From Tokyo: in a country where engineering ambition often speaks louder than corporate press releases, Fujitsu has just confirmed one of the semiconductor industry's worst-kept secrets. Its next-generation Monaka processor, a 144-core Arm-based chip designed for data centres, will be built using Broadcom's 3.5D XDSiP stacking technology. First samples have been shipped. The clock is now ticking toward a 2027 commercial launch, and the implications stretch well beyond Japan.
The Monaka chip represents a meaningful departure from Fujitsu's previous processor work. The company's Arm-based A64FX powered the original Fugaku supercomputer to the top of the global Top500 list in 2020, relying heavily on on-package high bandwidth memory. Monaka takes a very different approach: gone is the on-package HBM, replaced by an SRAM-heavy architecture. The chip is a system-in-package with four 36-core compute chiplets made on TSMC's 2nm process, stacked on top of SRAM tiles in a face-to-face manner using hybrid copper bonding. Those SRAM tiles are produced on TSMC's 5nm process technology.
The packaging technology at the heart of this collaboration is Broadcom's XDSiP, which stands for Extreme Dimension System in Package. Until recently, only AMD and Intel had the technology required to build a system-on-chip of this nature, and Broadcom's 3.5D XDSiP, introduced in 2024, was its attempt to change that. One of the core innovations is Broadcom's use of face-to-face 3D chip-stacking based on hybrid bonding, which connects copper wiring on the front of each silicon die directly without solder bumps, making it possible to sling signals between stacked dies far faster than previously achievable. The result is denser interconnects, lower latency, and a smaller overall package footprint.
What Australian observers often miss about Japan's chip industry is that it operates on very long planning horizons, backed by a mixture of corporate R&D and government subsidy. Fujitsu's Monaka project is partly subsidised by Japan's New Energy and Industrial Technology Development Organisation (NEDO) under a broader green data centre programme. That kind of structured, patient capital allows Japanese firms to pursue semiconductor bets that a purely commercial timeline might not support. The parallel with Australia's own hesitant attempts to build a domestic semiconductor strategy is instructive: Canberra has talked up the critical minerals angle, but the downstream chip design and fabrication capacity remains an open question.
Harish Bharadwaj, vice president of Broadcom's ASIC product division, confirmed to The Register that samples had already been dispatched to Fujitsu. "We've been working on this for almost five years as a technology," he said, adding that multiple other customers had also adopted the platform for their next-generation designs. Fujitsu may be among the first to embrace XDSiP publicly, but Bharadwaj noted that Monaka is only one of roughly half a dozen designs currently in development on the platform.
The commercial ambition behind the platform is considerable. Broadcom anticipates shipping at least one million units of its 3D stacked chip technology by 2027. While Monaka is a CPU platform, roughly 80 per cent of Broadcom's XDSiP design wins are for accelerators, or XPUs, with high bandwidth memory on board. When the platform was announced in 2024, it supported designs with up to 12 HBM stacks; designs with more than 12 stacks are now said to be in development. The trajectory suggests that the next generation of AI training hardware could dwarf even today's already formidable accelerators.
Sceptics will rightly point out that Monaka faces a steep commercial climb. Fujitsu's new Arm-based processor family will compete directly with AMD's fleet of EPYC processors and Intel's Xeon chips, both of which enjoy deeply entrenched software ecosystems and established data centre relationships. The server CPU market does not easily surrender share to newcomers, however technically impressive their silicon. Monaka will have standard Linux OS support, and Fujitsu is collaborating with Arm on open-source development tools, compilers, and libraries, which may help ease adoption, but convincing hyperscalers and enterprise customers to diversify away from x86 incumbents requires more than a compelling benchmark sheet.
The energy efficiency argument may prove Monaka's strongest card. Fujitsu has set ambitious goals, aiming to achieve twice the energy efficiency of current x86 processors by 2027 while maintaining air-cooling capabilities. For data centre operators grappling with soaring power costs and tightening sustainability commitments, a processor that genuinely halves the energy bill per unit of compute would be a significant proposition. Fugaku, Monaka's predecessor platform, is currently ranked the fourth most powerful supercomputer in the world, and Monaka is expected to feature in its successor, FugakuNEXT, planned for 2030.
The cultural significance extends beyond any single chip launch. For decades, the global narrative in advanced semiconductors has been written almost entirely by American and, to a lesser extent, European companies. Broadcom's decision to commercialise its most advanced packaging technology through a partnership with a Japanese firm, manufactured on TSMC's Taiwanese fabrication lines, reflects just how thoroughly the industry's centre of gravity has shifted to the Indo-Pacific. For Australia, sitting at the intersection of the supply chains, trade relationships, and geopolitical tensions that govern all of this, the rise of a credible Japanese challenger in high-performance computing is far from an abstract story.
Monaka will not ship until at least fiscal year 2027. Whether Fujitsu can convert an elegant engineering achievement into genuine market share remains an open question that only the data centres of 2028 will answer. But the fact that Broadcom has spent five years and delivered its most sophisticated packaging technology to a Japanese partner first, in a deal that both companies chose to disclose publicly when chip IP arrangements are usually treated as closely guarded secrets, tells you something important about where the competitive pressure in the semiconductor industry is really building. Reasonable observers can differ on whether Japan's state-backed chip strategy is a model worth emulating or an expensive industrial policy gamble. What is harder to dispute is that, in the race to define the next era of computing silicon, the contest has become genuinely global.