Apple Pushes AI Server Chip Development, Targets Data Center Dominance

Apple is taking a major step into AI infrastructure together with Broadcom to develop its own first server-grade AI chip, code-named “Baltra.”

The chip is expected to power advanced artificial intelligence (AI) workloads by 2026, reports The Information, citing three people with knowledge of the project.

The initiative marks a significant step for Apple as it expands its silicon expertise from consumer devices to enterprise-level AI infrastructure, positioning itself to compete with industry leaders like Nvidia, Microsoft, and Google.

The Baltra project is a continuation of Apple’s investment in proprietary silicon and advanced networking technologies. By incorporating Broadcom’s expertise in chiplet interconnects, Apple aims to address the increasing demand for high-performance hardware in data centers.

Apple’s Baltra Chip: Designed for AI in the Data Center

Apple is building its Baltra chip to tackle the growing demands of AI workloads in enterprise environments. Broadcom’s advanced 3.5D eXtreme Dimension System in Package (XDSiP) technology plays a key role in its design, improving communication between chiplets by reducing latency and increasing data transfer speeds.

Unlike conventional 3D stacking, XDSiP arranges chiplets horizontally, creating faster and more reliable connections.

The chip is expected to be manufactured using Taiwan Semiconductor Manufacturing Coorporation’s (TSMC) N3P process. This advanced semiconductor fabrication method improves power efficiency and chip density, enabling Baltra to support resource-intensive tasks such as training large language models and performing real-time analytics.

In addition to its technical innovations, Apple’s commitment to privacy could set Baltra apart. By incorporating privacy-focused features into its AI services, Apple aims to address enterprise concerns about data security, offering a competitive edge in the crowded AI server market.

Craig Federighi, Apple’s senior vice president of software engineering, commented earlier this year at WWDC, “We’re investing in both on-device intelligence and cloud-based AI to deliver better user experiences while maintaining our commitment to privacy.” The Baltra project aligns with this dual focus, enabling Apple to scale its AI infrastructure to meet growing enterprise demands.

Tracing the Roots: The ACDC Project

Apple’s work on the Baltra chip builds upon its earlier efforts under the Apple Chips in Data Centers (ACDC) project, first reported in May 2024. The ACDC initiative focused on creating AI accelerators optimized for inferencing—a process that enables AI models to generate outputs in real time.

These efforts drew from Apple’s experience with neural processing units (NPUs) in its M-series processors. The M4 Max chip, launched in October 2024, is a prime example of Apple’s silicon capabilities.

Featuring a neural engine capable of 38 trillion operations per second (TOPS) and a memory bandwidth of 546 GB/s, the M4 Max demonstrated Apple’s ability to handle complex AI workloads at scale. Baltra builds on this foundation, aiming to bring Apple’s silicon expertise to the data center market.

Manufacturing Challenges and Industry Dynamics

Despite the ambitious scope of the Baltra project, Apple faces logistical challenges in bringing the chip to market. Foxconn, a key Apple manufacturing partner, is currently committed to producing Nvidia’s GB200 superchips, which dominate the high-performance AI server market.

These chips, which combine dual GPUs with Nvidia’s Grace CPU, are critical for training large-scale AI models and have created bottlenecks in production capacity.

Foxconn’s expanded facilities in Mexico are operating at full capacity to meet Nvidia’s demands. These commitments could delay Apple’s production timelines for Baltra, underscoring the competitive and resource-intensive nature of the AI hardware market.

The AI server market itself is growing rapidly. In 2024, global server shipments are expected to reach 13.654 million units, with AI servers accounting for 12.1% of this volume.

The market is projected to generate $187 billion in revenue by the end of the year, reflecting increased investment in AI-driven applications. Companies like Microsoft, Google, and Amazon Web Services (AWS) have established themselves as leaders, collectively spending billions on proprietary hardware and data center expansions.

Differentiating Apple’s Approach

Apple’s entry into the AI server market is distinguished by its focus on proprietary silicon design and its emphasis on privacy and security. Broadcom’s 3.5D XDSiP technology is a significant differentiator, allowing Apple to create chips that enable faster communication between components.

TSMC’s N3P process further enhances Baltra’s performance by improving power efficiency and chip density, making it ideal for high-demand workloads.

The Bigger Picture: AI-Specific Hardware as a Growth Area

The development of the Baltra chip reflects Apple’s strategic shift toward enterprise solutions, extending its influence beyond consumer devices. This move aligns with a broader trend among tech companies investing in proprietary hardware to meet the growing demands of AI applications.

By collaborating with Broadcom, Apple demonstrates its willingness to partner with industry leaders to overcome technical and logistical challenges. As the project progresses, Baltra has the potential to position Apple as a significant player in the enterprise AI market, complementing its existing strengths in consumer-focused AI capabilities.