The Midrange GPU Is Being Squeezed Out — And the Memory Crisis Is Finishing the Job

Three Product Decisions, One Ugly Conclusion

The midrange discrete GPU — the $300–$450 card that powered two decades of PC gaming — is being methodically crushed between a memory crisis above and improving integrated graphics below. It isn't dying with a dramatic announcement. It's disappearing through a series of product non-events and quiet portfolio pivots, and three of them landed within roughly four months of each other.

That convergence is the story.

Intel Pulled the Plug on Its Own Card — Silently

The Intel Arc B770 was a no-show at CES 2026, and Intel's board partners had yet to receive test samples, with production delays and the memory crisis hanging over its uncertain launch. That silence was already telling. Then, in early February, the other shoe dropped.

A report attributed to anonymous sources claims Intel has permanently shelved the Arc B770 consumer graphics card, describing the project as no longer financially viable — and the framing matters: the claim is not that the GPU could not be built, but that launching and sustaining the product in today's market would not deliver an acceptable return.

The silicon itself was real. The BMG-G31 die that would power the Arc B770 definitely exists — Intel formally confirmed it in December 2025 software updates, and firmware for the GPU appeared in HP Panther Lake laptop drivers discovered at CES 2026. A chip with a confirmed die, existing drivers, and a leaked spec sheet simply never shipped as a consumer product. That's not a delay. That's a verdict on whether the midrange GPU business makes sense right now.

The BMG-G31 die is now being redirected: the latest rumors point to an imminent release of the AI-focused Intel Arc Pro B70, powered by the same silicon. That's the tell. The hardware didn't go away. It got repriced upward, repackaged for workloads that can actually absorb the memory cost, and pulled out of the consumer channel entirely.

AMD's 16GB Pivot — In Reverse

AMD's trajectory is similarly revealing, just playing out more gradually. When the Radeon RX 9060 XT launched at Computex 2025, the 8GB model was offered at a lower price point than the 16GB version — the implication being that 16GB was the real product and 8GB was the entry ramp.

By early 2026, that framing had quietly inverted. According to reports, AMD is shifting its Radeon focus to 8GB graphics cards, and for its RDNA 4 lineup that means a single GPU. Radeon GPU prices increased in early 2026, with further hikes expected soon, affecting supply and market dynamics.

For AMD's new RDNA 4 generation, the 8GB pivot means a single model: the mainstream Radeon RX 9060 XT 8GB. At the time of the RX 9060 XT's original launch, the focus was on the 16GB model, with 8GB cards primarily reserved for system integrators. The portfolio shift isn't subtle — it's a public acknowledgment that GDDR6 costs have made 16GB mainstream cards economically painful to build at prices consumers will actually pay.

AMD's partners can still obtain 16GB cards, but the focus shifts to the Radeon RX 9070 XT, with the baseline Radeon RX 9070 seeing its production significantly reduced. In other words: the midrange 16GB card is being rationed upward toward a price point where margins survive. The low-to-mid price tier is being hollowed out.

The Memory Crisis Is the Root Cause

None of this is mysterious. All available DRAM is currently being consumed by the AI data center buildout, which is absorbing the entire memory supply — and NVIDIA, AMD, and other GPU manufacturers are reportedly considering reducing their production of mid-to-high-end gaming cards, as memory now constitutes an unusually large portion of total cost.

Reports indicate that price increases for Radeon GPUs occurred in early 2026 for AMD's AIB partners purchasing GPU and memory bundles — and additional price increases are coming.

The report on the B770's cancellation points directly at the state of video memory supply and pricing: if VRAM remains constrained and costly, a 16GB design becomes expensive to build, and the final shelf price has to fight against well-established competitors.

This is the structural problem. A 16GB GPU at affordable price points requires cheap GDDR6. Cheap GDDR6 requires memory fabs that aren't already committed to HBM for AI accelerators. South Korean memory giants like Samsung and SK Hynix are currently facing supply constraints — with smaller OEMs and channel distributors informed to expect limited fulfillment through early 2026. That supply crunch cascades directly into consumer GPU pricing.

IDC expects 2026 DRAM supply growth to remain below historical norms, and the result of the supply/demand imbalance is twofold: DRAM and NAND prices have risen sharply, and availability is limited, forcing device manufacturers to navigate a fluid situation.

The consensus is clear: it's hard to launch the Arc B770 into a market where memory prices remain elevated for an extended period. That's not a launch delay. That's a generational skip.

What's Actually Left

The GPU market isn't disappearing — it's bifurcating hard. The economics still work at the extremes.

At the bottom, the 8GB cards (think under $300) can absorb elevated GDDR6 costs because there's less of it per unit. At the top, workstation and AI-targeted parts like Intel's Arc Pro B70 carry larger memory configurations and command margins that justify the memory spend. As DRAM becomes a rarer commodity, GPUs that do ship carry a premium built in. That premium is sustainable when the buyer is a data center or a professional workstation — not a gamer deciding between two consumer cards at similar price points.

The tier that's being squeezed out is specifically the $300–$450 discrete card: too much memory to be cheap, not enough margin to justify the memory cost, competing against integrated graphics that keep improving. Intel's Panther Lake Arc and AMD's Ryzen RDNA iGPUs are legitimately good at 1080p and increasingly credible at 1440p — which is the exact performance bracket where a $300 discrete card used to justify its existence.

Products for the PRO and workstation segment may simply be more lucrative for GPU manufacturers than gaming cards — and the industry is voting with its roadmaps.

This Is a Structural Shift, Not a Blip

The midrange discrete GPU wasn't killed by a single bad decision. It's being killed by three simultaneous pressures that reinforce each other: memory supply captured by AI buildout, improving APU performance reducing demand at the low end, and a GPU market in which the high-margin action has moved to professional and AI cards.

Intel's B770 cancellation, AMD's 8GB pivot, and the DRAM price trajectory all point the same direction. The gap between a budget card and a premium card is widening — and right now, there isn't much in the middle that pencils out for either the manufacturer or the buyer. That's a problem for PC gaming as a mainstream activity, and it isn't going to resolve itself until memory supply catches up to demand. The consensus timeline for normalization is multiple years away.

Until then, the most profitable segment PC gaming ever had is quietly going dark.