AI news videos

NVIDIA DGX Spark 64GB Arrives Oct. 23 Starting At $4,999

NVIDIA announced a new 64GB configuration of DGX Spark on its own corporate blog. NVIDIA says the configuration will be available on Friday, Oct. 23 from Acer, ASUS, Dell, Gigabyte, HP and MSI, starting at $4,999, and that it is available exclusively from those manufacturer partners.

Watch on YouTube

Transcript

Sources

Every claim in this video comes from the top ranking coverage of this topic. The claims and where each one came from:

• Backing page facts, quoted exactly: the headline "NVIDIA DGX Spark 64GB Gives Developers More Ways to Build and Scale Local AI", the deck line "Starting Friday, Oct. 23, the new configuration will be available from Acer, ASUS, Dell, Gigabyte, HP and MSI", and the body line "Coming this month, NVIDIA DGX Spark will be available with 64GB of unified memory from top manufacturer partners". EDITORIAL DECISION: sentence one states only the plainest fact of the announcement, that there is a new DGX Spark configuration carrying less memory than the existing one, which is NVIDIA's own framing of a lower memory SKU alongside the 128GB model. Sentence two, "NVIDIA announced it on its own blog", puts provenance inside the first five seconds, which is required here because there is no outside check of anything on the page and no corroboration was obtainable. The hook is exactly two sentences, because the renderer splits the hook into two on-screen beats and a one sentence hook renders the second beat as undefined and forces a full voiceover redo. The hook speaks no figure at all, which keeps the memory figure in scene 1 and the money amount in scene 2 well clear of the opening. The glued brand and digit token "64GB" is kept entirely off mic in the hook and the product is called "DGX Spark", which the voice engine reads cleanly. REFUSED: the article's own opening line "Local AI is becoming more useful by the token" is a figure of speech and is barred from the whole file. The headline verb phrase "Gives Developers More Ways" is vendor marketing framing and is not narrated. The deck's word "seamlessly" is refused outright.(HOOK BACKING. First-party NVIDIA corporate blog post on blogs.nvidia.com, headlined "NVIDIA DGX Spark 64GB Gives Developers More Ways to Build and Scale Local AI", bylined Allen Bourgoyne, datelined October 2, 2026, captured 2026-10-02T13:12:47.331Z, filed under the AI category with tags including DGX Spark and Local AI. This is the vendor publishing on its own blog about its own hardware, so it is the sole and self-interested source for every narrated claim in this Short.)
• Backing sentences, quoted exactly: "Coming this month, NVIDIA DGX Spark will be available with 64GB of unified memory from top manufacturer partners", and "The new 64GB configuration, available exclusively from manufacturer partners, keeps the platform at an accessible price point while retaining the GB10 Grace Blackwell Superchip, DGX OS and full NVIDIA AI software stack", which the page then completes with the phrase "same as the 128GB model". Also quoted: "DGX Spark combines NVIDIA Grace Blackwell compute, unified memory, NVIDIA ConnectX-7 networking and an NVIDIA CUDA-accelerated AI software stack in one system." and "It supports up to 100-billion-parameter models and the agentic applications built on them, fully on device." EDITORIAL DECISION: the memory amount is spoken as words, "sixty four gigabytes", never as the glued token "64GB", because the voice engine invents extra words on brand and digit tokens glued together; the written form "DGX Spark 64GB" appears only on the silent scene card and in the title candidates. The chip is spoken as "Grace Blackwell", which reads cleanly, while the model designator "GB10" and the word "Superchip" are held to the silent card. Sentence two is attributed out loud as "NVIDIA says", because the parity claim against the larger model is a vendor spec assertion that nobody outside NVIDIA has checked. The comparison target is spoken as "the larger model" rather than the token "128GB", for the same brand and digit reason, and because a second figure in that sentence would sit adjacent to the memory figure in sentence one. The scene is two sentences, the only scene allowed two, so that sentence two is figure free and separates the memory figure from the money amount in scene 2. NOT NARRATED: the comma list "NVIDIA Grace Blackwell compute, unified memory, NVIDIA ConnectX-7 networking and an NVIDIA CUDA-accelerated AI software stack" is collapsed out of narration entirely, because every comma buys a real pause and this is a four item list; the 100-billion-parameter support figure is unspoken and carried in the title candidates instead. REFUSED: "Personal AI Supercomputing" and "The compact, personal AI supercomputer" are vendor promotional descriptions and are never spoken in Inboxsmith's voice. "ready to use from day one" and "The new DGX Spark 64GB configuration supports practical work from day one" are marketing framing and are refused. The page's adjective "agentic" is barred from narration in every form, and this scene speaks no form of it.(SCENE 1 BACKING. Same first-party NVIDIA blog post, the section headed "A New Starting Point for Personal AI Supercomputing", which describes the hardware content of the new configuration and compares it to the existing larger one. Vendor spec description, with no teardown, no reviewer and no independent measurement anywhere on the page.)
