NVIDIA continues to lead the global artificial intelligence hardware market, with more than 90% of the graphics processing units market. Its CUDA software platform has blocked a vast developer base, which makes NVIDIA hardware the predetermined option for the training of AI in cloud suppliers, research laboratories and large business clients.
But the domain of the company now faces limits imposed by size. After crossing a market assessment of $ 4 billion, Nvidia’s growth rate is under scrutiny. Its income from the data center reached $ 39.1 billion in the most recent quarter, higher from previous years, but maintaining this rhythm will be more difficult as the mature market and competitors expand in adjacent categories.
AMD has increased its focus on the inference of AI: the deployment of models of the trained in real world applications, such as search, personalized recommendations and generative text services. One of the largest world -class development companies has adopted AMD GPUs for daily inference work loads, and the large cloud service providers have begun to incorporate AMD chips for specific AI services.
Nvidia is still the leader in AI training, but AMD’s position in inference is gaining ground as the costs and availability of chip become decisive factors for cloud operators. The AMD ROCM software platform, although not as mature as CUDA, has become enough for many inference work loads. The financial gap between the two companies highlights AMD’s growth potential. In the last quarter, AMD reported $ 3.7 billion in income of the data center, a fraction of NVIDIA, but large enough for small market share gains could translate into significant growth of income.
AMD is also generating impulse in the CPUs of the Data Center, where participation against Intel has increased. These chips handle memory, orchestration and other computer tasks that GPUs do not process directly. As Ia’s workloads expand, high -performance CPU demand will be expected to grow together with the demand for accelerators. The existing position of AMD on the server’s CPUs gives you another route to benefit from the AI ​​infrastructure cycle.
Additional development is AMD’s participation in the Uualink consortium, which is working to create an open standard for high -speed connections between AI chips. Nvlink patented Nvidia technology currently dominates this space, which forces the operators of the data center to build around Nvidia Hardware. If Uualink is successful, companies can integrate multiple suppliers processors, which would erode one of Nvidia’s key competitive advantages and give AMD new access to high performance AI groups.
Broadcom is taking a different path to the AI ​​market when focusing on the hardware that connects and admits large -scale AI systems. The company supplies Ethernet switches and optical interconnections, which move massive data volumes between processors within the data centers. As the AI ​​models grow, the networks capacity has become a limiting factor, and Broadcom has benefited from this trend. The company reported an increase of 70% in income from AI networks in its last profit period, driven by cloud operators and hyperscala data centers.
Broadcom is also expanding its custom chip business. The company designs specific integrated circuits of the application, the healthy ones, for technology companies that require optimized processors for specific workloads. These personalized chips generally offer better performance and lower energy use than general use GPUs for specific AI tasks. Broadcom contributed to the development of Google tensioner processing units and is now working with other data centers operators in large -scale personalized IA chips.
The management expects three of Broadcom’s largest custom customers clients to run up to one million groups of AI chips for fiscal year 2027. That deployment represents an estimated income opportunity between $ 60 billion and $ 90 billion, depending on the volume of final production and the implementation speed. Broadcom has also signed additional chips design agreements with new customers in the consumer technology sector, expanding its pipe beyond hyperscala data centers.
In addition, the acquisition of VMware by Broadcom has positioned the company to sell infrastructure management software that admits the implementation of AI. The VMware’s Cloud Foundation product helps large companies to administer IA workloads in private cloud data centers and public cloud environments, reducing complexity for companies that execute IA applications in mixed hardware. This complements the Broadcom hardware business by providing an integrated solution for AI clients that manage hybrid and multiple clouds configurations.
While Nvidia’s position in the AI ​​hardware remains strong, it is unlikely that the rapid growth of the company in the last two years will continue at the same rate. Its income from the data center expanded more than nine times during that period, a rhythm that is difficult to maintain as the market stabilizes.
On the contrary, AMD and Broadcom are starting from lower revenue bases in AI hardware, which makes their growth trajectories more pronounced if they continue to win traction in their respective markets. The progress of AMD in inference and CPU chips of the data center gives you a direct exposure to the growing demand for real world applications. The combination of network hardware Broadcom, custom Asic Design and virtualization software offers a multichannel approach to the AI ​​infrastructure that is aligned with the way the data centers are climbing their AI capacities.
Both companies are positioned to capture new expenses in Hardware categories from AI where NVIDIA has less direct control. For investors who look beyond the largest shares in the market, AMD and Broadcom present growth cases linked to specific changes in the way in which computer infrastructure is being built.
Also read: Nvidia reaches $ 4 billion, Dow, S&P 500, Nasdaq Post Gains, Trump issues rate notices
(Tagstotranslate) Forecision of Ambd 2025
Source link