploy it in a private environment rather than relying only on a hosted API. But the label does not, by itself, establish that the model’s training data, full training process, architecture, or evaluation materials are public.
Rank #2
- 【High-Performance APU】The MS-S1 MAX features an AMD Ryzen AI Max+ 395 APU, integrating a Zen 5 architecture CPU (up to 5.1GHz, 16C/32T, 64M L3 Cache), an RDNA 3.5 GPU, and an NPU (50 TOPS). The total system output is 126 TOPS. It provides powerful parallel computing capabilities for demanding AI workflows. It is ideal for running local LLMs, multimodal models, and computationally intensive tasks
- 【128GB UMA Memory】Equipped with up to 128GB of LPDDR5x-8000MT/s unified memory, it enables the CPU and GPU to access a shared, high-bandwidth memory pool with extremely low latency. Ideal for large-scale AI inference, 3D workloads, and complex timelines in video editing. It eliminates traditional VRAM bottlenecks, ensuring smoother data transfer during high-intensity computations. The UMA design maximizes performance stability under high loads
- 【Flexible Expansion】The MS-S1 MAX features USB4 V2 (up to 80Gbps), dual 10GbE LAN, HDMI 2.1 (up to 8K60), a full-length PCIe x16 expansion slot, and dual M.2 slots supporting up to 16TB RAID 0/1. Wi-Fi 7 provides stronger signal coverage and a more stable wireless experience. The slide-out design facilitates upgrades and maintenance. It easily adapts to personal, studio, or rack-mount enterprise environments
- 【High-Efficiency Cooling System】Utilizing an aerospace-grade aluminum alloy chassis, copper base plate, six heat pipes, dual turbine fans, and advanced PCM thermal conductive material, it maintains stable cooling performance even under continuous load. This system supports 130W continuous power and 160W peak power operation, with a built-in 320W power supply. It boasts multiple global certifications including CCC, FCC, UL, CE, and UKCA, ensuring stable and reliable operation in various environments
- 【Cluster Design】Two MS-S1 MAX units can be configured as a dual-unit cluster to run a large 235B Q4 model locally, achieving an output speed of 10.87 tok/s. Supporting 2U rack deployment, multiple MS-S1 MAX units can be cascaded into a distributed cluster to create a high-efficiency AI computing center. A cluster of four MS-S1 MAX units successfully ran a DeepSeek-R1 671B Q4 large model. A reserved cluster power-on interface allows for unified start-up and shutdown
Nor does “open-weight” automatically mean unrestricted commercial use or compliance with an open-source definition. A buyer or developer should check the specific license and materials released for each model instead of assuming that every use is allowed or that the model can be independently reproduced.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →NVIDIA’s model portfolio and the Nemotron push
WIRED reported that NVIDIA began releasing Nemotron models in November 2023. The company has also released models aimed at areas such as robotics, climate modeling, and protein folding. NVIDIA’s fiscal 2026 reporting described an expanding set of open-model efforts: Nemotron for agentic AI, Cosmos for physical AI, and Alpamayo for autonomous vehicles. It characterized the Nemotron 3 family as including open models, data, and libraries for specialized agentic-AI development. NVIDIA’s fiscal 2026 results provide the company’s description of that portfolio.
Nemotron 3 Super and benchmark claims
WIRED reported that NVIDIA launched Nemotron 3 Super alongside its March 2026 coverage and described it as a 128-billion-parameter model. The publication relayed NVIDIA’s claim of a score of 37 on the Artificial Intelligence Index, compared with 33 for GPT-OSS, and noted that some Chinese models scored higher. Those figures are reported company claims, not independent validation established by the filings.
Rank #3
- 【AI Max+ 395 AI Workstation】16 cores, 32 threads, up to 5.1 GHz boost and 80 MB cache. Integrated Radeon 8060S graphics with 40 CUs, RDNA 3.5, delivers performance close to RTX 4060/4070 laptop GPUs. Triple-engine design(CPU+GPU+XDNA 2 NPU) with up to 126 TOPS total, including 50+ TOPS dedicated NPU for local AI inference and machine learning acceleration. Ideal for AI development, content creation, virtualization, data analysis, and demanding multitasking. Compact, high-performance workstation.
- 【256-bit LPDDR5X MAX 128GB】The LPDDR5X onboard memory reaches 8400 MT/s - 1.5x faster than DDR5 SODIMM. Unlock the full potential of your graphics with massive 128GB memory pooling. This system allows you to manually assign up to 128GB of the onboard RAM to serve as video memory (VRAM) directly within the BIOS setup, delivering unparalleled performance for 4K video editing, and AI model training without the need for a discrete graphics card.
