The global trade data published Wednesday contains a number that reframes the entire consumer electronics market: Nvidia's AI processors, tracked under US Customs import category HTS 847150, crossed $165 billion in 2025 — larger than the combined $109 billion in smartphone and PC imports that year, according to Semiconductor Intelligence analyst Bill Jewell's analysis published on SemiWiki on July 23, 2026. The devices people actually use — the laptops they work on and the phones in their pockets — now rank below the AI chips feeding data centers in the hierarchy of what gets built, shipped, and prioritized. That inversion is why global PC shipments fell 4.9% year-over-year in the second quarter of 2026 and smartphone shipments dropped 6.7%, according to IDC data analyzed by Jewell and published on SemiWiki on July 23, 2026.

Neither decline reflects a demand problem. Three major PC makers put it on the record in their most recent earnings calls: Apple's 10-Q for its quarter ended in March 2026 stated the company is experiencing supply constraints on advanced semiconductors, NAND, and DRAM. HP, reporting its quarter ended in April, said memory and storage costs increased sequentially in Q2 and that trend is expected to continue in the second half of 2026. Dell, reporting its quarter ended in May, offered the most direct statement: the company acknowledged a supply issue, confirming it is supply constrained in the second half and that it is not a demand issue. When a company that sells laptops to millions of enterprise buyers says the problem is not demand, it means consumers who want to buy are unable to find the configurations they need. All three disclosures are sourced to Jewell's analysis of the earnings transcripts.

IDC projects the full-year damage at 11.3% for PCs and 13.9% for smartphones in 2026 — both categories that had posted growth in 2024 and 2025. These are not rounding errors. They represent tens of millions of devices that will not be built because the chips to build them are going elsewhere.

What Is HBM and Why Does One AI Chip Cost You Three Smartphones

The mechanism behind the shortage is specific and documented. High-bandwidth memory — the three-dimensional, stacked DRAM chip that sits directly alongside the processor in every Nvidia AI accelerator — and the low-power mobile DRAM inside a smartphone are produced by the same three companies, in the same fabrication facilities, using the same silicon wafers. Samsung, SK Hynix, and Micron together control more than 95% of global DRAM production.

Beginning in 2024, all three began systematically converting fabrication capacity toward high-bandwidth memory, because the economics are compelling: HBM generates approximately three to five times more revenue per wafer than standard consumer DRAM. More importantly, producing HBM is physically inefficient compared to conventional DRAM — the stacking of up to 16 DRAM dies using Through-Silicon Via technology, and the advanced packaging required to bond HBM to a GPU die, means each HBM wafer consumes roughly three to four times more fab capacity per usable memory bit than conventional DDR5 or LPDDR5X. Micron has publicly disclosed a 3:1 conversion ratio — meaning each HBM wafer effectively displaces the production of three conventional DRAM wafers.

What this means in practice: one AI accelerator chip is roughly three smartphones' worth of memory manufacturing capacity that no longer exists for consumers. By mid-2026, AI data centers were consuming approximately 70% of all memory chips produced worldwide, up from 20 to 30% as recently as 2022, according to IDC analysis. That is not a cyclical shift. It is a structural reallocation.

Dell, HP, Apple Are on Record: This Is Not a Demand Problem

The earnings disclosures from the three largest PC makers in the US market distinguish this shortage from every prior downturn in consumer electronics. A demand decline looks different: inventory piles up, prices soften, and manufacturers slow production voluntarily. What the Q2 2026 data shows is the opposite pattern.

Apple raised prices on MacBooks and iPads earlier in 2026 and acknowledged mounting costs from the chip shortage. Yet Apple also posted a 10.1% increase in PC unit shipments year-over-year in Q2, gaining market share to nearly 10% of global PC shipments — the highest in company history for a second quarter. The mechanism behind Apple's outlier performance is supply chain management: the company secured DRAM and NAND contracts 12 to 24 months in advance, and its premium product architecture keeps memory as a smaller share of total bill of materials than it represents in budget-tier devices, as detailed in TechTimes' coverage of Apple's Q2 PC market performance.

HP experienced the sharpest decline among the top three vendors at negative 9.0%, shipping 13 million units. Dell shipped 9.3 million units at a negative 5.0% rate. Lenovo retained the global lead with 16.6 million units despite a 2.1% decline. All figures are from IDC's Q2 2026 Worldwide Quarterly Personal Computing Device Tracker.

IDC's Jean-Philippe Bouchard noted the structural advantage that scale provides in a supply-constrained market, as reported in TechTimes' PC market coverage: the largest vendors, with their buying power and long-standing supplier ties, are best positioned to take share from smaller rivals. The negative 10.5% decline in the "Other" PC category — the long tail of smaller brands without procurement leverage — illustrates exactly what Bouchard described.

