Memory prices, not AI demand, are driving enterprise hardware costs

Enterprise hardware revenue can rise even as customers buy fewer machines. For technology leaders, separating price inflation from genuine demand is becoming essential to budget decisions.

Barclays’ October 14th downgrade challenged the enterprise AI growth story at Dell Technologies, Hewlett Packard Enterprise, and NetApp. Its analysts credited rising memory costs more than expanding enterprise AI adoption for strong enterprise revenues.

Hardware budgets are rising with memory prices whether or not an enterprise expands its AI spending.

PC Shipments Fell More Than 20% as Memory Neared 40% of Component Cost

A shrinking PC market offers little comfort when replacement configurations cost more.

Omdia’s October 9th figures showed global PC shipments falling 21.2% annually to 58.1 million units in the third quarter. Gartner’s October 12th preliminary results estimated a 22.4% decline to 55.4 million machines. IDC estimated a 20.1% decline to 62.7 million. The three firms differ on totals but agree on the direction. HP’s shipments fell about 31% and Dell’s 25%, according to IDC.

Omdia’s October 9th estimates put DRAM and SSDs near 40% of PC component costs. Their previous share was around 15%, following price increases exceeding fourfold. The figure combines memory and storage.

Gartner attributed the contraction to manufacturers passing higher component costs to customers. Its October 12th assessment put second-quarter DRAM costs at five times their level a year earlier.

IDC research director Jitesh Ubrani described manufacturers and distributors building inventory ahead of anticipated price increases. IDC’s October 9th figures showed shipments declining 9.1% from the preceding quarter, reversing the usual seasonal increase.

Gartner researcher Namrata Banerjee emphasized higher prices and manufacturers favoring expensive configurations over excess inventory. Ubrani expected temporary promotions, but no return to previous pricing. A discounted machine already in stock can have different economics from a newly configured fleet order. Omdia expects shipments to fall another 24% in the fourth quarter and 7% across 2027.

The Operator’s Lens

• Budget replacement quantities and quoted configuration prices separately.

• Compare stocked promotions with approved configurations, warranties, and delivery requirements before booking savings.

Component shortages can suppress equipment purchases while lifting profits for the suppliers behind those machines.

Samsung Guided to $80 Billion in Quarterly Profit as the Memory Shortage Deepened

Record-scale supplier profits can coexist with mounting pressure on customer budgets.

Samsung Electronics’s October 8th guidance projected quarterly operating profit between ₩107.3 trillion and ₩107.5 trillion, approximately $80 billion. The company projected revenue between ₩194 trillion and ₩196 trillion, approximately $145 billion. These preliminary figures cover the quarter ended September 30th. Detailed results were scheduled for October 29th.

The guidance does not separate divisional earnings, so memory’s precise contribution remains unconfirmed. Manufacturers are directing capacity toward more profitable AI infrastructure products. Sales of high-bandwidth memory for AI processors have grown from about $4 billion in 2023 to an estimated $77 billion this year. Consultancy Bain & Company says the shift is worsening shortages and raising the prices of smartphones and PCs. Reduced capacity for conventional devices carries the cost of AI investment into hardware bought for other purposes.

The constraints reach conventional DRAM and flash storage chips, alongside high-bandwidth memory used with AI processors. Samsung warned earlier this year that shortages could deepen during 2027 and continue into 2028. Micron Technology also expected tight conditions during 2027 and 2028.

TrendForce’s September 30th forecast projected conventional DRAM contract prices rising another 10% to 15% in the quarter ending December 31st. TrendForce expects the pace of those increases to moderate. Institutional investors cited in Taiwan expect overall DRAM contract prices to rise about 20% in the fourth quarter, down from about 80% in the second quarter and 40% in the third, as price ceilings in long-term supply agreements take effect.

Micron chief executive Sanjay Mehrotra said on September 30th that much of the company’s 2027 high-bandwidth memory volume is already sold out at prices “much higher than 2026 prices.” He added, “We do not have line of sight to when supply and demand will return to balance.”

Micron’s Taiwan unit signed a pay deal with its Taichung union on October 21st, adding a quarterly profit-sharing bonus. The Taoyuan union began a symbolic 24-hour strike the next day. Only 11 union officials took part, and the union said the action was not meant to disrupt production.

Slower price increases still mean higher bills. Samsung faces both sides of this tradeoff, with memory prices helping semiconductors while raising expenses for its electronics operations. Enterprise buyers can encounter the same expense pressure without increasing their own AI requirements.

