Hardware Outlook · Canada · September 20, 2026

No New Gaming GPUs Until 2028, No New CPUs Until 2027: What the Hardware Drought Means for Your Next Build

Three roadmaps just went blank at once. The GPU you can buy today is the newest GPU you will be able to buy for roughly 30 months, and it has to carry you through GTA 6’s PC launch. We stacked Nvidia’s, AMD’s, and Intel’s schedules into one timeline and ran the buyer math.

Every piece of conventional PC buying advice assumes the same rhythm: something better is always about twelve months away, so patience is free. That rhythm is dead, and September 2026 is the first month where you can see the full replacement at once. The Information reported that Nvidia will ship no new gaming GPUs in 2026 and that the RTX 60 series has slipped to 2028. TechSpot followed with word that nearly all of AMD’s RDNA 5 stack slips with it. Within the same fortnight, both Intel Nova Lake and AMD Zen 6 desktop confirmed 2027 arrivals, and the memory makers told Intel’s CEO there is no price relief until 2028.

Each story has been reported separately. Stacked together, they describe a window with no exit ramps: from now until early 2028, anyone building or upgrading a PC is choosing from hardware that already exists, at prices that are moving up rather than down. The practical question stops being “should I wait for next gen” and becomes “which current part survives a 2028 horizon, and what should I pay for it today.”

The drought, stacked on one timeline

The three clocks run on different schedules, which is exactly why the picture has been slow to land. The GPU lane is empty the longest. The CPU lane fills in during 2027. The memory lane, the root cause of the other two, stays red through all of it.

Count the GPU lane. The newest graphics card anyone can buy today launched in the first half of 2025. The next generation is reported for 2028, with late 2027 as the optimistic case floated by GameGPU after Gamescom. That is a 30-month freeze from this month, and it stretches the generational cadence PC gamers have lived with for eight years by roughly half:

The mid-cycle rescue that normally breaks a long wait is also gone. The RTX 50 Super refresh was not delayed. It was completed, shipped to board partners, and then frozen in place. VideoCardz reported in July that at least one partner physically holds the cards, and Gamescom confirmed the cancellation outright in late August. For buyers who kept telling themselves a VRAM-rich Super refresh was coming to fix the 12GB midrange, that plan no longer exists.

One root cause: memory went to the datacenter

The drought is not three coincidences. It is one shortage propagating through three product lines, and the mechanism is visible enough to trace step by step.

The arithmetic that killed the Super refresh is worth seeing directly, because it also explains why the next generation slipped. A 24GB card on a 256-bit bus carries eight memory chips:

CPUs caught the same disease by a different route. When Tom’s Hardware reported both desktop lineups slipping in February, it noted Intel had already been shifting production capacity from consumer chips toward datacenter parts, and that RAM had tripled in price while SSD and GPU prices climbed alongside. AMD’s server version of Zen 6, EPYC “Venice,” shipped first and has reportedly captured 46.2 percent of server CPU revenue. The desktop version of the same silicon comes later, because datacenter orders pay more and arrive first.

The CPU drought is shorter, and its dates come with caveats

Nothing on the CPU side is vendor-confirmed, so the honest presentation separates paper from rumor. One fact is hard: AMD’s own developer documentation, spotted by leaker InstLatX64 on August 27, confirms Zen 6 desktop chips, codenamed Olympic Ridge, run on the existing AM5 socket. Everything about dates is leaks, though several independent ones now converge.

Two buyer-relevant details sit inside those leaks. First, Nova Lake lands on a new socket that Intel promises will support future generations, so buying it early does not strand you. Second, and more important for anyone on AM5 today: waiting is cheap in a way it never is on the GPU side, because a Zen 5 X3D bought now drops into the same motherboard that will take Zen 6 in late 2027.