AI Processor Imports Now Larger Than All Consumer Devices Combined

The most arresting single data point in Jewell's analysis arrives from an unlikely source: US Customs import category HTS 847150, which covers digital processing units that are not complete computer systems and do not include input or output devices. This is where Nvidia's AI processors appear in trade statistics.

The numbers are stark. HTS 847150 grew from $37 billion in 2023 to $165 billion in 2025. That $165 billion is 51% larger than the $109 billion in combined smartphone and PC imports in 2025. The geographic sourcing of those imports points directly at the Nvidia supply chain: 49% came from Mexico and 42% from Taiwan.

Nvidia's AI processors are fabricated by TSMC in Taiwan. Foxconn operates factories in Mexico, including a $900 million assembly plant for Nvidia's GB200 server systems, to package those chips into modules before they are sold to the companies building AI systems. The importing company buys the module, adds its own memory, storage, and networking, and then builds the AI server. This supply chain — TSMC Taiwan to Foxconn Mexico to US hyperscaler data center — is now larger by dollar volume than the entire US smartphone and PC import economy combined.

The growth rate in HTS 847150 imports closely tracks Nvidia's own revenue trajectory. Imports grew more than fourfold from 2023 to 2025, which matches Nvidia's revenue growth of approximately 3.5 times over the same period. This correlation is not coincidental — it confirms that the HTS 847150 category is effectively a real-time trade ledger for AI infrastructure spending in the United States.

Where Your Next PC and Phone Are Now Made

The supply constraint is not the only major structural shift captured in Jewell's analysis. The data also document a fundamental reorientation of where the world's consumer devices are assembled — a geographic shift that has unfolded rapidly under the pressure of tariffs and trade policy.

On the PC side, the transformation is dramatic. In 2022, China accounted for 92% of US PC imports. By 2025, that share had fallen to just 19%, with Vietnam now supplying 60% of the total. Major PC manufacturers — HP, Lenovo, Apple, Asus, and Quanta — have been shifting PC production from China to Vietnam and other countries.

The smartphone picture mirrors this trend with India emerging as the primary beneficiary. China's share of US smartphone imports fell from 79% in 2022 to 40% in 2025, while India climbed from 2% to 39% over the same period. For the first five months of 2026, smartphone imports from India were worth more than twice the value of those from China. Apple, Samsung, Xiaomi, Oppo, and Vivo all operate production in India.

An important distinction deserves emphasis here. The assembly shift from China to Vietnam and India affects the final-stage manufacturing of consumer devices — where components are put together into finished products. It does not affect where the critical silicon is made. Semiconductor fabrication — the production of the DRAM, NAND, and logic chips that go into these devices — remains concentrated in Taiwan (TSMC) and South Korea (Samsung, SK Hynix). Tariff policy can redirect assembly operations; it cannot quickly relocate the fabs.

When Does Relief Arrive — and Why Prices May Not Return to What They Were

The uncomfortable answer is that relief from the shortage itself is two or more years away, and recovery to pre-shortage pricing may not happen at all.

Micron and Samsung are both doubling capital expenditures versus the prior year. But adding meaningful memory capacity takes years: Samsung's P5 fabrication plant in Pyeongtaek is not expected to reach volume production until approximately 2030; SK Hynix's M15X facility at Yongin is targeting Q1 2027 for its first cleanroom phase; Micron's new ID1 fab in Idaho is not expected to come online before 2027, and will not reach full production yield for years after, as detailed in TechTimes' prior coverage of the PC market impact. Jewell's assessment, published on SemiWiki on July 23, 2026, is that the situation will probably take at least two years to resolve through some combination of increased memory capacity and AI demand leveling off.

Gartner projects memory prices will climb approximately 130% by the end of 2026, pushing PC prices up 17% and smartphone prices up 13% against 2025 levels. Ranjit Atwal, a senior director analyst at Gartner, concluded that the sub-$500 entry-level PC segment will disappear by 2028 — that memory's rise to 23% of total PC bill-of-materials, up from 16% in 2025, removes vendors' ability to absorb the cost increase in low-margin configurations.

Intel CEO Lip-Bu Tan stated bluntly at a Cisco Systems conference in February 2026 that "there's no relief until 2028," citing direct conversations with key players in the memory industry. SK Group Chairman Chey Tae-won has separately warned that supply may struggle to keep pace with demand in the years ahead.

But the deeper problem is not just when relief arrives — it is what "relief" means in this market. When new fab capacity does come online in 2027 or 2028, Samsung, SK Hynix, and Micron will face the same economic calculation that drove the reallocation in the first place: HBM generates three to five times the revenue per wafer that consumer DRAM does. A manufacturer choosing between building HBM for a hyperscaler on a long-term contract and building LPDDR5X for a smartphone OEM faces a straightforward optimization problem. New capacity restores their ability to make that choice — it does not change which option maximizes per-wafer returns.