- 【Lastest GPU 8060S & XDNA 2 NPU】Built on the RDNA 3.5 architecture, the AMD Radeon 8060S Graphics iGPU features 40 compute units (2,560 stream processors). It delivers performance on par with NVIDIA's mobile RTX 4070, efficient encoding/decoding for AVC, HEVC, VP9, and AV1 video codecs. And It can connect 4 screens via HDMI & DisplayPort & Full Featured USB4 x2 to efficiently handle your tasks and meet your specific needs. Supports 8K/4K resolution displays.
- 【Dual LAN (2.5GbE+10GbE)& WiFi 7】The computer has double LAN, one is 2.5GbE (I226), the other is 10GbE(AQC113). provides more applications, such as firewall, soft routing, multichannel aggregation. Built-in WiFi module, support WiFi 7 and Bluetooth5.4. Known as 802.11be, Wi-Fi 7 promises up to 46Gbps theoretical throughput, making it 4.8x faster than Wi-Fi 6. and computer has 4 built-in NVMe SSD slots, 1 SD card slot, allowing you to expand its storage capacity.
- 【Engineered to Endure】The computer measures 7.13 x 7.24 x 2.99 inches. AI mini pc is encased in a premium all-aluminium chassis. Dual turbo CPU fans deliver silent, ultra-efficient cooling, To enable the computer to maintain stable operation for a long time. We offer up to 2 years warranty and lifetime professional customer service. Please feel free to contact us if any issues happened. thanks
When assessing such comparisons, distinguish total parameter count from the number of parameters active for a given inference, and a benchmark score from cost, speed, and quality in a real deployment. Results depend on the exact model version, test conditions, and hardware; they do not establish that one model is best for every task.
Why a chipmaker would invest in open models
Models can help shape the hardware roadmap
A model is both an AI product and a demanding workload. NVIDIA executive Kari Briski told WIRED that the company builds models to stress-test compute, storage, networking, and its broader data-center architecture roadmap. Developing models in-house can help NVIDIA understand how its infrastructure performs and which hardware and software improvements matter to users.
Open releases can draw developers into NVIDIA’s stack
A downloadable model can lower the barrier to experimenting with NVIDIA systems, while tools such as CUDA, NeMo, NIM, TensorRT-LLM, and DGX Cloud can provide paths to training, fine-tuning, and deployment. NVIDIA’s fiscal 2025 filing describes DGX Cloud as a managed AI-training service and identifies NIM, NeMo, and AI Blueprints as parts of its AI software strategy.
Rank #4
- 【AI Max+ 395 AI Workstation】16 cores, 32 threads, up to 5.1 GHz boost and 80 MB cache. Integrated Radeon 8060S graphics with 40 CUs, RDNA 3.5, delivers performance close to RTX 4060/4070 laptop GPUs. Triple-engine design(CPU+GPU+XDNA 2 NPU) with up to 126 TOPS total, including 50+ TOPS dedicated NPU for local AI inference and machine learning acceleration. Ideal for AI development, content creation, virtualization, data analysis, and demanding multitasking. Compact, high-performance workstation.
- 【256-bit LPDDR5X MAX 128GB】The LPDDR5X onboard memory reaches 8400 MT/s - 1.5x faster than DDR5 SODIMM. Unlock the full potential of your graphics with massive 128GB memory pooling. This system allows you to manually assign up to 128GB of the onboard RAM to serve as video memory (VRAM) directly within the BIOS setup, delivering unparalleled performance for 4K video editing, and AI model training without the need for a discrete graphics card.
- 【Lastest GPU 8060S & XDNA 2 NPU】Built on the RDNA 3.5 architecture, the AMD Radeon 8060S Graphics iGPU features 40 compute units (2,560 stream processors). It delivers performance on par with NVIDIA's mobile RTX 4070, efficient encoding/decoding for AVC, HEVC, VP9, and AV1 video codecs. And It can connect 4 screens via HDMI & DisplayPort & Full Featured USB4 x2 to efficiently handle your tasks and meet your specific needs. Supports 8K/4K resolution displays.
- 【Dual LAN (2.5GbE+10GbE)& WiFi 7】The computer has double LAN, one is 2.5GbE (I226), the other is 10GbE(AQC113). provides more applications, such as firewall, soft routing, multichannel aggregation. Built-in WiFi module, support WiFi 7 and Bluetooth5.4. Known as 802.11be, Wi-Fi 7 promises up to 46Gbps theoretical throughput, making it 4.8x faster than Wi-Fi 6. and computer has 4 built-in NVMe SSD slots, 1 SD card slot, allowing you to expand its storage capacity.
