NVIDIA DGX Spark Launch Delay Explained: Current Price, Availability, and Specs

iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more

The NVIDIA DGX Spark is already on sale. Its retail launch began on October 15, 2025—not October 15 as an upcoming date—and the original May 2025 availability target was missed by roughly five months. As of August 12, 2026, NVIDIA’s U.S. marketplace lists the compact AI workstation at $4,699, although prices, configurations, and stock can vary by seller and region.

The system began life as Project DIGITS, a $3,000 personal AI supercomputer announced at CES 2025. NVIDIA renamed it DGX Spark in March 2025, moved partner availability first to July, and ultimately opened ordering in October. Here is what changed, what the machine actually offers, and who should consider buying one.

The short version

• Historical launch date: October 15, 2025.
• Original target: May 2025, announced as Project DIGITS.
• Approximate delay: Five months from the original May target.
• Current U.S. NVIDIA price: $4,699 as listed on August 12, 2026.
• Core platform: NVIDIA GB10 Grace Blackwell superchip with a 20-core Arm CPU, Blackwell GPU, and 128 GB of unified memory.
• Best suited to: Local AI development, inference, model fine-tuning, research, education, and data-science workflows.
• Not suited to: Buyers expecting a conventional gaming mini-PC, a general-purpose desktop replacement, or an across-the-board 1-petaFLOP FP32 machine.

The launch delay was real, but the more important buying story is that the product also arrived at a substantially higher price than NVIDIA’s original announcement suggested. Project DIGITS was introduced at a starting price of $3,000. The Founders Edition was later listed at $3,999 and then increased to $4,699 in February 2026, with NVIDIA attributing the increase to worldwide memory-supply constraints rather than a hardware or configuration change.

DGX Spark’s launch timeline

This chronology matters because some older reports still describe May or July 2025 as the expected launch window. Those dates were announced targets, not the final retail release date. October 15, 2025 is now the historical launch date.

Current availability and price

NVIDIA’s U.S. marketplace currently lists the NVIDIA DGX Spark at $4,699. The page also identifies authorized retail and channel partners, including Amazon, Micro Center, and PNY. That does not mean every seller has the same stock, warranty terms, storage configuration, or price. Buyers should verify the exact seller, model, storage capacity, warranty, shipping location, and return policy before ordering.

For readers comparing local-AI workstations, the NVIDIA DGX Spark is the relevant product listing to check first. A live retailer listing may differ from NVIDIA’s direct price, and the price is especially important here because the final system costs substantially more than the $3,000 starting price announced for Project DIGITS.

NVIDIA’s February 2026 price increase is also worth separating from the launch delay. The delay moved availability from May to October 2025. The later price increase was attributed to memory-supply constraints and, according to NVIDIA’s developer forum, did not accompany a hardware or configuration upgrade.

What hardware do you get?

DGX Spark is built around NVIDIA’s GB10 Grace Blackwell superchip. It combines a 20-core Arm CPU and a Blackwell GPU in a small desktop enclosure. Unlike a conventional PC with separate system RAM and graphics memory, the platform uses a large unified memory pool that both the CPU and GPU can access.

The small enclosure is one of DGX Spark’s strongest selling points. It provides a substantial unified-memory pool without requiring a tower workstation, rack server, or external GPU enclosure. The trade-off is that this is a specialized integrated system: buyers should not assume the same upgradeability, component replacement options, or expansion flexibility available from a conventional desktop.

What does “1 petaFLOP” mean here?

NVIDIA rates DGX Spark at up to 1 petaFLOP at FP4 precision with sparsity, as well as up to 1,000 TOPS of inference performance. Those figures describe AI workloads under specific precision and sparsity conditions. They should not be interpreted as a general-purpose FP32 benchmark or as a direct comparison with every desktop GPU, workstation GPU, or server accelerator.

FP4 uses four-bit floating-point computation, which can be useful for suitably optimized AI inference workloads. Sparsity can further reduce the amount of computation required when the model and software can take advantage of it. Real-world results will depend on the model, quantization, framework, batch size, memory behavior, and software optimization.

In practical terms, the headline number is best understood as an AI-compute capability claim—not as a promise that every application will run at one petaFLOP or that DGX Spark will outperform a larger system in every task.

How large a model can DGX Spark run?

NVIDIA says one DGX Spark can support models of approximately 200 billion parameters. Two connected systems can support models of up to approximately 405 billion parameters. These are platform-positioning figures, not guarantees that every model of that size will run comfortably at every context length, quantization level, or performance target.

