NVIDIA has slashed gaming GPU production. To cope with the memory constraints, NVIDIA reportedly reduced gaming and consumer GPU production by an estimated 30-40% in early 2026, redirecting silicon and memory allocation to its far more profitable data center AI business . This is not a temporary adjustment — it reflects a strategic pivot where AI GPU revenue now dwarfs NVIDIA's gaming division
.
AMD has paused new GPU launches until 2027 for the same underlying reason: there simply isn't enough GDDR7 memory available to launch new consumer cards . This effectively means an entire generation of potential GPU releases has been shelved.
TSMC's CoWoS advanced packaging is a chokepoint. Even when chips and memory exist, they must be physically assembled. TSMC's Chip-on-Wafer-on-Substrate (CoWoS) packaging capacity is fully allocated through mid-2027 . Compounding this, Japan's Ajinomoto — which supplies more than 95% of the global market for ABF substrate, an insulating material crucial for advanced packages — raised prices 30% in 2026 and projects a supply gap extending through 2027
.
Every Blackwell GPU is already spoken for. Every unit NVIDIA can package through mid-2027 is already contracted to hyperscale cloud providers like Microsoft, Google, Amazon, and Meta . The gap between demand and supply is currently estimated at 1.4x to 1.6x, and is not expected to close soon
.
The impact on consumers has been immediate and severe. GPU retail prices have climbed to 2–2.5 times launch MSRP in some regions . In early August 2026, Asus and Gigabyte confirmed a second price increase within the year across both NVIDIA RTX 50-series and AMD Radeon RX 9000 lines
. The RTX 5090D V2 alone saw increases of US$592 at Gigabyte and US$666 at Asus. Other models were also affected: the RTX 5080 rose by US$207 to US$222, the RTX 5070 Ti by US$222 to US$252
.
Japanese distributor CFD Sales flagged 20–40% price hikes on certain Gigabyte orders effective August 1, 2026 . In China, RTX 5070 Ti listings jumped 21.8% and RTX 5070 listings climbed 16.2% inside a single 24-hour window in late July, while RTX 5060 listings reportedly rose $75 overnight
.
Entry-level and budget GPUs are hit hardest because they carry the thinnest margins. The proportional cost impact of memory price increases is steepest at the low end, making affordable PC building nearly impossible . PC Partner Group has publicly warned that worsening shortages of GPUs, graphics memory, CPUs, and other components will raise PC prices broadly in late 2026, especially for entry-level models
.
Consumer PC demand is weakening as higher prices push buyers out of the market. However, this softening does nothing to relieve supply, because manufacturing capacity is already locked into AI contracts that prioritize hyperscalers over individual consumers .
NVIDIA — CFO Colette Kress warned on an earnings call that it will be difficult to boost consumer GPU shipments for multiple quarters, with retail shortages likely to persist past summer 2026 .
NVIDIA data-center GPU lead times — For H100, H200, and GB200 models, lead times remain at 36–52 weeks as of mid-2026. All Blackwell capacity coming online through at least August and September 2026 has already been booked .
Broadcom flagged that TSMC's advanced-node capacity will limit chip supply into 2026, affecting not just GPUs but high-performance networking and custom ASICs .
Inventec warned that the AI memory crunch has reached servers, with lead times for key memory components exceeding 40 weeks .
Aetrix reports that even 28nm-node semiconductor lead times have stretched to 20–30 weeks. Mature-node FPGAs, automotive MCUs, and industrial networking chips are under visible stress as the AI buildout consumes not just cutting-edge capacity but the broader electronics supply chain .
Storage and server segments remain strong — AI infrastructure spending has spread supply tightness across enterprise memory, high-speed networking (with 800G optical transceiver lead times pushing past 40 weeks), power ICs, and storage. This strength stands in stark contrast to the softening consumer PC side .
PC Partner Group — the parent company of Zotac and Inno3D, and a major NVIDIA board partner — reported H1 2026 results that sharply illustrate the diverging market dynamics.
Attributable profit soared ~118% to at least HK$500 million (approximately SGD $86 million), more than double the HK$250 million reported in H1 2025 .
The earnings surge was driven entirely by higher average selling prices (ASPs) for its branded Zotac and Inno3D graphics cards. The scarcity of GPU supply enabled PC Partner to command significantly higher prices for every card it could secure . Revenue grew while shipment volumes were flat or declining — the monetization of scarce GPU allocation at inflated prices more than offset lower unit sales
.
Analysts at DBS and KGI project FY26 double-digit revenue growth for PC Partner, led by ASP expansion. A key factor in their outlook is PC Partner regaining access to top-tier RTX 5090 allocation after its Singapore headquarters move and SGX secondary listing .
The bottom line is clear: the companies that hold access to scarce GPU supply are making record profits on dramatically reduced volumes. For end consumers, that translates to sticker shock and extended waits. For investors, it means a starkly different picture depending on which side of the supply chain you sit on.
The 2026 GPU shortage is not a repeat of past cycles. It is a structural memory and packaging crisis driven by AI demand that shows no signs of easing. It is pushing consumer PC prices sharply higher while dampening demand — but companies with access to scarce GPU allocation are seeing record profits. For buyers, the best strategy may be to wait unless you find a card at a price you can stomach, because the structural constraints show no signs of easing in the near term.