Global Semiconductor Sales Up 94% Year-Over-Year — HBM4E, RTX Spark, and Intel 18A Are Rewriting the AI Chip Architecture Map

The HBM4E Race Is the Memory Market's Most Consequential Competition in a Decade

The acceleration of Samsung and SK Hynix in shipping HBM4E seventh-generation samples to major AI chip customers defines the near-term competitive boundary for the AI infrastructure buildout in a way that is frequently underweighted in coverage focused on compute chips. High Bandwidth Memory is the performance bottleneck that limits how fast AI models can actually execute inference — a compute chip that can perform more FLOPS than its memory can feed data to is an underutilized asset, which means memory bandwidth is the binding constraint on AI system performance at the architectural level. HBM4E's advance beyond the HBM4 generation — which was itself only beginning to sample earlier this year — reflects a development pace that the memory market has not sustained since the DDR3-to-DDR4 transition, and the speed of the transition matters commercially because the AI infrastructure operators building out capacity in 2026 are making purchasing decisions today about system configurations that will be in production for three to five years. Locking in HBM4E-capable systems over HBM4 systems is a meaningful infrastructure investment that affects AI performance economics over that entire period, which is why Samsung and SK Hynix's aggressive sampling timeline — and NVIDIA's confirmation of LPDDR6 use in the Rubin generation — are commercially significant events rather than routine product roadmap announcements.

Intel's milestone on its 18A-P process node entering risk production is the foundry development with the most geopolitical and commercial weight in the current semiconductor cycle, because it represents Intel's first credible attempt to re-enter the leading-edge process competition that TSMC has dominated since Intel's manufacturing stumbles began in 2018. The reported interest from Apple and Google in 18A-P capacity, and Intel's strategic partnership with UMC to challenge TSMC's foundry dominance, suggest that the major hyperscalers are actively hedging their TSMC concentration even as TSMC's Phoenix fabs come online and its panel-level packaging capabilities for AI chips advance. For the European semiconductor sovereignty agenda, Intel's Ohio and Arizona investment, TSMC's Phoenix expansion, and the EU Chips Act's strategic reset together define a period in which the geographic distribution of leading-edge semiconductor manufacturing capability is actively in flux for the first time since the 1990s — and the policy and investment decisions made in 2026 will shape that distribution for at least fifteen years.

The EU Chips Act Reset — From Supply to Demand Stimulation

The EU's strategic reset of its Chips Act, announced alongside the US closing of a key export loophole restricting Chinese firms' access to AI chips abroad, reflects a recognition that the original Chips Act's supply-side focus — subsidizing leading-edge fab construction in Europe — was producing results on a timeline too slow to address the immediate competitive gap in AI infrastructure. The reset toward demand stimulation — creating conditions under which European companies build, buy, and deploy AI chips at a rate that justifies domestic semiconductor ecosystem investment — is a materially different policy logic, because demand stimulation operates on a one-to-three year commercial timeline rather than the five-to-ten year fab construction timeline. Catalonia's effort to build a unified semiconductor cluster spanning photonics, packaging, AI, and chip research, and GlobalFoundries and Qualinx delivering what they describe as Europe's first fully secure chip supply chain, are expressions of this demand-stimulation logic applied at the regional scale. For US semiconductor companies with European operations or customers, the EU Chips Act reset creates both opportunity — accelerated European demand for AI chips — and competitive complexity, as European industrial policy increasingly distinguishes between domestically-produced semiconductor capability and imported chips, even from allied nations.

The Section 48D Advanced Manufacturing Investment Credit expiration at the end of 2026 remains the most consequential near-term policy uncertainty for the US semiconductor supply chain — and the most overlooked. The SIA has been explicit that the credit, which provides a 25 percent investment credit on domestic semiconductor manufacturing capital expenditure, is essential to the investment case for the specialty chemicals, advanced packaging, and test and assembly capacity that the anchor fabs TSMC and Intel are building in the US require to function as a complete domestic supply chain rather than as isolated islands of leading-edge manufacturing surrounded by imported inputs. The STAR Act introduced by a bipartisan Senate group to extend the credit has not advanced past committee, creating a planning environment where companies making three-to-five-year capital investment decisions cannot assume the credit will be available for the full operational life of the facilities they are building.

Companies to Watch