【译文】OpenAI声称计划自研AI芯片

原文地址：https://techcrunch.com/2023/10/06/openai-said-to-be-considering-developing-its-own-ai-chips

OpenAI, one of the best-funded AI startups in business, is exploring making its own AI chips.

OpenAI，作为商业领域资金充裕的最佳人工智能初创公司之一，正在考虑开发自己的AI芯片。

Discussions of AI chip strategies within the company have been ongoing since at least last year, according to Reuters, as the shortage of chips to train AI models worsens. OpenAI is reportedly considering a number of strategies to advance its chip ambitions, including acquiring an AI chip manufacturer or mounting an effort to design chips internally.

据路透社报道，至少自去年开始，OpenAI公司内部一直在讨论AI芯片战略，因为用于训练AI模型的芯片短缺情况加剧。OpenAI据称正在考虑多种策略来推进其芯片计划，包括收购AI芯片制造商或着手内部芯片设计的工作，以满足其雄心壮志。

OpenAI CEO Sam Altman has made the acquisition of more AI chips a top priority for the company, Reuters reports.

据路透社报道，OpenAI的CEO萨姆·奥尔特曼已将获取更多AI芯片置于公司的首要任务之中。

Currently, OpenAI, like most of its competitors, relies on GPU-based hardware to develop models such as ChatGPT, GPT-4 and DALL-E 3. GPUs’ ability to perform many computations in parallel make them well-suited to training today’s most capable AI.

目前，OpenAI像其大多数竞争对手一样，依赖基于GPU的硬件来开发模型，例如ChatGPT、GPT-4和DALL-E 3。GPU的并行计算能力使它们非常适合训练当今最先进的人工智能模型。

But the generative AI boom — a windfall for GPU makers like Nvidia — has massively strained the GPU supply chain. Microsoft is facing a shortage of the server hardware needed to run AI so severe that it might lead to service disruptions, the company warned in a summer earnings report. And Nvidia’s best-performing AI chips are reportedly sold out until 2024.

但是，生成式人工智能的蓬勃发展已经极大地压紧了GPU供应链，对于像Nvidia这样的GPU制造商来说，这是一笔巨大的财富。微软在夏季财报中警告称，他们正面临严重的服务器硬件短缺问题，可能导致服务中断。而据报道，Nvidia最高性能的AI芯片已售罄，直到2024年。

GPUs are also essential for running and serving OpenAI’s models; the company relies on clusters of GPUs in the cloud to perform customers’ workloads. But they come at a sky-high cost.

GPU对于运行和提供OpenAI的模型至关重要；该公司依赖云中的GPU集群来处理客户的工作负载。但这也伴随着极高的成本。

An analysis from Bernstein analyst Stacy Rasgon found that if ChatGPT queries grew to a tenth the scale of Google Search, it’d require roughly $48.1 billion worth of GPUs initially and about $16 billion worth of chips a year to keep operational.

来自伯恩斯坦分析师斯泰西·拉斯冈的分析发现，如果ChatGPT的查询规模增长到谷歌搜索的十分之一，就需要大约480.1亿美元的GPU初始投资以及每年约160亿美元的芯片运营费用。

OpenAI wouldn’t be the first to pursue creating its own AI chips.

OpenAI并非首家追求自主研发AI芯片的公司。

Google has a processor, the TPU (short for “tensor processing unit”), to train large generative AI systems like PaLM-2 and Imagen. Amazon offers proprietary chips to AWS customers both for training (Trainium) and inferencing (Inferentia). And Microsoft, reportedly, is working with AMD to develop an in-house AI chip called Athena, which OpenAI is said to be testing.

谷歌拥有一款处理器，名为TPU（缩写为“张量处理单元”），用于训练大规模生成式AI系统，如PaLM-2和Imagen。亚马逊为AWS客户提供专有芯片，用于训练（Trainium）和推理（Inferentia）。而据报道，微软正在与AMD合作开发一款名为Athena的内部AI芯片，OpenAI据称正在进行测试。

Certainly, OpenAI is in a strong position to invest heavily in R&D. The company, which has raised more than $11 billion in venture capital, is nearing $1 billion in annual revenue. And it’s considering a share sale that could see its secondary-market valuation soar to $90 billion, according to a recent Wall Street Journal report.

当然，OpenAI有充分的资金来大力投资研发（研究与开发）。这家公司已经获得了超过110亿美元的风险投资，年度收入接近10亿美元。根据最近的《华尔街日报》报道，该公司正考虑进行股权出售，这可能会使其在二级市场的估值飙升至900亿美元。

But hardware is an unforgiving business — particularly AI chips.

但硬件是一个不容忽视的领域，特别是在AI芯片方面。

Last year, AI chipmaker Graphcore, which allegedly had its valuation slashed by $1 billion after a deal with Microsoft fell through, said that it was planning to job cuts due to the “extremely challenging” macroeconomic environment. (The situation grew more dire over the past few months as Graphcore reported falling revenue and increased losses.) Meanwhile, Habana Labs, the Intel-owned AI chip company, laid off an estimated 10% of its workforce. And Meta’s custom AI chip efforts have been beset with issues, leading the company to scrap some its experimental hardware.

去年，据称与微软的交易未能达成导致估值下调了10亿美元的AI芯片制造商Graphcore表示，由于“极具挑战的”宏观经济环境，他们计划进行员工裁减。在过去几个月里，Graphcore报告了营收下滑和亏损增加的情况，局势变得更加严峻。与此同时，英特尔旗下的AI芯片公司Habana Labs裁减了估计10%的员工。而Meta（前Facebook）的自有AI芯片项目也遇到了一些问题，导致该公司取消了一些实验性的硬件项目。

Even if OpenAI commits to bringing a custom chip to market, such an effort could take years and cost hundreds of millions of dollars annually. It remains to be seen if the startup’s investors, one of which is Microsoft, have the appetite for such a risky bet.

即使OpenAI决定推出自家的芯片，这样的努力可能需要数年时间，并且每年成本可能高达数亿美元。目前尚不清楚这家初创公司的投资者，其中之一是微软，是否愿意承担这样高风险的赌注。