In June, while discussing memory chips, Intel CEO Lip-Bu Tan said: “There’s no relief until 2028.” He was referring to memory chips – not a product Intel manufactures, which is precisely why the remark carried weight – a chip executive with no direct stake in DRAM pricing was confirming what buyers across the industry had started to suspect. The factories built to fix America’s memory shortage are real, funded, and under construction. They are also, by their own builders’ timelines, three to four years away from producing anything at volume.
That gap has become the story. The CHIPS and Science Act, signed in 2022, was publicly framed to fix exactly this kind of supply crisis: reshore memory manufacturing, reduce dependence on a handful of Asian suppliers, insulate American industry from exactly the kind of price shock now unfolding. Micron and SK Hynix took the money and broke ground. But construction schedules, environmental permitting, and equipment installation cycles run on a timeline measured in years, not quarters – and the shortage they were meant to prevent arrived first.
The fabs, and when they actually open
Micron’s Boise, Idaho facility was originally supposed to start DRAM production in early 2026. The company’s most recent SEC filing indicates first wafer output is now projected for mid-2027. A second Idaho fab, announced in June 2025 specifically to meet AI-driven demand, won’t be operational until the end of 2028. Micron’s New York campus near Clay – the flagship of its $100 billion U.S. expansion – has slipped further: construction on the first fab began this year, but the facility isn’t expected to produce supply until 2030, five years behind the original target, according to the company’s final environmental impact statement. Micron has since redirected $1.2 billion of its CHIPS Act funding away from New York and into Idaho, an acknowledgment of where the company believes it can get chips to market faster.
SK Hynix’s $3.87 billion Indiana facility, built alongside Purdue University, broke ground this year and is targeting mass production in the second half of 2028. It’s worth noting what that plant actually makes: not commodity DRAM, but advanced packaging for HBM4E and HBM5 – the high-bandwidth memory that feeds AI accelerators, not the memory that goes into a laptop or a car’s infotainment system. Even when it opens, it won’t touch the segment of the shortage hitting consumer electronics and industrial buyers hardest.
Add it up, and the earliest a CHIPS-funded U.S. fab adds meaningful DRAM volume to the market is mid-2027, with the bulk of new capacity – including the highest-profile New York project – not arriving until 2029 or 2030. SK Hynix CEO Kwak Noh-Jung told Bloomberg in July that the broader memory crunch could persist “until the next decade.” That’s more than a forecasting caveat – it’s the person running one of the world’s three DRAM makers saying the shortage may outlast the fabs built to end it.
Why the gap years are so brutal
Unlike the usual triggers behind memory crunches – a factory fire, a natural disaster – the cause this time is a structural reallocation. Producing a gigabyte of HBM consumes three to four times the wafer capacity of standard DDR5, and HBM commands two to three times the margin. Samsung, SK Hynix, and Micron control over 70% of global DRAM supply between them, and all three have shifted capacity toward AI-grade memory because that’s where the money is. TrendForce’s Avril Wu, an analyst who has tracked the memory market for two decades, called the current period “the craziest time ever” in the industry’s history.
The numbers back her up. DRAM contract prices increased by an estimated 90–95% quarter-over-quarter in Q1 2026, according to TrendForce, followed by another 58–63% jump in Q2. Bloomberg reported spot prices up nearly 700% over the trailing year. Data centers are on pace to absorb roughly 70% of global memory chip output through the rest of 2026 and into 2027 – a run-rate long enough, one industry report noted, to leave telecom equipment makers and other non-AI buyers short of supply for at least two more years.
Who’s actually exposed
The exposure isn’t evenly distributed, and that unevenness is the part CFOs and procurement teams should be tracking closely. Apple secured long-term DRAM supply agreements through the first quarter of 2026 and has weathered the shortage better than most of its competitors, though it still raised prices across its device lineup in June. Dell has not been as fortunate: Morgan Stanley downgraded the stock from Overweight to Underweight in late 2025, citing the company’s exposure to rising server memory costs. Lenovo, HP, Acer, and Asus have all warned customers of 15–20% price increases in 2026, with some analysts projecting hikes as high as 30%. HP disclosed that memory now accounts for roughly 35% of a laptop’s bill of materials, up from 15–18% just one quarter earlier.
Automakers face a comparable squeeze with less room to pass costs along. S&P Global Mobility projects DRAM prices could rise 70–100% in 2026 versus 2025, with lead times on new orders stretching past 58 weeks. Tesla’s Elon Musk framed the choice bluntly in January: automakers must “hit the chip wall or make a fab.” Honda has already cited semiconductor shortages in cutting North American production by 110,000 units. Yamaha told investors in February it has “no line-of-sight” on when memory supply catches up with demand.
Medical device makers and telecom equipment manufacturers are quieter but no less exposed. A multi-industry coalition – spanning automotive, medical device, and telecommunications firms – sent a letter to the Trump administration this year warning of production disruptions and delays affecting federal contractors, marking what Tom’s Hardware described as the first coordinated push for federal intervention beyond the original CHIPS funding. Compounding the problem, Samsung is exiting MLC NAND flash entirely, with final shipments in June 2026 – a segment analysts expect to see capacity drop more than 40% this year, hitting industrial control systems, automotive electronics, and medical devices that depend on it.
The strategic takeaway
For enterprise buyers, the lesson isn’t that prices will eventually normalize – most analysts now treat that as unlikely before 2028 at the earliest, and possibly not until well beyond it. The lesson is that security of supply, not price optimization, is now the variable that determines whether a production line keeps running. Companies without Apple’s negotiating power or long-term contracts are, as Counterpoint Research’s Tarun Pathak put it, effectively standing in line behind the hyperscalers. The CHIPS Act fabs will eventually change that math. They just won’t do it in time for the years that matter most.
