经过六年的低调研发，Atomic Machines终于浮出水面，并成功筹集到了2.5亿美元。该公司称，其系统能够直接根据代码制造出可正常运作的微型机器。这家位于旧金山湾区的初创企业于周三宣布了这一消息，其首款产品是一款专为人工智能数据中心设计的电源开关。
Atomic Machines has come out of six years in stealth with $250M raised. It says its system builds working micro-machines straight from code. The Bay Area start-up announced the launch on Wednesday. Its first product is a power switch for AI data centres.
创始人兼首席执行官杰夫·霍尔登曾是亚马逊的第10位工程师，并参与创建了Amazon Prime服务。之后他又出任优步的首位首席产品官，负责组建该公司的自动驾驶技术实验室、人工智能研究部门以及空中出租车项目。
Founder and chief executive Jeff Holden joined Amazon as its tenth engineer and built Amazon Prime. He then became Uber’s first chief product officer. There, he set up its self-driving, AI labs and air-taxi units, according to the company.
“3D打印技术让各种形状变得可编程——只要有相应的文件，就能打印出任意几何形状。而我们正在让机器本身也变得可编程，”霍尔登说道。
“3D printing made shapes programmable: any geometry, straight from a file. We’re making machines programmable,” Holden said.
这家公司将其系统命名为“物质编译器”（Matter Compiler），这一名称源自尼尔·斯蒂芬森的小说《钻石时代》中的一台机器。据称，该系统完全由代码来决定要制造何种设备。它所制造的机器具备可活动的部件，由多种材料构成，其精细度可达个位数微米级别，且无需使用任何模具或固定工具。
What the Matter Compiler does The company calls its system the Matter Compiler, after a machine in Neal Stephenson’s novel The Diamond Age. It says code alone decides which device the system builds. The machines it makes have moving parts, several materials and features down to single-digit microns. No moulds, masks or other fixed tooling are needed.
Atomic Machines表示，该系统能实时测量并修正每一个制造步骤，其内置的人工智能还能直接在硬件上开展实验。该公司的研发目标包括制造出沙粒大小的齿轮以及能够进入人体内部工作的微型机器人，还有将冷却功能直接集成到人工智能芯片中的技术。
The system measures and corrects every step in real time, and its AI runs its own experiments on the hardware, Atomic Machines said. Targets include gears the size of a grain of sand and robots small enough to work inside the body. Another is cooling built into AI chips.
据该公司介绍，传统制造工艺在微米级尺寸上便已遇到瓶颈；再小的尺寸则需借助芯片制造设备来实现。而每款采用此类工艺制造的新设备（即微机电系统，MEMS）都需要耗费数年时间进行定制化工艺开发。瑞典的Silex Microsystems公司正是从事此类MEMS制造的企业。
Conventional manufacturing hits its limits at a few tens of microns, the company said. Below that, chipmaking tools take over. Each new device made with them, known as MEMS, needs years of custom process work. Sweden’s Silex Microsystems runs a MEMS foundry business of that kind.
在官网的一篇文章中，霍尔登解释了创办这家公司的初衷。他将这套系统比作核糖体——后者能够根据遗传密码来合成蛋白质。
Holden explains why he started the company in an essay on its website. He compares the system to the ribosome, which builds proteins from a genetic code.
首款产品方面，该公司推出的首款设备名为PrimeSwitch PS-150，是一款能在50微秒内完成开关动作的继电器。据称其响应速度比传统接触器快约1000倍，可承载150安培的连续电流。该产品的外形尺寸为直径9.5毫米、厚度3毫米。
The first product The first device is PrimeSwitch PS-150, a relay that opens in 50 microseconds. The company says that is about 1,000 times faster than a conventional contactor. It carries 150 amps of continuous current. The package measures 9.5mm across and 3mm thick.
这款开关专为那些正逐步采用800伏直流供电方案的人工智能数据中心而设计。据该公司介绍，在如此高的电压环境下，一旦发生故障，便可能在数微秒内释放出极具破坏性的能量；而机械保护装置则需数毫秒才能启动。
The switch is meant for AI data centres that are moving to 800-volt direct current. At those voltages, a fault can release destructive energy in microseconds, the company said. Mechanical protection takes milliseconds to open.
目前，PrimeSwitch已开始向首批体验客户发货以供测试。Atomic Machines并未透露这些客户的身份。该产品将于10月12日至15日在圣何塞举行的开放计算项目全球峰会上首次公开亮相。
PrimeSwitch is shipping to early-access customers for evaluation. Atomic Machines did not name them. It will show the switch in public for the first time at the Open Compute Project Global Summit in San Jose, from 12 to 15 October.
数据中心供电领域也吸引了众多初创企业涉足。今年8月，Emerald AI成功筹集了1.5亿美元，用于开发能够减轻电网负荷的软件。
Data centre power is drawing other start-ups too. Emerald AI raised $150m in August for software that eases the load on grids.
投资方阵容强大：OneIM、Gigafund、XTX Ventures、KAS Venture Partners、Valor Equity Partners以及Sozo Ventures均参与了投资。此外，加利福尼亚大学董事会也提供了资金支持。不过该公司并未公布自身估值，也未说明这2.5亿美元资金在各轮融资中的具体分配情况。
Who is backing it Investors include OneIM, Gigafund, XTX Ventures, KAS Venture Partners, Valor Equity Partners and Sozo Ventures. The Regents of the University of California also invested. The company did not give a valuation or say how the $250M was split across rounds.
“如果说‘氛围编程’代表了当下软件开发的趋势，那么这项技术则标志着‘氛围制造’时代的开端，”Gigafund的执行合伙人卢克·诺塞克说道。
“If vibe coding is how software gets made now, this is the start of vibe manufacturing,” said Luke Nosek, managing partner at Gigafund.
所谓“氛围编程”，即人们用通俗语言描述所需软件的功能，再由人工智能负责编写代码。这一趋势目前已催生出多家独角兽初创企业。
In vibe coding, people describe software in plain language and AI writes it. The trend has already produced start-up unicorns.