{
  "id": 8169216,
  "title": "Full speed ahead: Despite calls to slow AI down, its support structure is on the fast track",
  "url": "https://urgent.news/2026/09/18/full-speed-ahead-despite-calls-to-slow-ai-down-its-support-structure",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-18T03:38:59.000Z",
  "source": {
    "name": "SiliconANGLE",
    "slug": "siliconangle",
    "url": "https://siliconangle.com/2026/09/17/full-speed-ahead-despite-calls-to-slow-ai-down-its-support-structure-is-on-the-fast-track/"
  },
  "original_language": "en",
  "account": "At Salesforce Inc.'s Dreamforce event in San Francisco, debates over slowing down the deployment of artificial intelligence (AI) took place, while a separate group of tech experts met in Santa Clara to discuss accelerating AI infrastructure development. Companies like Amazon Web Services, Oracle, Broadcom, Qualcomm, and d-Matrix were among those outlining their efforts to optimize AI infrastructure to meet the growing demands of the technology.\n\nAI systems are notoriously power-hungry and memory-intensive, with token costs soaring and leaving many enterprises struggling. Tony Pialis, Qualcomm's executive vice president and general manager of datacenter and AI, emphasized that \"tokens per watt\" has become the new key metric, urging the industry to improve infrastructure. The rapid demand for AI has led to a diverse range of computing solutions appearing in a short time, with central processing units gaining prominence in inference workloads.\n\nAmazon Web Services, an industry leader, is focusing on its Graviton CPU portfolio to meet the increasing inference workload demand. AWS' strategy is centered around Graviton5, a CPU designed to power real-time AI reasoning and multistep task orchestration. By using Graviton, the majority of AWS workloads run cost-effectively and faster than before, with AI expected to unlock even more specialized software applications.\n\nMemory has emerged as a critical component in AI processing, with AI servers using roughly eight times more memory than traditional servers. The cost of AI server memory is projected to increase fivefold from $35 billion in 2025 to between $175 billion and $190 billion by 2027. Qualcomm has responded to this challenge by shifting its focus from traditional high-bandwidth memory (HBM) to a new architecture called high-bandwidth compute (HBC). HBC provides six times the bandwidth per watt compared to HBM and 200 times the capacity per watt compared to SRAM solutions, effectively bringing compute closer to memory and overcoming the memory wall.\n\nAn alternative approach to memory challenges comes from d-Matrix Corp., which incorporates higher-throughput 3D DRAM into its next-generation chip architecture called Raptor. This solution offers higher storage density and improved performance compared to other options like SRAM. d-Matrix has also partnered with Nvidia to integrate Raptor into its rack reference architecture, NVLink Fusion, targeting ultra-low-latency premium-level token services for AI labs, hyperscalers, and neoclouds.\n\nNetworking has become an essential aspect of AI infrastructure, with Oracle focusing on improving performance, scalability, and cost through high-performance network virtualization architecture and converged SmartNIC technology called Acceleron. By collaborating with Nvidia and AMD, Oracle has developed Acceleron RoCE, which reduces data movement through central processing units and boosts performance. Broadcom has also played a significant role in shaping networking tech for AI deployment, using Ethernet technology to create a network architecture specifically designed for AI applications.",
  "summary": "While tech titans at Salesforce Inc.’s Dreamforce event in San Francisco this week engaged in a spirited debate over whether the pace of deployment for artificial intelligence should be slowed, a group of high-powered tech experts were meeting at the same time about an hour’s drive south to describe how they were building AI’s infrastructure […] The post Full speed ahead: Despite calls to slow AI…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}