đź”’ Amazon, Google scramble to keep pace with OpenAI despite struggle to innovate

By Parmy Olson

Of all the questions that ChatGPT has raised about the future of artificial intelligence, one still reverberates through Silicon Valley: Why couldn’t the industry’s largest technology firms breed an innovative service with a similar kind of impact, especially after amassing some of the world’s largest AI teams?    

Exclusive new data from a London-based analytics startup show that the five biggest tech firms have an estimated army of 33,000 people working directly on AI research and development, with Amazon boasting the largest pool of AI-focused employees, at 10,113. Microsoft Corp. has 7,133 AI staff and Google has 4,970, according to Glass.ai, which used machine-learning technology to scrutinize tech company websites and thousands of LinkedIn profiles of their AI-focused employees. The numbers might not yet account for recently announced layoffs at Amazon, which were expected to affect AI staff, but they are also a conservative estimate, excluding software engineers who might well be working on AI, too. [1]

___STEADY_PAYWALL___

The numbers underscore how seriously the world’s biggest technology firms have been taking their work on artificial intelligence, but also how slow and cautious they have been to create services with the technology until a tiny firm, San Francisco-based OpenAI, prodded them to act. 

Read more: Economic slump impacts Apple, Alphabet, and Amazon with reduced demand

Just a few months after research lab OpenAI released ChatGPT , the chatbot became the fastest-growing online service of all time, sparking a race  between Google and Microsoft to plug generative AI into many parts of their software. (Microsoft also has a partnership with OpenAI and has invested more than $10 billion in the smaller company.) Adobe Inc., meanwhile, has unveiled an AI image generator after the success of OpenAI’s DALL-E 2, Snap Inc. recently launched a chatbot similar to ChatGPT and Facebook’s Meta Platforms Inc. is racing to build similar “AI personas.” Most of this is in response to the work of OpenAI’s tiny team of artificial intelligence experts, who number just 154 people, according to Glass.ai.

The scuttlebutt in the AI community is that OpenAI’s success really comes down to clever marketing. The company has been on a promotional tear for the last two years with glowing media coverage for earlier projects like its language model GPT-3 and DALL-E 2. But its success also comes down to the direct access its most specialized researchers have to the public. 

OpenAI isn’t a product company but a research and development lab. That means it launched ChatGPT without the battalions of engineers and product managers who would normally have their hands in the development of a product at a larger tech firm — and who inadvertently can create bottlenecks inhibiting the development of new technology.

Amazon makes an interesting alternative case study for turning AI research into successful products. Despite having the largest team of AI researchers in the industry, it has had mixed success turning its AI research into popular or innovative products. 

Nearly a decade ago, Amazon introduced Alexa to the world as an exciting new virtual assistant. But the service, available for free through the $99 Echo speaker, has been something of a flop, having drained the company of billions of dollars. Many of Alexa’s customers use it to play music or set timers and nothing more.

Read more: AI Chatbot race between Microsoft, Google is loaded with risk

Why are Alexa and other digital assistants like Siri and Google Assistant so limited when chatbots like ChatGPT are so humanlike and versatile? Because the latter tools are powered by large language models, which are trained to generate text based on huge datasets scraped from the web. Alexa is powered by a more limited command-and-control system, which is programmed to recognize certain commands like, “What time is it?”

An Amazon spokeswoman said the company disagreed that Amazon was limited in what it could do. She added that AI permeates Amazon’s operations, noting that AI staff also worked in areas like product recommendations, its cloud-computing business Amazon Web Services (AWS) and its warehouse logistics systems.

Amazon’s cloud-computing business AWS has been striking partnerships recently with some of the hottest names in machine learning, including Stability AI, which used AWS’s computing infrastructure to build and deploy its image-generating tool Stable Diffusion. In February Amazon partnered with Hugging Face, which is building a rival to ChatGPT on Amazon’s cloud-computing system.

These promising partnerships might offer a new strategic path for the company after its own internal efforts to build large-scale AI services struggled.  

On Monday, Amazon announced it was laying off 9,000 more staff in the coming weeks, adding to some 18,000 previously disclosed job cuts. But the company has been building up its AI workforce for many years, and even went on a hiring spree in the field over the last five years. Glass.ai’s data show that after 2018, Amazon pulled well ahead of Google, Microsoft, Meta Platforms Inc. and Apple to hire more staff in artificial intelligence.

It’s possible that many of Amazon’s latest layoffs will be concentrated in its struggling Alexa business. Amazon had about 10,000 staff working on Alexa back in November 2022, according to the Wall Street Journal, and while that number will likely include many sales and marketing staff, there will also be significant overlap with the 10,000 AI-focused staff that Glass.ai identified. [2]

Read more: Google’s plan to catch ChatGPT is to stuff AI into everything

The downside to OpenAI’s straight-to-market approach is that it is releasing a powerful and largely untested tool to the public. In treating its first millions of users like guinea pigs, the company also risks creating social harms it didn’t anticipate. GPT-4, its latest language model, underwent six months of safety testing before being unleashed to the wild. But it is still unclear whether or to what extent these tools will spread misinformation, upend livelihoods or be misused by bad actors. 

Amazon Chief Executive Officer Andy Jassy is now trying to streamline Amazon to make its processes similarly efficient. He may have to think more carefully about where to direct Amazon’s army of AI expertise, but maintaining some caution isn’t such a bad idea, either. 


  1. To identify AI employees, Glass.ai’s software crawled websites of technology companies and scanned the LinkedIn profiles of hundreds of thousands of employees. It shortlisted “core” AI teams by looking for the following phrases in their job titles and descriptions: AI, AI Architect, AI Developer, Artificial Intelligence, Autonomous Vehicle, Big Data, Computer Vision, Data Analyst, Data Science, Data Scientist, Deep Learning, Language Models, LLM, Machine Intelligence, Machine Learning, ML, Natural Language, Neural, NLP, Research Scientist. It excluded people working in finance, sales, marketing, HR and other “non-technical” departments, as well as more generic technical roles like “software engineer” or “software developer,” even though people with those titles could be working on AI, too.
  2. Apple, at least, is probably avoiding the chatbot race. Glass.ai’s analysis of Apple’s most recent AI recruits showed that many were focused on computer vision technology, which ties better into Apple’s work on a mixed-reality headset. Apple did not respond to a request for comment.

© 2023 Bloomberg L.P.