IDC has described the current reallocation as "potentially permanent" — a structural reset, not a temporary bottleneck. Analysts at multiple firms project the shortage will persist through 2028 or beyond. The difference between a cyclical shortage and a structural reallocation matters for buyers: in a cyclical shortage, waiting is rewarded when prices eventually fall; in a structural reallocation, waiting means paying more for a device that may have fewer features than the one available today.

What This Means for Consumers Buying Now

Gartner expects PC replacement cycles to lengthen by 15% for business buyers and 20% for consumers in response to price increases — which is to say, most people facing today's prices will rationally hold onto their existing hardware longer. That is a defensible choice when the alternative is paying 17% more for equivalent specifications.

For consumers who need a device in the near term, the practical guidance from analysts across Gartner, IDC, and independent retail analysts is consistent: buying sooner is financially better than waiting in the sub-$500 PC and sub-$400 smartphone categories. Devices available today were built with memory contracts signed before the full price shock reached the market; devices launching through late 2026 and into 2027 will incorporate components procured at the elevated prices now flowing through the supply chain, as Gartner's February 2026 analysis makes clear.

Gadjo Sevilla, an analyst covering technology for eMarketer, flagged the certified refurbished market as an unusual source of value in this environment — devices from 2024 and early 2025 carry pre-shortage memory cost structures and typically deliver better specifications per dollar than new 2026 equivalents.

What is becoming undeniably clear from the combined import, shipment, and earnings data is that the AI supercycle is not reshaping the semiconductor market in the abstract. It is physically redirecting the world's most critical chips — away from the devices consumers use and toward the racks in hyperscale data centers — with measurable, documented consequences for the rest of the electronics economy. Waiting for the market to correct itself back to 2025 pricing is, on the current evidence, a plan built on a structural misunderstanding of what has changed.

Frequently Asked Questions

Why are Nvidia AI chip imports bigger than all PC and smartphone imports combined?

Because AI data centers are consuming an unprecedented share of the world's semiconductor manufacturing capacity. Nvidia's AI processors, assembled into modules by Foxconn in Mexico from chips fabricated by TSMC in Taiwan, crossed $165 billion in US imports in 2025 — compared to $109 billion for smartphones and PCs combined, per US Customs data analyzed by Semiconductor Intelligence. This inversion reflects how much hyperscalers like Microsoft, Amazon, Google, and Meta are spending to build AI infrastructure: AI chip spending has grown more than fourfold since 2023, while consumer device trade has contracted. The same three companies that make consumer DRAM — Samsung, SK Hynix, and Micron — also produce the high-bandwidth memory that goes into those AI chips, and they are choosing to make more of the latter because it earns three to five times more revenue per wafer.

Will memory prices return to 2025 levels when new fabs open in 2027 or 2028?

Probably not in full — and possibly not at all for budget-tier devices. When new manufacturing capacity comes online, the three companies controlling 95% of DRAM production will face the same economic calculation that drove the reallocation: high-bandwidth memory for AI data centers earns three to five times more per wafer than consumer DRAM. New capacity restores their ability to choose; it does not change which choice maximizes revenue. IDC has explicitly described the current reallocation as "potentially permanent," and Gartner projects even post-shortage prices will be permanently above 2025 levels. Ranjit Atwal at Gartner expects the sub-$500 PC to disappear entirely by 2028, not as a temporary effect of the shortage but as a structural outcome of memory now representing 23% of total PC bill of materials, up from 16% in 2025.

Should I buy a PC or smartphone now, or wait for prices to fall?

For sub-$500 PCs and sub-$400 smartphones, buying sooner is the better financial strategy based on current evidence. Devices available today incorporate memory procured at earlier contract prices; devices launching through late 2026 and 2027 will carry higher component costs that manufacturers are already warning will flow through to sticker prices or specification downgrades. Gartner projects PC prices will be 17% higher by end of 2026 versus 2025 levels. The certified refurbished market — particularly 2024 and early 2025 devices — offers the best specifications-per-dollar in the current environment because those devices carry pre-shortage component cost structures, according to eMarketer analyst Gadjo Sevilla via CNN. If you do not need a new device in the near term, the most financially rational response, per Gartner, is extending the life of existing hardware.

Does the assembly shift from China to Vietnam and India change where my phone is made — and does it fix the shortage?

It changes where your phone is assembled, but not where the critical components are made — and no, it does not fix the shortage. Major manufacturers including HP, Lenovo, Apple, Asus, and Quanta have shifted PC final assembly from China to Vietnam (now 60% of US PC imports). Apple, Samsung, Xiaomi, Oppo, and Vivo have shifted smartphone assembly to India (now 39% of US imports, up from 2% in 2022). But the DRAM, NAND, and logic chips inside those devices are still fabricated in Taiwan and South Korea, at fabs that have no short-term spare capacity to redirect toward consumer products. Tariff policy reshaped where screwdrivers are turned; it did not add a single new DRAM wafer start. The silicon that determines whether the device has enough memory to work is still coming from the same constrained fabs it was before, as Semiconductor Intelligence's analysis of US import data makes clear.