Shulou(Shulou.com)11/24 Report--

According to an internal memo on April 25, Meta CEO Mark Zuckerberg (Mark Zuckerberg) convened a team of executives to analyze the company's computing power, especially its ability to deal with cutting-edge artificial intelligence, for five hours in the late summer of 2022.

The memo points out that despite Meta's high-profile investment in artificial intelligence research and the company's increasing reliance on artificial intelligence to support its growth, the social media giant is slow to use expensive artificial intelligence to optimize hardware and software systems in its main business, hampering its ability to keep up with innovation as it expands. To support artificial intelligence, Meta needs to "fundamentally change our physical infrastructure design, software systems, and ways to provide a stable platform."

According to the company, the restructuring increased Meta's capital expenditure by approximately $4 billion per quarter (CTOnews.com Note: currently about RMB 27.6 billion), almost double the level in 2021, and led to the suspension or cancellation of plans to build data centers in four locations.

Meta, which is facing severe financial difficulties, has been cutting jobs on an unprecedented scale since November.

At the same time, the emergence of ChatGPT in November last year triggered competition among tech giants, which released generative AI products. According to five sources, generative AI requires a lot of computing power, which aggravates the urgency of Meta expansion.

Sources revealed that the slow application of GPU in artificial intelligence by Meta is one of the main problems. GPU chips are well suited for artificial intelligence processing because they can perform a large number of tasks at the same time, reducing the time required to process billions of pieces of data. However, according to sources, GPU chips are expensive and chipmaker Nvidia controls 80 per cent of the market and maintains a leading position in the corresponding software.

Until last year, Meta mainly used a large number of normal CPU to run AI workloads. CPU is the main chip in the computer world. Although it has dominated the data center for decades, it does not perform well in artificial intelligence.

This has led competitors to outgrow Meta in the AI field. They use GPU chips and have better AI software, so they can develop new AI products and services more quickly.

According to two sources, Meta also began to use its own custom chips designed internally to train AI. But by 2021, this two-pronged approach has proved to be slower and less efficient than the GPU-centric approach. GPU chips are also more flexible than Meta chips in running different types of models, the two sources said.

Later, as Zuckerberg turned the company to meta-universe, the lack of computing power made it impossible for the company to cope with threats, including the rise of TikTok and Apple-led changes in advertising privacy.

These issues have attracted the attention of former Meta board member Peter Thiel (Peter Thiel). At the beginning of 2022, he resigned without explaining why. At a board meeting before he left office, Mr. Thiel pointed out that Mr. Zuckerberg and his executives were too focused on meta-universe at the expense of Meta's core social media business, leaving the company vulnerable to competitors such as TikTok, according to two people familiar with the matter.

Meta had planned to launch custom chips in 2022, but gave up and ordered billions of dollars of Nvidia GPU chips in the same year. Meta has lagged behind peers like Google, which began deploying its own customized version of GPU, called TPU, in 2015.

Meta then began to restructure its artificial intelligence department, appointing two new engineer leaders. During that time, dozens of executives left Meta and almost all of them changed the leadership of artificial intelligence infrastructure.

Next, Meta began to transform its data center to accommodate the introduction of GPU, these chips need more power and generate more heat, must be closely gathered together, and make a special network connection between them. This work requires a large amount of network capacity and new liquid cooling systems to manage the heat of the cluster, so they need to be "completely redesigned".

As the work progressed, Meta began internal plans to develop a more ambitious new chip, similar to GPU, that could train artificial intelligence models and reason. The project will be completed around 2025, according to two sources.

Jon Carvill, a spokesman for Meta, declined to comment on the chip project.

While Meta is expanding its GPU, companies such as Microsoft and Google are promoting business-generated artificial intelligence products, and Meta hasn't made much progress in this regard.

Meta's chief financial officer admitted in February that the company is not currently devoting most of its computing power to generation jobs. "basically all of our artificial intelligence capabilities are used in advertising, dynamic messaging and Reels," she said. " Reels is Meta's short video similar to TikTok, which is popular with young users.

According to four sources, Meta did not give priority to developing generative artificial intelligence products until the launch of ChatGPT in November last year. Although the company's AI research department has been releasing technology prototypes since the end of 2021, it has not focused on turning them into products. However, with investor interest rising, Zuckerberg announced in February the formation of a new high-level generative artificial intelligence team, which he said would "accelerate" the company's work in this area.

Chief technology officer Andrew Bosworth also said this month that generative artificial intelligence is the area where he and Zuckerberg spend the most time, and predicts that Meta will launch new products this year.

According to two people familiar with the new team, the team's work is at an early stage, mainly building the basic model, a core program that can be fine-tuned and adapted to different products later.

Meta spokesman Carvill said the company has been developing generative artificial intelligence products on different teams for more than a year. He confirmed that the work had accelerated in the months after the launch of ChatGPT.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.