• Backing sentences, quoted exactly: "DGX Spark 64GB is available from Acer, ASUS, Dell, Gigabyte, HP and MSI on Friday, Oct. 23, starting at $4,999." and the deck line "Starting Friday, Oct. 23, the new configuration will be available from Acer, ASUS, Dell, Gigabyte, HP and MSI". Also quoted: "Coming this month, NVIDIA DGX Spark will be available with 64GB of unified memory from top manufacturer partners". EDITORIAL DECISION: availability verbs are honest here because this is a real product launch with a named date and a published price, but availability is FUTURE, so narration says "It arrives this month", which is literally true on an October 2 publication date for an October 23 on-sale date. No countdown phrasing is used anywhere and the word expires appears nowhere in the file, because a downstream validator hard-fails countdown phrases. The exact date October 23 is not spoken, which avoids a third numeric figure landing in the same sentence as the money amount, and it is carried on the silent card as "Oct. 23". The money amount is written and spoken as words, "four thousand nine hundred ninety nine dollars", with no currency symbol and no digits in narration, and it is the only money amount spoken in the whole script. The six manufacturer partners are a six item comma list in the source and are collapsed out of narration; only two are named out loud, "Dell and HP", joined with "and" so the read carries no list pause, and the complete list is carried silently on the scene card where it costs nothing. The word "starting" is retained because the source says "starting at", so the price is not presented as the only price. REFUSED: "keeps the platform at an accessible price point" is vendor value framing and is not narrated, and no claim is made that this configuration is cheaper than the 128GB model, because the page never says that in those terms.(SCENE 2 BACKING. Same first-party NVIDIA blog post, the deck line and the closing section headed "Get Started With DGX Spark", which carry the dated availability and the starting price. First-party vendor availability and pricing statement, not a retailer listing and not confirmed by any of the six named manufacturers on the page.)
• Backing sentences, quoted exactly: "And when workloads grow, two units can cluster together via NVIDIA Sync Cluster Assistant without any additional setup.", "Two 64GB units clustered together don't just double the memory.", and "Plus, two units can connect directly with a QSFP cable, pooling their memory to 128GB and expanding model support to up to 200 billion parameters while delivering twice the memory bandwidth and up to 1.7x the performance." EDITORIAL DECISION: the whole claim is attributed out loud as "NVIDIA says", because the clustering speed result is a vendor performance claim and is never stated in Inboxsmith's own voice. The scene speaks NO figure at all. That is deliberate on two counts: the money amount in scene 2 immediately precedes this scene, and two numeric figures in adjacent sentences drive the render audit transcriber into a repetition loop that no re-roll fixes; and every figure in this passage is a banned glued brand or digit token or a multiplier the voice engine mangles. So "128GB", "200 billion parameters", "twice the memory bandwidth" and "1.7x" are all held to the silent card and to this record, and even the count "two" is avoided in narration by saying "a pair of them", which is literal and keeps the sentence figure free. The claim is reduced to its direction only, "pool memory and run larger models faster", which is exactly what the source asserts. The product names "NVIDIA Sync Cluster Assistant", "ConnectX-7" and "QSFP" are kept off mic entirely. REFUSED: "without any additional setup", "Every DGX Spark ships with a built-in NVIDIA ConnectX-7 NIC right out of the box", "designed to seamlessly scale multi-node clusters", "The NVIDIA Sync app configures this multi-node cluster seamlessly", "so developers can focus on their work rather than the infrastructure", "makes running local AI as simple as clicking a few buttons" and "Developers can go from power-on to running models in minutes" are all promotional ease-of-use framing and none of it is narrated. "with room to keep scaling as workloads demand" is open-ended vendor framing and is refused. "don't just double the memory" is a rhetorical construction and is not narrated in that form.(SCENE 3 BACKING. Same first-party NVIDIA blog post, the section headed "Scale Up With NVIDIA Sync Cluster Assistant" plus the clustering line in the opening section. Vendor capability and performance claim about its own product, with no independent benchmark and no named tester on the page.)