Memory requirements rise with parameter count, context length, runtime overhead, activation storage, and the precision used. A model that technically fits may still be too slow for an interactive workflow or may leave insufficient room for datasets, compilation, caching, and other services. Buyers should therefore treat NVIDIA’s model-size claims as an indication of the system’s intended scale, not as a substitute for testing their specific model.

Who should buy DGX Spark?

DGX Spark is aimed at people who need a local NVIDIA AI development environment in a compact form factor:

• AI developers who want to prototype and test models locally before moving workloads to a larger server or cloud environment.
• Researchers working with large language models, multimodal models, inference pipelines, or fine-tuning workflows.
• Data scientists who need local GPU-accelerated experimentation and a large shared memory pool.
• Educators and students who need a self-contained platform for learning modern AI frameworks.
• Small teams that want a local development or demonstration system without immediately purchasing a multi-GPU server.
• Organizations that value NVIDIA’s software ecosystem and want to keep some development or inference workloads on premises.

It is less compelling if your main use is gaming, office productivity, web browsing, video editing, or ordinary desktop computing. DGX Spark’s price is difficult to justify as a generic mini-PC, and its value depends heavily on using the accelerated AI hardware and software stack.

Software: more than a small computer with a GPU

NVIDIA positions DGX Spark as a complete AI development platform rather than merely a compact hardware box. The software environment is intended for machine learning, data science, inference, deployment, and fine-tuning, with support for tools and frameworks including PyTorch and TensorRT-LLM.

NVIDIA also provides a 90-day NVIDIA AI Enterprise—DGX Spark trial or license path. NVIDIA describes this software as an enterprise-grade platform for AI development, deployment, and optimization on DGX Spark and other GB10-based systems.

Licensing deserves attention before using the machine commercially. NVIDIA’s EULA distinguishes evaluation, development, testing, production, and enterprise-license use cases. Some NIM components designated as free SDKs may be used on qualifying personal systems under stated conditions, while other commercial, production, or multi-user deployments may require a subscription or another license. The exact terms depend on the software component and use case, so organizations should review the applicable NVIDIA license rather than assuming that all included tools are unrestricted for production use.

Partner systems and DGX Spark alternatives

NVIDIA did not launch only one chassis. Partner systems were announced from Acer, ASUS, Dell Technologies, GIGABYTE, HP, Lenovo, and MSI. NVIDIA’s certification documentation lists examples including the ASUS Ascent GX10, GIGABYTE AI TOP ATOM, HP ZGX Nano AI Station, Lenovo ThinkStation PGX, and MSI EdgeXpert among certified DGX Spark or GB10 systems.

These machines are useful comparison points, but they should not automatically be treated as identical to NVIDIA’s branded Founders Edition. A partner system can differ in:

• Internal storage capacity and drive configuration
• Chassis design and cooling system
• Noise characteristics and thermal limits
• Power-delivery design
• Warranty length and support process
• Operating-system image and preinstalled software
• Regional availability and pricing

Certification establishes an important relationship with NVIDIA’s platform, but it does not eliminate product-level differences. Compare the complete specification sheet and support terms for the exact OEM model.

For buyers evaluating DGX Spark alternatives, the ASUS Ascent GX10 and Lenovo ThinkStation PGX are especially relevant named systems, but availability and pricing should be checked in the buyer’s country. The partner ecosystem may offer a better warranty, distribution channel, chassis, or configuration than NVIDIA’s own unit, depending on the market.

Planning the physical setup

DGX Spark includes four USB-C ports, HDMI 2.1a, 10 GbE, Wi-Fi 7, Bluetooth 5.4, and ConnectX-7 networking. That is a useful selection for a compact development station, but the right accessories depend on how the system will be used.

Display and peripherals

A monitor, keyboard, mouse, and other peripherals can be connected directly. If you need to attach several USB devices, a USB-C dock for DGX Spark may simplify the desk setup. Verify the dock’s data, display, Ethernet, and power behavior for this specific system; a dock should not be treated as an NVIDIA-required accessory or as automatically compatible with every peripheral combination.

Datasets, models, and checkpoints

The internal storage options are 1 TB and 4 TB. Large model files, datasets, container images, checkpoints, and generated outputs can consume that space quickly. An external NVMe SSD for AI models can be useful for moving datasets or keeping checkpoints separate from the operating-system drive, but verify the enclosure interface, sustained-transfer behavior, filesystem support, and intended connection method before buying.

Fast local networking

The documented 10 GbE and ConnectX-7 networking make DGX Spark suitable for workflows involving a fast local file server, network-attached storage, or another AI system. A 10GbE switch for local AI workstation can be relevant when several machines or a high-speed storage system need to communicate, but it is unnecessary for a simple standalone setup. The rest of the network—including the switch, cabling, storage server, and remote system—must also support the desired throughput.

