U.S. Foundry Capacity Arrives Too Late — Intel's Ohio Fab Finally Lifts Its First Truss, Even as Memory Prices Surge

Semiconductors 22 OCT 2026 1,710 words 8 min read By news.jualin.id-publisher

Intel's Ohio Fab Finally Lifts Its First Truss — Five Years Late

The United States has spent years promising to bring leading-edge chipmaking home. On October 21, 2026, Intel announced that crews had installed the first major structural truss at its new chip fabrication plant in New Albany, Ohio — a milestone that finally puts something tangible behind the rhetoric. Yet the project is now slated for operations no earlier than late 2030 or 2031, five years behind the original 2025 target. The truss is a beginning, not a finish line, and the delay coincides with the most intense AI-driven demand cycle the industry has ever seen.

Intel's Foundry head Naga Chandrasekaran framed the moment optimistically: "This represents another significant step toward expanded U.S. leading-edge semiconductor capacity," he wrote on the company's website on Wednesday. But the same announcement quietly acknowledged that the first Ohio fab, originally expected to begin delivering chips in 2025, is now aimed for completion in 2030, with a second fab at the same site planned for completion in 2031. A second Arizona fab is expected to begin operations by the end of 2028.

Intel's first offshore assembly plant in Penang, Malaysia, under construction in 1972 — a historical illustration, not the Ohio site; the Intel "ELECTRONICS FACTORY" sign, steel truss skeleton and stacks of lumber are visible. Photo: Intel Free Press, CC BY-SA 2.0.

The photo above is a historical illustration rather than the Ohio site: Intel's first offshore assembly plant in Penang, Malaysia, under construction in 1972 on what had been a muddy paddy field, captured by the Intel Free Press. The vintage image — showing the Intel "ELECTRONICS FACTORY" sign, a steel truss skeleton and stacks of lumber — underscores how the physical scale of these facilities has always been the easy part; the hard part is filling them with customers and keeping them on schedule.

Why the Delay: CEO Lip-Bu Tan's Cautious Pivot

Intel's slowdown is not an engineering accident; it is a deliberate strategic recalibration. When Lip-Bu Tan took the helm in 2025, he replaced former CEO Pat Gelsinger's aggressive foundry expansion with a more cautious turnaround: wait for customer demand to solidify before forging ahead with massive fab plans. Under Gelsinger, Intel had launched its external foundry ambitions; under Tan, the Ohio factory was deliberately slowed while the company refined its most advanced 18A node.

By late 2025, Intel was in high-volume production at its Arizona fab, producing its most advanced 18A chips — but exclusively for its own processors. Intel has yet to land a major external customer for those 18A chips, a gap that explains why the Ohio project was deferred. The company's earlier delays on 7- and 10-nanometer nodes, which triggered its worst stock year ever in 2024, have made customers reluctant to commit again.

There is a possible external lifeline. In September, reports surfaced that SK Hynix — the world's leading maker of high-bandwidth memory — could team up with Intel in Ohio to lease foundry space for memory chips. Neither company confirmed any plans, so the arrangement remains unconfirmed, but it would fit Intel's need for a flagship customer and SK Hynix's need for U.S. capacity.

Legacy DRAM Pricing Surges: Powerchip's Niche Strategy

While the leading-edge story dominates headlines, the mature-node segment is telling a very different tale. Powerchip Semiconductor Manufacturing Corp., which bills itself as poised to become the world's sole memory manufacturer producing legacy DRAM from DDR1 to DDR3 using 63-nanometer to 20-nanometer processes, is negotiating further price hikes of 10 to 20 percent next year, citing persistent capacity constraints across the industry.

The numbers are striking. Last quarter, Powerchip raised prices by 23 percent from the previous quarter and expects another 30 percent increase this quarter. Net profit nearly doubled to NT$6.5 billion (US$200 million) from NT$3.29 billion, reversing a loss of NT$2.73 billion in the third quarter last year. Gross margin improved to 48 percent from 28 percent a quarter earlier, and earnings per share rose to NT$1.38 from NT$0.76.

Powerchip president Martin Chu framed the outlook in terms of AI: "Provided that artificial intelligence [AI] momentum remains strong, we hold a highly optimistic outlook for 2027 revenue and profitability." Chairman Brian Shieh added that the company anticipates its niche DRAM foundry pricing will remain robust, showing no signs of a downward adjustment.

The reason for the pricing power is reallocation. Aggressive AI server buildouts have spurred massive demand for high-bandwidth memory, squeezing mature DDR3 and DDR4 DRAM production as the major memory makers reallocate capacity to HBM. As Chu put it: "As major memory makers aggressively reallocate capacity to HBM and exit mature DRAM, we expect supply tightness for customers." Powerchip's legacy technologies, with an extended lifecycle of 10 to 15 years, are now effectively the only source for customers who still need older DRAM.

Four Samsung 8GB DDR4-2400 registered ECC server memory modules — not Powerchip's own products, but illustrative of the server-grade mature DRAM that today's capacity crunch spans; Samsung labels and Supermicro warranty stickers are visible. Photo: Wikimedia Commons, CC BY-SA 4.0.

