Nvidia doubles RTX PRO 6000 Blackwell price to $16,000
The AI buildout's memory shortage keeps hitting every tier of hardware — this time it is the workstation GPU on your desk, not the data center. Nvidia quietly raised its flagship Blackwell card to $16,000, and a new report shows the human cost behind robot training data in India.
Nvidia has doubled the price of its RTX PRO 6000 Blackwell workstation GPU to $16,000 on its US Marketplace, up from $13,250 in June — the card that started pre-orders below $8,000 last year now lists at 87% above its original MSRP. The 96GB flagship, announced in March 2025 as Nvidia's biggest Blackwell workstation part, has climbed in steps: roughly $8,000 at launch, $13,250 two months ago, and now $16,000 — a 20.8% jump in just two months, per VideoCardz. Tom's Hardware flagged the change as another silent increase on Nvidia's own store page, with no announcement attached.
The driver is the same memory crunch squeezing the whole industry: VRAM and NAND prices have surged as AI demand for high-bandwidth and high-capacity memory outpaces supply, and a 96GB workstation card is where that shortage lands hardest. The pattern is broader than one SKU — consumer cards, datacenter accelerators, and now professional GPUs are all repricing on memory costs, even as the market digests record AI infrastructure spending elsewhere.
Why it matters: this is the clearest signal yet that the AI compute crunch has reached the desktop. For individual researchers, startups, and hobbyists running local inference or fine-tuning, the price of admission to serious on-prem AI work has doubled in a year — and the trend points one direction as long as memory supply stays tight. It also feeds the circular-financing concern now circulating in the industry: if GPU prices keep inflating, the collateral value behind AI compute loans only gets more volatile.
A new Bloomberg report looks at workers in India who are paid extra to wear devices that capture first-person video of factory and other work tasks — footage that becomes training data for AI robots. The practice, known as egocentric data collection, has thousands of Indian homemakers and factory workers filming everything from slicing mangoes to folding towels with head-mounted cameras, per earlier reporting from AFP and Al Jazeera. Startups like Objectways run dedicated studios in Tamil Nadu's Karur district, while Silicon Valley's Human Archive pays gig workers through India's services economy to capture synchronized video, motion, and tactile data for physical AI labs.
Why it matters: the humanoid-robot boom runs on human demonstration data, and the cheapest way to produce it at scale is low-wage labor wearing cameras. It is the data-labeling economy of the 2010s reborn for embodied AI — and a reminder that every "autonomous" robot fleet is, at training time, deeply dependent on people.
What to watch: whether memory prices ease as new fabs come online, and whether egocentric-data wages become a labor story of their own.
Does a $16,000 GPU price the individual researcher out of frontier AI work — or is local compute still worth it? Tell us in the comments.
Sources: Tom's Hardware · VideoCardz · WCCFTech · Techmeme · Al Jazeera · TechCrunch
