bibibi.ccLoading…
Connect scattered updates into a story you can understand.
Vitalik Buterin stated that capital expenditure on Bitcoin mining equipment and electricity costs each account for approximately half, while AI inference costs are almost entirely dominated by capital expenditure on equipment.
A reposted article from the PANews channel states that Pearl the public blockchain uses GPU matrix operations to serve both AI inference and mining, and argues that inference service providers have an economic advantage over pure miners.
AMD CEO Lisa Su said that agent AI workloads will drive explosive growth in the CPU market over the next several years, and characterized 2026 as the “Year of Inference.”
Microsoft reportedly plans to expand deployment of its self-developed Maia chips starting in 2027, with Maia 300 possibly exceeding 1000000 units next year.
Banbury Road released a system consisting of 32 models Kardashev-0.7, saying that joint reinforcement learning enables them to develop different specialties and can reduce inference costs and memory requirements.
BlockBeats reports that Antseed has launched a decentralized P2P AI inference network and stated that its nonprofit foundation completed a 2400000 USD token funding round, led by Spark Capital.
Renowned trader Kyle believes that the 2026 crypto market may place greater emphasis on tokens backed by fundamentals, and is optimistic about Hyperliquid, Ethena,AI inference, and RWA and other directions.
Prime Intellect releases the open-source model inference service Prime Inference, with an API compatible with OpenAI API, and the first publicly deployed model is GLM-5.3. The company says the service processes nearly 1000000000000 tokens。
GMI Cloud announced the completion of 668000000 USD financing, including 223000000 USD Series B equity financing and 445000000 USD credit financing;ARCHIV led by the investment firm, with NVIDIA participating.
A SemiAnalysis report says MoE inference efficiency depends more on memory bandwidth, scheduling, and tiered KV cache management, and forecasts that a single pipeline stage may require approximately 400 to 500GB of local high-speed memory in 2027.
