Trump says Iran war "close to over" amid hopes for more negotiations
Investing.com -- Apple is making a significant bet on an advanced chip packaging technology, System on Integrated Circuit (SoIC), as it builds out the infrastructure for its artificial intelligence ambitions.
SoIC is TSMC’s advanced 3D packaging platform that enables high-performance chiplet designs.
Until recently, AMD was the technology’s single largest customer, using it across its MI-series AI GPUs and high-end gaming processors, and now Apple is also moving aggressively into that space.
According to Morgan Stanley, Apple has placed orders equivalent to 36,000 wafers at TSMC for 2026 and 60,000 wafers for 2027. The scale of those commitments points away from Apple’s existing Mac product line and toward something larger.
Apple shipped roughly 80,000 Mac Pro and Studio units powered by M2 and M3 Ultra processors in 2025, which, at an estimated 50 Ultra processors per wafer, would account for at most around 1,600 wafers of SoIC demand — a fraction of what Apple is now reserving.
Morgan Stanley analyst Erik Woodring believes that "the bulk of SoIC capacity is for Apple’s 3nm AI ASIC for Private Cloud Compute," an in-house chip the company has internally dubbed "Baltra."
“This is to replace the current M-series Ultra processors used in PCC today for better AI inference performance and efficiency, consistent with reports of Apple’s multi-year collaboration with Broadcom,” he added.
Private Cloud Compute is the server-side backbone of Apple Intelligence, the AI feature suite Apple introduced for its devices. By developing its own silicon for these servers rather than relying on third-party cloud providers, the tech behemoth is pursuing greater cost control and a more vertically integrated technology stack.
The wafer volumes involved are substantial by industry standards. For comparison, AMD’s 42,000 SoIC wafers in 2026 are estimated to produce nearly 800,000 MI-series AI GPUs. Apple’s 36,000 wafers in 2026, rising to 60,000 in 2027, are dedicated largely to Private Cloud Compute infrastructure, Woodring noted.
He pointed out that "performance and efficiency of Apple’s AI ASIC at scale remain key unknowns," even as the strategic direction becomes clearer.
Apple’s spending on this infrastructure, the analyst added, appears more likely to be treated as capital expenditure than as an ongoing operating cost — a structure that could prove more cost-efficient over time.