Tether announced an investment in AI computing company Tenstorrent on October 20, extending CEO Paolo Ardoino's effort to provide developers and businesses with greater control over computing infrastructure. However, a cost study released by QVAC, Tether's local AI platform, demonstrates that purchasing a laptop to run AI models locally does not provide cost savings for most users.
QVAC's October 12 analysis examined the economics of running the dense Qwen3.8-27B model on a $2,399 laptop with 32 GB of memory. Spread across four years, this hardware cost amounts to approximately $50 per month. The study estimates the model can process about 55 million tokens in a 176-hour working month, with 94% input tokens and 6% output tokens.
The study's break-even analysis reveals that users need approximately 77 million tokens of monthly usage to offset the laptop's monthly cost allocation. According to QVAC's estimates, processing 55 million tokens would generate an API bill of $35.75 if using Alibaba Cloud's pricing, placing most users below the break-even threshold.
Tenstorrent closed a $1.5 billion Series E funding round at a $6.2 billion post-money valuation on October 19. Tether's specific commitment amount was not disclosed.
Who Can Justify Local Hardware Investment
Users who already own capable devices face a different economic equation. Since the hardware purchase has already occurred, local AI does not require recovering a new laptop's full cost. Sustained users of AI agents may achieve savings when handling workloads comparable to cloud-based alternatives, particularly if using faster model configurations.
The analysis notes that QVAC's faster Qwen3.6 35B mixture-of-experts scenario can reach approximately 200 million tokens in the same 176-hour window, which could accommodate heavier coding workloads that exceed the dense model's processing capacity.
Beyond Token Pricing
Ardoino has framed local AI ownership in terms beyond pure cost comparison. Running models locally provides offline availability and reduces dependence on a service provider's continued operation. Processing prompts and files on a device can offer privacy benefits, though QVAC acknowledges that connected tools may still transmit data externally.
The study explicitly states that it assumes local and cloud versions are suitable for the same work without establishing equal output quality and excludes operating costs such as electricity. Regional pricing variations and caching discounts could alter the break-even calculation for different users.
