On April 10th, Google unveiled a new chip called Axion, boasting powerful capabilities suitable for a range of tasks from precise ad targeting on YouTube to complex data analysis, aimed at helping Google tackle the rising costs of artificial intelligence.
The introduction of Axion marks a significant breakthrough for Google in its path of developing proprietary chips, signaling a crucial step forward in the field of commonly used chips for data centers. Over the years, Google has been continuously exploring new computing resources, particularly dedicated chips for artificial intelligence. Since the release of ChatGPT by OpenAI at the end of 2022, igniting a new race in artificial intelligence, Google has accelerated its pace in developing proprietary chips, aiming to gain a competitive edge in the internet domain.
Industry analysts widely believe that Google's efforts in the chip sector will help reduce reliance on external suppliers while enabling it to compete with chip manufacturing giants like Intel and NVIDIA. However, Google executives hold a different view, with Amin Vahdat, Vice President of Chip Operations, stating, "We don't see this as competition but as an opportunity to grow the market pie."
With the rapid development of artificial intelligence technology, the demand for computing resources is increasing. Google's cloud computing rivals like Amazon and Microsoft are also intensifying their efforts in developing proprietary chip technology.
Google's early success partly stems from investments in critical chip technology, providing a solid foundation for its network search algorithms. This often involves innovative combinations of low-cost commercial hardware. Faced with the rapid development of artificial intelligence technology and the huge demand for computing resources, Google is shifting towards developing more specialized and efficient custom chip solutions. The company believes that its self-developed artificial intelligence-specific chip, the Tensor Processing Unit (TPU), demonstrates significant cost advantages.
Since 2016, Google has been closely collaborating with semiconductor giant Broadcom to jointly produce custom hardware products. In a recent internal report, Broadcom CEO Chen Fuyang revealed that with Google rapidly increasing TPU production, Broadcom's custom chip division has seen a surge in business. Chen Fuyang stated that Microsoft's integration of artificial intelligence capabilities into its Bing search engine, directly challenging Google's core position in the search field, has driven the sharp growth of Broadcom's custom chip business. It is reported that the operating profit of Broadcom's custom chip division exceeded $1 billion in the last quarter, with most of the revenue coming from Google's orders.
Google's Chief Financial Officer, Ruth Porat, projected a significant increase in spending on technology infrastructure such as artificial intelligence chips at the beginning of the year. She revealed that Alphabet, Google's parent company, saw a nearly 50% year-on-year increase in capital expenditures in the fourth quarter, reaching $11 billion.
As a CPU developed by Google, Axion's broad applicability not only powers Google's search engine but also supports various artificial intelligence-related functions. Company executives stated that Axion will play a crucial role in the field of artificial intelligence by efficiently processing massive amounts of data to serve billions of users worldwide.
It is worth noting that Axion is based on the architecture of British chip design company Arm, making Google the third major tech company to adopt Arm architecture-designed data center CPUs after Amazon and Microsoft. This shift breaks the long-standing dominance of large server operators relying almost exclusively on CPUs supplied by Intel and AMD.
In response to market changes, traditional chip manufacturers are also launching CPUs with built-in artificial intelligence computing functions and independent chips designed specifically for artificial intelligence to meet the growing market demand. On Tuesday, Intel unveiled its third-generation Gaudi artificial intelligence chip, planning to start shipping to customers this year.
Although Google refuses to directly sell chips to customers for use in their data centers, this strategy increases Google's flexibility and autonomy in the semiconductor field, enabling it to compete more directly with chip manufacturers such as Intel and NVIDIA in future market competition. As the biggest beneficiary of the artificial intelligence technology boom, NVIDIA currently holds over 80% of the market share for chips used in developing and servicing artificial intelligence technology.
Vahdat pointed out that there is a fundamental difference between being an excellent hardware company, a leader in cloud services, and a global information navigator.
To adapt to market changes and meet customer demands, Google has decided to lease custom chips to cloud customers. It is reported that later this year, external customers will have the opportunity to use Google's Axion chips developed, while the latest generation of TPU chips has already been widely deployed.
In November of last year, Google announced the successful connection of over 50,000 TPUs, building unprecedentedly large-scale artificial intelligence systems and using TPUs to create Gemini, dedicated to handling user queries.
However, as Google's cloud computing business grows, balancing the demands of internal team competition with those of external customers, such as Anthropic, becomes complex, especially against the backdrop of widespread GPU supply constraints. Sources familiar with the matter revealed that due to the continued growth in demand for artificial intelligence services, some teams within Google will be unable to obtain additional computing resources this year. Facing challenges in resource allocation, Vahdat stated that Google will prioritize products and services with the fastest growth.
Google's internal chip development began in 2013, when breakthroughs were made in speech recognition technology, prompting Google to realize the significant impact widespread use of the technology would have on data center chip demand. Engineering Director Jeff Dean told the Systems Infrastructure Department that to address the challenge, the number of data center chips needed to be increased by about double. He said, "This is really the first time we deeply felt this impending issue." Years later, when Google designed the first generation of TPUs, Dean successfully convinced senior management to purchase more TPUs, which became crucial for researchers to create the Transformers software system, now the foundation of generative AI products like ChatGPT.
In attracting external clients, Google achieved some success. Despite collaborating with several well-known startups like Character, a chatbot manufacturer, and Midjourney, an image generation company, developers found it challenging to develop software for these chips.
To address this challenge, Google announced a collaboration with tech giants like NVIDIA to advance the OpenXLA software project, aiming to simplify the complex process of developing AI systems.
After Amazon pledged to invest up to $4 billion, Anthropic, one of TPU's main users, began shifting some of its AI needs to custom chips developed by Amazon last September. In response, Google promptly committed $2 billion in funding support for Anthropic and announced an expansion of their partnership.
Early last year, due to difficulties in obtaining GPUs, Assembly AI, a startup focused on speech-to-text technology, chose to build its latest tech version on TPUs. CEO Dylan Fox remarked, "In terms of availability, we are very pleased with the performance of TPUs."
Internal data from Google shows that the Axion processor offers performance improvements of up to 30% compared to the fastest Arm chips in the cloud computing market. Several clients, including Snap, plan to test this new hardware.
Mike Gualtieri, Chief Analyst at market research firm Forrester, commented, "Even if Axion only achieves half of the performance improvement Google claims, this investment is definitely worth it."
He pointed out that Google faces fierce competition from other major cloud computing companies, in a battle for network services among super-scale enterprises, with each participant striving for upstream dominance.
