According to CoinMeta, Cerebras has released a new generation of AI inference servers, CS-4. Although the name has been upgraded from CS-3 to CS-4, the core remains the WSE-3 series chips, with no new WSE-4 being introduced. The main upgrades come from the overall design; each CS-4 integrates 3 WSE-3 Turbo, and there have been refinements to power supply, cooling, and inter-chip communication. Cerebras claims that the inference speed of CS-4 can reach up to 30 times that of traditional GPU solutions, and the throughput per unit of power consumption is up to 10 times that of CS-3. The first batch of CS-4 will be delivered in the third quarter of this year, with plans to launch the next generation of chips and servers in 2027.