Two Hikes, One Word
Why Intel and AMD's CPU price hikes are two opposite stories
Steve Williams · babnews.org
The news is easy to summarize and easy to misread. Intel is raising CPU prices about 10 percent, its third increase in a year, and AMD is expected to follow in 2027. The reflex read is that both companies are flexing pricing power in a hot market. That read is half right, and the half it gets wrong is the more important half, because there are actually two different price hikes buried in this story, and they are opposite in almost every way that matters. One is pricing power. The other is its mirror image. Telling them apart takes exactly one number, and once you see it you can't unsee it.
The server hike is the real thing
Start with the half that genuinely is pricing power, because it's spectacular and it's worth understanding on its own.
Agentic AI turned out to need CPUs, not just GPUs. An AI agent breaks a goal into steps and works through them, querying databases, calling tools and APIs, running code, invoking the GPU cluster for the reasoning, checking outputs, and looping. Most of that orchestration runs on general-purpose CPUs. Intel's CEO has said some frontier customers are now buying as many as four CPUs for every GPU, and analysts expect the CPU-to-GPU ratio to narrow from roughly one-to-four in the training era toward one-to-one. That's a structural shift in what a data center is made of.
The market math follows. The server-CPU total addressable market, around $35 to $43 billion in 2026, is projected to reach north of $210 billion by 2030, a roughly fivefold expansion, with AI-specific CPUs accounting for about 86 percent of it. And supply hasn't come close to keeping up. AMD's EPYC is effectively sold out through the end of 2026. Intel is reportedly filling only about 40 percent of its yearly backlog allocations and under-shipping real demand by roughly 20 percent, with lead times stretched to six months. Both companies' stocks have run enormously on the strength of it.
Raise prices into that, a market that is sold out, supply-constrained, and begging for more, and you are exercising real pricing power. The defining signature is right there in the data: price up, and demand up with it. Buyers pay the higher price and still can't get enough. That is a company charging what its product is worth, because there is no substitute and nowhere else to go.
The client hike is the opposite, wearing the same word
Now the other half, and the details quietly invert the whole story.
On the client side, the PC side, desktop CPU shipments cratered about 20 percent. The PC market is shrinking, and Intel itself has described it as a tale of two kingdoms with the mainstream taking a beating. And yet Intel raised client prices, and reported client revenue up 13 percent year over year. The reflex says pricing power again. But Intel's own CFO gave the game away: that revenue increase came from higher average selling price, not higher volume. Fewer chips, each sold for more.
So why raise prices into a market that's actively shrinking? Not because you can. Because your costs went up. Intel's Panther Lake requires expensive high-speed LPDDR5X memory as a minimum, and the prior-generation Lunar Lake carries memory directly on the package. The memory shortage is driving both of those costs up, and it hits fully built systems like laptops hardest. Intel is passing that input-cost increase through to buyers. The hike isn't a flex. It's a cost being handed down.
The signature here is the exact inverse of the server side: price up, and units down. That is not pricing power. It is margin defense in a declining market, a company protecting its gross margin as volume falls by charging the remaining buyers more. It can look identical to strength on the income statement, revenue up, ASP up, but the mechanism underneath is the opposite of the server story, and it will behave the opposite way when conditions change.
The tell is volume
Here is the one number that separates the two, and it's worth internalizing as a general rule, not just for this story. When a company raises prices, ask whether volume moved with the price or against it.
Pricing power raises the price and keeps or grows the volume, because demand is inelastic and buyers have no alternative. That's the server CPU. Margin defense raises the price and loses volume, because the hike is a response to rising cost in a market where buyers are already leaving. That's the client CPU. Same two companies, same quarter, same reported direction on price and revenue, and completely different underlying health. The income statement alone won't tell you which one you're looking at, because both show revenue and ASP rising. The volume line tells you. Units with the price, or units against it.
That distinction matters because the two ages very differently. The server hike compounds: demand keeps rising, supply stays tight, pricing holds. The client hike is fragile: it's propping up revenue on a shrinking base, and if the memory cost eases or the volume decline accelerates, there's nothing underneath it. One is a durable advantage. The other is a bridge over a falling market.
A second tell, hiding inside the strong half
There's a subtler layer worth adding, because even the genuine-pricing-power side isn't uniform. Within the booming server market, the composition of Intel's and AMD's growth differs in a way that's its own tell.
AMD's server growth is coming substantially from volume, roughly two-thirds units rather than price, as it wins share with EPYC. Intel's is more ASP-led, and Intel is projected to lose server value share over the decade, from around 34 percent toward the low 20s, as AMD holds around 31 percent and Arm-based designs climb toward nearly half. So even on the strong side, there's a quality difference: AMD is earning its growth by taking share, while Intel is leaning harder on price as its unit position erodes. The same volume test that separates server from client also separates AMD from Intel inside the server market. Growth from units is winning. Growth from price alone, while share slips, is a milder version of the same margin-defense pattern, just in a rising market instead of a falling one.
Which is why "Intel and AMD are showing pricing power" is doubly imprecise. It bundles a real pricing-power business (server) with a cost-pass-through one (client), and inside the real one it bundles a share-gainer (AMD on volume) with a share-loser leaning on price (Intel on ASP). Three different stories, one headline.
The memory tax reaches the CPU
The client hike also completes a pattern I've been tracing through this whole buildout. The memory shortage created by AI's appetite for high-bandwidth memory keeps relocating downstream as a cost, into phones, into cars, into laptops, and now into the CPU itself. Panther Lake's memory requirement and Lunar Lake's on-package memory mean the processor now carries the memory cost directly, so a CPU price hike is, in part, the memory tax wearing a new label. A 10 percent Intel increase with an AMD follow-on is that tax handed to every laptop and desktop buyer, one more rung of the same cost working its way to the person at the checkout who has no one to pass it to.
The honest reading
None of this makes the client hike a mistake. Raising price to defend margin in a shrinking market is exactly the right move for Intel; the criticism isn't of the decision, it's of the label. Calling it pricing power misreads the company's actual position, which is a business managing decline in its oldest market while a genuinely booming one carries the story. And the server pricing power is real but not evenly Intel's, the share math says Intel is the weakening incumbent in the very market where the pricing-power narrative is strongest.
So the precise version is the useful one. Intel and AMD are raising CPU prices for two entirely different reasons at once. On the server side, because demand is inelastic and they can, real pricing power, with AMD earning it on volume and Intel leaning on price as its share erodes. On the client side, because the memory shortage raised their costs and they're passing them through a market that's shrinking, margin defense that looks like strength only until you check the volume.
You do not remove a bottleneck. You relocate it. The memory shortage just relocated into the processor, and it produced two price hikes that read as one. Only one of them is a company charging what its product is worth. The other is a company charging what it costs, and the difference between those two sentences is the whole story.