Flatkey Launches Flatkey Compute, a NeoCloud for Dedicated NVIDIA B300 Clusters - AiThority

Flatkey Compute signals a structural shift in compute allocation power, not just another cloud launch. Technically, B300's 1400W+ per-node power makes liquid cooling mandatory, while the dedicated model strips multi-tenant scheduling overhead, cutting inference latency 15-20%. Upstream, TSMC (Taiwan, China) 3nm allocation tilts toward B300; downstream, CUDA lock-in deepens, compressing AMD MI400 and Intel Gaudi 3's replacement window. Compliance-wise, BIS controls on B300 exceed B200's, structurally locking the customer base to North America, the Middle East, and Southeast Asia. In geopolitical stress, the asset-light NeoCloud model is more fragile than hyperscalers on Taiwan, China fab delivery risk. Competitively, CoreWeave holds B200 long-term contracts; Lambda and Crusoe will likely follow the dedicated-B300 narrative within two quarters. Flatkey's edge is architectural nativity. But if AWS or Azure opens bare-metal B300 within 12 months, NeoCloud premiums erode fast. 18-month outlook: expect 2-3 M&A consolidations. Survivors bind to 1-2 frontier LLM customers, forming compute-model symbioses. The 36-month B300 depreciation cycle is the cash-flow watershed—operators missing utilization targets get abandoned by capital.