What the delay and price increase mean for buyers

The delayed launch does not by itself indicate that the platform is unsuitable. Hardware launches can move as manufacturing, software, validation, supply, and partner-planning work develops. The more concrete buyer concern is the gap between the original announcement and the final product economics.

Project DIGITS was presented as a personal AI supercomputer starting at $3,000. The current NVIDIA-branded system costs $4,699 in the U.S. marketplace listing, a major increase over that original entry point. The machine may still make sense for someone who would otherwise rent substantial cloud compute, needs local data handling, or values a ready-made NVIDIA development environment. It is harder to justify for occasional experimentation or buyers who already have access to a capable workstation or cloud GPU.

Cloud-versus-local economics also depend on utilization. A local system has an upfront cost, electricity, maintenance, and eventual obsolescence. Cloud hardware has recurring usage charges but can scale up or down and may provide access to larger accelerators. DGX Spark’s strongest case is consistent local use, not necessarily a few hours of experimentation each month.

Buying checklist

1. Confirm the date and model. The October 15, 2025 date is historical. Make sure the listing is for the current DGX Spark or a clearly identified GB10 partner system.
2. Check the exact price and storage. NVIDIA lists 1 TB and 4 TB configurations in its documentation, and retail prices can vary.
3. Verify the seller and warranty. Authorized-channel status, return rights, regional support, and warranty coverage differ by seller and country.
4. Match the memory to the workload. The 128 GB unified pool is central to the platform’s value, but model fit still depends on precision, context, runtime overhead, and other memory demands.
5. Review licensing. Confirm whether your intended use is evaluation, development, testing, production, or multi-user enterprise deployment.
6. Plan storage and networking. Decide whether your models and datasets need external storage, 10 GbE, or a connection to another machine.
7. Compare certified partner systems. An OEM system may provide a different chassis, cooling solution, support package, or regional purchase path.
8. Do not buy based only on the petaFLOP headline. Look for workload-specific evidence using your own model, framework, precision, and batch size.

Bottom line

NVIDIA DGX Spark did go on sale, but later than originally promised. Project DIGITS was announced for May 2025 at $3,000, partner availability was later described as starting in July, and the actual retail launch began on October 15, 2025. By August 2026, NVIDIA’s U.S. price for the branded Founders Edition was $4,699.

The final product is a compact GB10 Grace Blackwell AI workstation with 128 GB of unified memory, fast networking, and an NVIDIA-focused software stack. Its advertised ability to handle models approaching 200 billion parameters on one unit—and roughly 405 billion across two connected systems—makes it interesting for serious local AI development. But the 1-petaFLOP figure is an FP4-with-sparsity AI metric, not a universal performance rating.

Buy DGX Spark for its unified-memory AI platform and local NVIDIA workflow—not simply because it is small or because it carries a petaFLOP headline.

Frequently Asked Questions

Is NVIDIA DGX Spark delayed?

It was delayed relative to NVIDIA’s original schedule. Project DIGITS, later renamed DGX Spark, was announced for May 2025. NVIDIA later described partner availability beginning in July, but ordering and retail availability ultimately began on October 15, 2025.

How much does NVIDIA DGX Spark cost?

As of August 12, 2026, NVIDIA’s U.S. marketplace lists the branded DGX Spark at $4,699. Retailer prices, stock, storage configurations, taxes, and shipping can vary.

Why did the DGX Spark price increase?

NVIDIA’s developer forum attributed the February 2026 Founders Edition MSRP increase—from $3,999 to $4,699—to worldwide memory-supply constraints. NVIDIA said the increase was not accompanied by a hardware or configuration change.

Can DGX Spark replace a gaming PC?

It can perform ordinary desktop tasks, but that is not its purpose. DGX Spark is designed for local AI development, inference, fine-tuning, research, and data science. Its price and specialized platform make it a poor value as a conventional gaming or office mini-PC.

What is the 1-petaFLOP claim based on?

NVIDIA’s rating is up to 1 petaFLOP at FP4 precision with sparsity, plus up to 1,000 TOPS of inference. It is a conditional AI-compute figure, not a general FP32 benchmark or guarantee of application performance.

What are the main DGX Spark partner alternatives?

NVIDIA has identified systems from Acer, ASUS, Dell Technologies, GIGABYTE, HP, Lenovo, and MSI. Certified examples include ASUS Ascent GX10, GIGABYTE AI TOP ATOM, HP ZGX Nano AI Station, Lenovo ThinkStation PGX, and MSI EdgeXpert. Their storage, cooling, warranty, support, price, and availability can differ from NVIDIA’s branded unit.