- 【Engineered to Endure】The computer measures 7.13 x 7.24 x 2.99 inches. AI mini pc is encased in a premium all-aluminium chassis. Dual turbo CPU fans deliver silent, ultra-efficient cooling, To enable the computer to maintain stable operation for a long time. We offer up to 2 years warranty and lifetime professional customer service. Please feel free to contact us if any issues happened. thanks
That integration may make development easier for teams already using NVIDIA GPUs. It can also encourage reliance on NVIDIA-specific tooling. Whether that becomes a lock-in problem depends on the model license, available serving software, and how well the workload runs on other accelerators.
Open models broaden the market for AI infrastructure
If more companies run, fine-tune, and deploy models themselves, demand can grow for inference capacity, data-center systems, and related software. NVIDIA’s fiscal 2026 results said inference providers had cut AI costs by up to 10 times using open-source models on Blackwell. That is a company-reported claim, not a universal cost comparison: actual economics depend on the model, workload, hardware, utilization, and deployment choices.
Competition from other open-model ecosystems matters
WIRED framed NVIDIA’s push partly as a response to the popularity of open models from DeepSeek, Alibaba, Moonshot AI, Z.ai, and MiniMax. If developers build around models and tools optimized for other hardware, NVIDIA could face pressure on its platform position. Releasing models and supporting their deployment gives the company a way to compete for developers’ attention as well as for chip sales.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
The Nemotron Coalition puts the strategy into practice
On March 16, 2026, NVIDIA announced the Nemotron Coalition, a collaboration involving companies and labs including Mistral AI, Perplexity, LangChain, Cursor, Black Forest Labs, Sarvam, Reflection AI, and Thinking Machines Lab. NVIDIA said participants would contribute data, evaluations, expertise, and domain knowledge. The first project is a base model co-developed by NVIDIA and Mistral AI, trained on DGX Cloud and intended to be shared with the open ecosystem as the basis for Nemotron 4. These are plans described in NVIDIA’s coalition announcement; the announcement does not establish the model’s eventual license or performance.
How developers and enterprise buyers should evaluate the models
Open weights can give organizations more deployment choices, but they do not eliminate the work or cost of running AI. Before adopting a model, assess the operational details as carefully as the benchmark results.
- License: Check commercial-use, redistribution, fine-tuning, and deployment terms for the specific model.
- Openness: Identify what is actually released—weights, code, training data, recipes, model documentation, and evaluation details.
- Hardware portability: Test whether the model performs acceptably on the accelerators and serving stack you can use, rather than assuming NVIDIA optimization is vendor-neutral.
- Inference economics: Compare memory needs, throughput, quantization support, utilization, and total operating cost for your workload.
- Task quality: Evaluate the tasks that matter to you, including tool use, reasoning reliability, multilingual performance, and error rates.
- Deployment and data governance: Confirm support for your chosen serving environment and whether data can remain within the required private or regulated boundary.
- Maintenance and risk: Review the model’s update and security commitments, regulatory obligations, and the consequences of deeper dependence on one supplier’s hardware and software.
Downloadable weights can support local or private deployment, fine-tuning, and greater control over latency and data residency. They do not guarantee low operating costs: large models can still require expensive compute, and licensing or hardware constraints may narrow the practical options.
Is NVIDIA becoming a direct rival to OpenAI and Anthropic?
NVIDIA is moving further into model development and can compete for research talent, developer mindshare, and enterprise workloads. But its open-model effort is not the same product strategy as operating a proprietary consumer chatbot or selling access to a closed frontier-model API. NVIDIA’s central business remains accelerated computing and infrastructure; its models can also serve as reference systems and tools that make that infrastructure more useful.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →WIRED described the move as a potential route toward becoming a frontier lab capable of competing with companies such as OpenAI and DeepSeek. That is a strategic possibility, not evidence that NVIDIA has matched the leading proprietary labs. Even if NVIDIA’s models do not lead every benchmark, the company can benefit if they encourage more developers to build on its platforms.
What remains unclear
The public disclosures cited here do not identify the portion of the $26 billion or later $27 billion commitment allocated specifically to open-weight models, the terms of every cloud agreement, or the budget for individual training runs. They also do not establish whether every future NVIDIA model will ship with weights, what licensing and data transparency Nemotron 4 will provide, or what return NVIDIA expects from these investments.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