Prices are already moving the wrong way for waiters

The standard argument for waiting assumes the current stack gets cheaper while you wait. During a shortage with production cuts, the opposite is happening, and the receipts are in the trackers. Tom’s Hardware’s monthly median of every in-stock RTX 50-series listing on Newegg, as of June 2026:

Canada takes a steeper hit at the top. The cheapest verified RTX 5070 Ti in the country in August was an ASUS Prime at $1,089.99 CAD at Canada Computers, with Best Buy Canada listing the same card up to $1,409.99 CAD. By September, price trackers had the card at $1,829 CAD new on Amazon.ca, 60 percent above its $1,140 CAD twelve-month average, while the RTX 5070 listed at $1,139 CAD against a $771 average. The supply side explains the direction: PCMag reported a 30–40 percent production cut for early 2026, and TechTimes reported in June that supply had been trimmed further with prices climbing and no refresh coming to relieve them. Nvidia is also reportedly prioritizing lower-VRAM SKUs, which squeezes exactly the cards this window pushes you toward.

The budget tier tells the same story from below: it rose roughly 40 percent in three months this summer, which is what made a five-year-old RTX 3060 the rational budget buy in our pricing breakdown earlier this week. When the floor moves up 40 percent in a season, the assumption that time is on the buyer’s side needs evidence, and right now the evidence points the other way.

The GTA 6 collision

One demand event sits inside this window and changes what the purchase is for. Grand Theft Auto VI launches on console November 19, eight weeks from now, and no PC version launches with it. Rockstar’s pattern since GTA 5 puts the PC port a year or more later, and the trackers’ consensus is late 2027 or early 2028, with reports from TechPowerUp and Wccftech that console shortages could pressure Rockstar to move the PC launch earlier. Either date lands the single most hardware-hungry game of the decade inside the GPU drought.

That turns the purchase decision around. A PC built this holiday is not a machine you will upgrade before GTA 6 arrives on PC. It is the machine you will play GTA 6 on, through its first years of patches and online content, with no new GPU generation appearing to rescue an underbought card. The console reference point matters too: the game runs at a locked 30fps on PS5-class hardware with 16GB of shared memory, which tells you where the minimum spec anchors and where the comfortable PC headroom starts. Whatever tier you choose, choose it for the 2027–2029 software curve on frozen hardware, not for this month’s benchmarks.

Which tier survives a 2028 horizon

Nobody can benchmark a card against games that have not shipped, but the logic for a 30-month carry is checkable. Games through 2028 will keep targeting current consoles, because the next console generation is itself reportedly slipping under the same memory costs. That keeps the software ceiling roughly where it is: comfortable at 16GB of VRAM, strained above it, compromised below it. Performance headroom on top of that is what you actually pay for, tier by tier:

Within the 16GB tier, the tiebreakers are raster headroom and the upscaling stack, because frame generation is the lever that stretches a frozen card through two more years of game engines. Nvidia’s DLSS multi-frame generation currently leads there, which is a real argument for the 5070 Ti and 5080 over the RX 9070 XT even where raster performance trades blows, and AMD’s FSR 4 is improving on its own schedule. The 5090 remains in a category of one at more than double its MSRP, a card that makes sense only if money is genuinely no object. The defensible default for a build meant to last to 2028 is the cheapest 16GB card you trust, bought before the shortage prices another 20 percent onto it.

The 2022 instinct is the wrong instinct

The reason “just wait, prices always crash” feels plausible is 2022, when the mining bubble burst, used cards flooded the market, and the RTX 40 series arrived on schedule that October. Waiting then was rewarded. The mechanisms this time are inverted on every axis that mattered then:

The decision, one component at a time

The wait decision is not one decision. It is three, and the drought gives them different answers:

Five public signals will tell you when any branch of this flips. AMD announcing and pricing the AT2, the only rumored GPU inside the window. Nvidia or a board partner confirming an RTX 60 production date, which moves the right edge of the timeline. The 3GB GDDR7 chip price, because if it falls back toward twice the 2GB price rather than triple, cancelled products can un-cancel. Hyperscaler capital-spending guidance flattening or dropping on earnings calls, the one force that bends the whole shortage. And Rockstar announcing GTA 6’s PC date, the day the midrange demand spike begins. Each will update in public, whether or not anyone prices it in ahead of time.

The drought is not a rumor anymore. The Super cards are built and boxed and frozen at board partners, the CPU leaks converge on 2027 from four directions, and the memory makers have told the industry’s biggest CPU vendor there is no relief until 2028. For the person staring at a parts list this week, that collapses to one sentence: the PC you buy now is the PC you keep, so buy the tier that survives the window, buy it at a fair price when you see one, and stop paying the waiting tax on a next generation that does not exist yet.