Meta has taken another step in its artificial intelligence strategy by extending its collaboration with Broadcom to develop custom chips, a move that reinforces its commitment to having its own hardware on a large scale to power its AI services.
The agreement, which extends the relationship between the two companies until 2029 , includes an initial deployment of more than one gigawatt of computing capacity dedicated to training and inference, with a roadmap that aims for a deployment of several gigawatts in the coming years.
A long-term agreement for MTIA chips
As both companies have detailed, the expansion of the agreement focuses on the Meta Training and Inference Accelerator (MTIA) program , the family of internal accelerators that Meta is using to strengthen its data center infrastructure specifically geared towards AI workloads.
The initial commitment exceeds one gigawatt of computing power, a quantity comparable to the electricity consumption of hundreds of thousands of homes , which will be deployed across training and inference clusters distributed throughout Meta's global network of data centers. The company has already indicated that this is not a one-off project, but rather the first phase of a sustained deployment toward several gigawatts.
Within this program, the first chip, the MTIA 300 , is already being used in content classification and recommendation systems, while Meta plans to incorporate up to three additional generations of MTIA accelerators by 2027, with versions specifically optimized for inference tasks, that is, for the phase in which models generate answers to user queries.
Broadcom, for its part, has emphasized that Meta's MTIAs will be the first AI silicon manufactured using a 2-nanometer process , a node leap that should translate into higher transistor density and better energy efficiency, critical factors when talking about large-scale data centers and sustained power consumption.
Networks, advanced packaging, and a massive computing base
Beyond the chip design itself, the alliance encompasses Broadcom's advanced packaging, interconnection, and Ethernet networking technology —key elements for uniting thousands of accelerators in large AI clusters. Meta will use Broadcom's networking technology to connect its rapidly growing AI data centers .
Mark Zuckerberg, co-founder and CEO of Meta, explained that this collaboration with Broadcom will allow them to "build the massive computing foundation we need to deliver personal superintelligence" to a potential user base of billions. This statement encapsulates the company's ambition: to equip itself with its own infrastructure capable of supporting increasingly demanding AI models.
This investment in custom silicon comes alongside a significant increase in Meta's infrastructure spending . The company has budgeted up to $135.000 billion for AI in the near term, an effort that includes both purchasing GPUs from third parties (such as Nvidia and AMD) and deploying internally developed chips.
In terms of physical capacity, Meta already operates more than thirty data centers spread across various regions, with a significant concentration in the United States but with a global impact, including Europe, where the company operates infrastructure serving the European market. The need to coordinate chip design, packaging, networks, and power supply on a large scale is one of the reasons why it has partnered with companies like Broadcom.
Hock Tan's departure from the Meta board and new advisory role
The announcement of the new agreement was accompanied by changes in Meta's governance. The company stated that Hock Tan, Broadcom's CEO , has decided not to seek re-election to Meta's board of directors, which he joined in 2024.
Although he is formally leaving his seat on the board, Tan will move into an advisory role focused on Meta's custom chip strategy , working more directly on defining the silicon roadmap and key computing infrastructure decisions.
This move is interpreted in the market as a further step in deepening the strategic collaboration between both companies: Broadcom will not only be a hardware supplier, but a key partner in the design of Meta's computing architecture in the medium and long term.
At the same time, Meta has indicated that Tracey Travis , a board member since 2020 and former chief financial officer of Estée Lauder, will also not be seeking re-election at the next annual shareholders' meeting. The simultaneous departure of two board members adds another dimension of change to the company's corporate governance structure.
Fierce competition for AI capabilities and less dependence on Nvidia
The context of the agreement is a global race to secure AI computing power . Large tech companies like Meta, Google, and Amazon are pushing to design their own chips to reduce their reliance on Nvidia and AMD processors , which remain the de facto standard for many AI workloads but are expensive and in limited supply.
In this environment , application-specific integrated circuits (ASICs) , such as Meta's MTIA or Google's TPUs, have gained prominence . Compared to general-purpose GPUs, these chips are designed for a more limited set of tasks, which can make them more efficient and cost-competitive for specific uses, although they sacrifice flexibility.
Meta unveiled its first MTIA chips in 2023 and has since been strengthening its internal product line , in parallel with giants like Google and Amazon, which have also deployed their own AI processors. The difference is that, for now, Meta is targeting its chips for internal use, while Google and Amazon are selling some of their silicon through their cloud computing platforms.
The expansion of the agreement with Broadcom comes shortly after Broadcom itself announced a long-term contract with Alphabet (Google) for the production of its TPUs and network solutions, as well as an agreement that will allow AI developer Anthropic to access several gigawatts of computing capacity based on this hardware starting in 2027. Together, these alliances reinforce Broadcom's role as a central cog in the AI ​​hardware supply chain.
Impact on Broadcom and market analysis
In the financial markets, the news of the new agreement with Meta resulted in moderate gains in Broadcom shares , which advanced around 3-3,5% in after-hours trading, while Meta shares barely moved.
The semiconductor company has experienced several consecutive quarters of strong growth driven by demand for generative AI chips . In its last fiscal quarter, AI-related revenue doubled year-over-year to several billion dollars and now represents a significant portion of its total revenue.
Industry analysts expect that, thanks to long-term contracts like those signed with Meta and Alphabet, Broadcom's AI revenue will easily exceed tens of billions of dollars in the coming years. Forecasts indicate that the role of artificial intelligence in its business will continue to grow well into the next decade.
In Europe and Spain, where investors and fund managers closely monitor the evolution of the US technology sector, Broadcom has established itself as one of the benchmark stocks linked to the expansion of AI . The visibility provided by these contracts—some of which extend until 2031—is a factor that analysts highlight in contrast to the typical volatility of the semiconductor market.
A firm commitment to its own AI infrastructure
For Meta, the extension of the agreement with Broadcom is a clear sign of continuity in its commitment to MTIA , at a time when the market was wondering whether the company would maintain its roadmap of in-house chips or backtrack in the face of high costs and technical complexity.
Far from slowing down, Meta has confirmed that the next generation of accelerators will scale to several gigawatts of capacity starting in 2027 , and that the deployment of its own silicon will coexist with significant purchases of third-party GPUs. The strategy is hybrid: combining in-house hardware tailored to its needs with large volumes of standard processors to maintain flexibility.
This approach has direct implications for the availability of AI services and tools worldwide, including the European Union and the Spanish market , where Meta's products and functions depend on a global data and computing infrastructure that is largely built on these types of agreements.
In a landscape where European regulations on data, privacy and the use of artificial intelligence are becoming increasingly demanding, having its own more efficient and controlled AI infrastructure also allows Meta to adapt more quickly to regulatory and security requirements, something that is of particular concern to EU regulators.
By extending its alliance with Broadcom until 2029, Meta consolidates a relationship that combines chip design, networking, and strategic advice around its AI infrastructure, reinforcing its position in the race for advanced computing and sending a signal to the market that its MTIA program is not only still going strong, but is preparing to scale to another level of power and reach.
