NVIDIA might do for desktop AI what it did for desktop gaming
At CES this year, the company's usual host of announcements included a home AI 'supercomputer' that could unlock 'on prem' super intelligence for everyone
One of the highlights of the annual Consumer Electronics Show (CES) has been the NVIDIA keynote for as long as they’ve been around – it’s a much different affair from the usual efforts by Samsung, LG, Sony and others which just feel like marketing decks for home electronics buyers but performed on stage with a relatively inflated A/V budget. NVIDIA’s keynote, delivered by its charismatic CEO Jensen Huang, offers a more wide-ranging and ambitious look at the future – as powered by NVIDIA tech, of course.
This year, NVIDIA’s keynote focused on AI, which is to be expected given that company’s foundational role in underpinning much of the advancements made in the era of the LLM. Alongside new foundational world models, and a platform for helping individuals train and test the next generation of AI-powered robots, Huang also revealed ‘Project Digits,’ a new product based on its Grace Blackwell AI-specific architecture that aims to offer at-home AI processing capable of running 200 billion-parameter models locally for a projected retail cost of around $3,000.
My former colleague Kyle Wiggers has a great rundown of the specs and capabilities of the initial ‘Project Digits’ offering at TechCrunch, which NVIDIA is developing and shipping with partners (much like it does with its consumer GPUs currently – an important point of comparison). The bottom line is that it can provide unprecedented local AI compute at a – while still expensive – extremely affordable price point, relatively speaking.
Based on Huang’s statement to Kyle, the initial target market for Project Digits isn’t necessarily just your average home PC user; he specifically called out the intent of putting AI supercompute capability in the hands of “every data scientist, AI researcher and student,” so there’s an assumption that at least partial specialty and technical familiarity will be part of the ideal customer profile out the gate.
There are many exciting things about Project Digits, including the fact that two can be paired to offer 405 billion-parameter model support for ‘just’ $6,000 (again, sounds like a lot, but tiny compared to what has previously been available), and its ability to play nice with both Windows and Mac – as well as to operate independently using its built-in Linux-based DGX OS. But what’s most exciting about it is probably what it indicates in terms of NVIDIA’s strategy around consumer AI and what comes next.
If NVIDIA and its partners can deliver a capable product that lives up to its promises on Project Digits at its target price point, that will drive a lot of interest, and may start turning the flywheel for future cost reductions, along with performance and efficiency improvements that make possible a more mature product line with varying price points depending on needs. It could, in fact, look a lot like NVIDIA’s current GPU line, which offers tiers of performance at different price levels depending on the needs and budgets of its gaming customers.
These could fit neatly with different parameter-sizing in terms of model support, and NVIDIA has a well-established playbook when it comes to working with licensed hardware partners to co-brand and distribute this kind of product lineup.
The remaining challenge in terms of this growing from an interesting but minuscule side-business aimed at professional researchers and a few amateur enthusiasts lies on the software side; in GPUs, NVIDIA benefits from an extremely mature and diversified gaming market to drive demand for its products – in AI, and generative AI in particular, the user-facing software product side is less well-established. Most of the paradigms we’ve seen so far either focus on A) simple, cloud-based products like ChatGPT and Claude or B) more complex, specialist-only implementations like local installations of Stable Diffusion, Meta’s Llama or similar.
If we see a proliferation of AI products with an accessible and easy-to-learn user experience, paired with local offline backend, then we could definitely see this become an area of rapidly ramping investment and interest for NVIDIA. With Project Digits, we at least see the company looking to solve one part of that particular chicken and egg problem.