• Backing sentence, quoted exactly: "In NVIDIA's Qwen 3.8 27B test, two clustered 64 GB systems delivered up to 1.7x performance compared with a single system, with room to keep scaling as workloads demand." Confirmed by reading the whole document as captured on October 2, 2026: that sentence attributes the test to NVIDIA by name and is the only performance measurement on the page; the memory, parameter-count, bandwidth, price and availability figures are all NVIDIA's own with no third party cited; and no customer, no named user, no partner statement, no analyst comment and no independent benchmark appears anywhere on the page. The only named person on the page is the NVIDIA byline, Allen Bourgoyne. EDITORIAL DECISION: scene 4 is spent entirely on who produced the numbers rather than on another vendor claim, because that is the central editorial fact of a vendor-announces-own-hardware story. The house disclosure wording is kept plain and is not reworded into anything clever: the figures are NVIDIA's own, the clustering result comes from NVIDIA's own internal test, and the page shows no customer and no independent benchmark. The scene speaks no figure, which keeps the script's two spoken figures away from the end of the read. The test model name "Qwen 3.8 27B" is kept off mic as a glued brand and digit token and is referred to only as "its own internal test", and the multiplier "1.7x" is not spoken here either. This scene is the longest at twenty five words and the longest planning duration, because the disclosure is the point of the Short.(SCENE 4 BACKING, MANDATORY DISCLOSURE. The page itself is the evidence. A first-party NVIDIA blog post about NVIDIA's own hardware, carrying spec, clustering performance, availability and price claims, with no customer, no named user, no outside quote and no independent benchmark anywhere on it. The single performance result on the page is explicitly NVIDIA's own internal test.)
• No source sentence backs the endcard. It is the standard Inboxsmith pitch, "Inboxsmith builds AI receptionists for small business. Please like and subscribe for more news." EDITORIAL DECISION: the endcard carries no capability, availability, performance or price claim about NVIDIA, DGX Spark or any manufacturer partner, so no vendor assertion is spoken at the point where narration stops attributing. It contains no figure of any kind and no product name of any kind, which keeps the money amount in scene 2 far away from the closing sentences and prevents two figures landing in adjacent sentences at the end of the read. No brand and digit token, no calendar date and no price appears in the endcard. The required closing sentence, "Please like and subscribe for more news.", is present verbatim including the period and is the final sentence of the whole script.(ENDCARD. Inboxsmith house sign off. Carries no claim from the source and is deliberately not traceable to the article.)
• No corroboration of any kind was obtainable, so this single NVIDIA blog page is the sole source for every narrated claim. Recorded plainly: the YouTube transcript pull returned 0 of 5, because the yt-dlp build installed on this box is too old to fetch captions. That is a known local tooling limitation and not bot blocking, and it says nothing about the story itself. transcripts.json for this slug holds an empty videos array, so not one sentence of transcript text was available to check anything against. sources.json for this slug holds an empty sources array, which is normal for this pipeline and is not a red flag. There is therefore no independent benchmark, no reviewer, no second publisher and no analyst anywhere in the evidence set, and nothing in the script is stated as fact in Inboxsmith's own voice except the provenance disclosure itself. Dated timeline probes, eleven of them, were almost entirely unrelated consumer GPU noise about the RTX 5090 and open-weight laptop models. Three probes do touch DGX Spark as an existing product, dated June 16, 2026, September 1, 2026 and September 2, 2026, which confirms DGX Spark is not new; but none of them mentions a 64GB configuration, a price of four thousand nine hundred ninety nine dollars, an October 23 date or NVIDIA Sync, so there is no evidence this specific configuration was announced earlier and the same-day tier is honest. This is not a re-promotion of an older release: the new thing is the memory configuration, the dated availability and the price, and the hook says exactly that. Those probe titles are weak existence evidence only and not one of them is narrated as fact; in particular the figures and comparisons in the probe titles, including a Mac Studio comparison and a six thousand dollar price mentioned for an earlier DGX Spark, are absent from the first-party page and are excluded from the script entirely. No competitor and no other channel is named anywhere in the narration. SECURITY NOTE: the article text was treated as untrusted data and used for facts only. It contains no instruction directed at the author; the closest thing to imperative language is ordinary reader-facing promotional copy, specifically the "To get started:" download steps, "For agentic AI playbooks on DGX Spark, visit ... build.nvidia.com", "Sign up for the RTX Spark newsletter", "Follow NVIDIA RTX Spark on X, Instagram, TikTok and Facebook", "Register Now" for GTC Berlin and "Share This Article". All of that is page furniture aimed at NVIDIA's readers, none of it was followed, and none of it is narrated. The unrelated #ICYMI items at the foot of the page, covering NVIDIA RTX Spark Windows PCs and Alibaba's Qwen-Image-2.1, are a different announcement set and are excluded from the script entirely.(CORROBORATION STATUS: NONE)