The photo above shows four Samsung 8GB DDR4-2400 registered ECC server memory modules — not Powerchip's own products, but an illustration of the server-grade mature DRAM that today's capacity crunch spans. The Samsung labels and Supermicro warranty stickers are visible on the modules.

Advanced Packaging and IPD: Customer-Funded Capacity

Powerchip's expansion plan is not just about legacy DRAM. The company is scouting a site to build a small-scale chip facility in Taiwan, Japan or Singapore, focused on advanced packaging-related technologies: post-wafer-finish (PWF) service, interposers used in advanced chip-on-wafer-on-substrate packaging, and integrated passive devices (IPD) used in AI-related applications. Demand for IPD is expected to surge from 10,000 wafers per month currently to as many as 50,000 wafers per month in the second half of next year.

To mitigate capital expenditure risks, Powerchip aims to secure 30 to 60 percent of funding for new production line expansions from customers via leasing or prepayments. Micron Technology has set a precedent by fully funding the manufacturing equipment for Taiwan's PWF mid-to-back-end packaging expansion through prepayments. The logic is that customer-funded capacity de-risks the buildout and aligns supply with committed demand.

Schematic illustration of multiple silicon dies mounted on an IC substrate — the kind of advanced packaging that interposers and post-wafer-finish services enable. 3D render, not a photograph. Image: Wikimedia Commons, CC BY-SA 4.0.

The illustration above depicts the conceptual assembly of multiple silicon dies onto an IC substrate — the kind of advanced packaging that interposers and PWF services support. It is a 3D schematic rather than a photograph, showing blue and red chips stacked and placed onto a green substrate.

Supply Crunch Mechanics: HBM Reallocation and Legacy Pricing

The through-line of this market is migration, not growth. Leading-edge capacity is arriving years late: Intel's Ohio fab will not deliver high-volume wafers until 2030 or 2031, while TSMC's lines remain so contested that even the largest chipmakers — Nvidia, AMD, Apple and Broadcom — face "extremely fierce" competition for space, according to Mizuho Securities analyst Jordan Klein. Intel could offer customers an alternative source, but only if it can prove its process technology and attract a flagship tenant.

On the mature side, the HBM boom is squeezing legacy DRAM. Powerchip's mild pricing weakness in DDR3, attributed to a temporary supply influx from an enterprise refresh cycle where outdated computers were replaced with AI-capable models, is expected to be short-lived. The company expects revenue to expand by a double-digit percentage sequentially this quarter, after revenue expanded 23 percent last quarter.

Policy Context: "Too Far Away" and New U.S. Investment

The policy backdrop is explicit. U.S. Commerce Secretary Howard Lutnick said Tuesday that Taiwan is "too far away" for the United States to rely on for its chip production: "It's all made in Taiwan, right? And that's just too far away for America," he told The Hill. The Trump administration has said it wants to move 40 percent of Taiwan's chip supply chain and production to the U.S. during the current term, and has warned that chips made outside the U.S. could face tariffs of up to 100 percent.

Taiwan rejected the proposal in February, with Vice Premier Cheng Li-chiun saying the industry could not simply be relocated and that Taiwan would keep its most advanced chip technology at home while allowing companies to expand overseas.

The investment side is moving. The Pentagon will invest about NT$47 billion (US$1.5 billion) in semiconductor production in North Carolina, and materials company Sibelco will build a quartz manufacturing facility there — quartz being a critical input for semiconductor manufacturing — with an expected investment of hundreds of millions of dollars. TSMC has committed about NT$8.61 trillion to expand its Arizona operations and is considering further investment in Texas; the company reported a 69 percent annual increase in third-quarter net profit, driven by strong AI demand.

Micron's Taichung Labor Agreement and China's AI Export-Control Draft

Two additional developments round out the supply picture. Micron Memory Taiwan and a union representing workers at its Taichung operations signed a labor-management agreement covering multiple issues, prompting both sides to withdraw from mediation scheduled for the following day. The terms were not disclosed. Micron employs about 15,000 people in Taiwan and has said it invested more than NT$1.6 trillion (US$49.6 billion) in the country; President William Lai had urged both sides to negotiate "sincerely" ahead of a possible strike over a bonus-scheme dispute. Any disruption to Micron's Taiwan operations could affect already strained global supply.

Separately, China proposed adding advanced artificial intelligence technologies to its export-control list, potentially restricting overseas transfers of AI models and chip designs. The draft covers 23 AI technologies, including large language models, agents, chip designs and computing clusters, with restricted technologies to include model weights and source code for products exceeding certain performance thresholds.

What It Means

The Intel Ohio truss is a visual step forward, but the larger story is about timing. The United States is still years away from the leading-edge capacity it needs, while the existing global supply chain is already straining under AI demand. Powerchip's pricing power shows that legacy nodes are not immune to the squeeze — as HBM reallocation drives mature DRAM tightness, niche foundries are capturing margin and expanding packaging capacity on customer-funded terms.

For buyers of semiconductors, the implication is a sustained period of tightness across both leading and mature nodes, with pricing power firmly on the supplier side. For the U.S. policy agenda, the lesson is that announced capacity and delivered capacity are different things, and the gap is measured in years.