The Operator’s Lens

• Put explicit price adjustment limits and written supplier notification periods into major hardware contracts before renewal.

• Retain a component inflation reserve until supplier quotations show sustained relief.

Higher component prices also make equipment revenue a less reliable signal of expanding enterprise demand.

Barclays Downgraded Dell, HPE and NetApp Over Memory-Driven Revenue

Revenue growth alone cannot validate an enterprise AI replacement cycle.

Barclays lowered Dell, HPE, and NetApp from overweight to equal weight, broadly equivalent to moving from buy to hold. Its reasoning separates higher selling prices from more equipment shipped.

The bank argued that enterprise AI spending had not accelerated enough to support expectations of a broad hardware replacement cycle. The dispute is about timing.

Barclays’ October 14th analysis estimated that businesses typically devoted just 5% of their IT budgets to AI. It estimated hardware captured about 25% of enterprise AI spending. These are analyst estimates, not disclosed spending figures from every customer.

Barclays’ October 14th call lowered Dell’s price target to $574 from $603. Dell shares fell about 17% across October 14th and 15th. Analyst Tim Long challenged both the explanation for current revenue and expectations for future demand.

Michael Dell, chief executive of Dell, argued on October 13th that enterprise AI adoption remained at an early stage. Dell also maintained that demand for AI computing exceeded available supply.

Barclays identified another exposure if memory prices stabilize or decline. Equipment vendors could lose pricing support even without weaker unit demand.

The Operator’s Lens

• Require workload capacity estimates and utilization targets for AI infrastructure approvals.

Expectations for future memory supply can move hardware valuations before additional production reaches enterprise buyers.

China’s CXMT and 2029 Oversupply Fears Hit Memory Stocks

Investors can price in future oversupply long before procurement teams see cheaper quotes.

Memory stocks remained under pressure on October 15th amid concerns about Chinese competition and possible oversupply. Over five days, SK hynix fell 26%, Samsung 22% and Micron about 11%. The share of global DRAM held by ChangXin Memory Technologies (CXMT) has risen to about 10%, from about 4% a year earlier, according to HSBC.

CXMT plans server memory with a denser memory-cell architecture, targeting introduction by the end of 2026. The proposed products use fifth-generation double data rate memory. Analysts questioned whether CXMT can produce it in volume.

Kiwoom Securities researcher Han Ji-young described manufacturing yields, performance, and customer qualification as unverified. Qualification determines whether components meet buyers’ requirements. Quilter Cheviot researcher Ben Barringer emphasized manufacturing yield, the share of produced chips that work correctly. Density delivers limited commercial advantage if too many chips fail.

A South Korean newspaper column cited Ahn Ki-hyun of the Korea Semiconductor Industry Association and a Seoul National University professor predicting a flood of new supply by 2029, a forecast other analysts have not echoed. Rosenblatt Securities analyst Kevin Cassidy expected CXMT’s developments to leave the supply-demand balance largely unchanged. Citi analysts wrote in August that the US government is unlikely to allow China-made memory to be sold in the US, though sales to data centers in regions such as Europe could still pressure Micron.

On October 16th, Goldman Sachs told clients the market had overestimated how quickly CXMT can close the gap in advanced memory. SK hynix and Sandisk still fell 3% that day, and Micron 2%. The shares recovered part of the loss this week. On October 21st, SK hynix rose 7% and Samsung about 6% in Seoul after South Korea reported a 222.5% jump in chip exports in early October.

Counterpoint Research’s forecast reported on October 15th projected DRAM prices declining approximately 23% for 2028 compared with 2027. The forecast still placed prices approximately 230% above their 2025 level. Even that decline would provide only partial relief from the elevated cost baseline.

Counterpoint analyst Jeongku Choi argued that longer supply agreements could moderate a downturn through advance payments and negotiated pricing boundaries. New competition and existing contractual commitments can pull prices in different directions.

The Operator’s Lens

• Keep essential replacements on schedule with staged orders. New memory factories take three to five years to come online.

• Require documented qualification status and warranty coverage before approving memory substitutions in enterprise configurations.

SK hynix reports third-quarter results on October 28th and Samsung on October 29th, the first earnings since the selloff. Enterprise planning needs separate views of component prices, purchase volumes, and productive capacity. AI investment influences supply available to every buyer, but purchasing decisions still need to follow actual workloads and replacement obligations. Supplier earnings and stock movements cannot substitute for that assessment.