If you blinked through the past month, you missed one of the most interesting shifts in PC hardware in a decade. Computex 2026 is barely in the rearview and the aftershocks are still rolling. NVIDIA basically walked into the PC market through the front door, memory prices are doing things nobody predicted, and there is a Chinese chipmaker nobody in the West talks about quietly dropping server parts that actually look competitive.
Let me unpack what actually matters.
NVIDIA Just Stopped Being a Data-Center-Only Company
The big one. Jensen Huang stood on stage at Computex and unveiled the RTX Spark superchip — and I do not think people outside the enthusiast bubble fully grasp what this means. It is a 70-billion-transistor chip built on TSMC 3nm, fusing a Blackwell GPU with 6,144 CUDA cores and a 20-core Grace CPU into one unified SoC. The trick is the shared memory pool: 128GB of unified memory at 300GB/s bandwidth. CPU and GPU stop copying data back and forth. It just lives in the same space.
And it is Windows on Arm. That part is going to rattle the industry more than the hardware specs.
Microsoft, Dell, HP, ASUS, Lenovo, Acer, and MSI are all lining up to ship RTX Spark machines this fall. Huang actually held up a tiny MSI prototype and said an AI agent could "run 24/7, meter free" — a direct jab at the cloud API pricing model. Whether local AI agents turn out to be something normal people actually use is another conversation, but the hardware bet is real.
Here is what I am less optimistic about: the Arm transition on Windows has a long track record of promises and delays. Remember Windows RT? Yeah. Qualcomm has been chipping away at this for years with Snapdragon X and the adoption curve is still modest. NVIDIA brings better GPU drivers and actual developer mindshare, but the ecosystem gap does not vanish overnight. Games compiled for x86 will need emulation. Some creative software stacks will stutter. I would not call this a slam dunk yet — but the raw silicon is impressive.
Meanwhile, on the data center side, Vera Rubin has entered full production with shipments starting fall 2026. Samsung, SK Hynix, and Micron are confirmed HBM4 suppliers. Samsung stock jumped 10% on the news. Huang said a single Grace Blackwell rack now assembles in about five minutes — a flex on manufacturing efficiency more than anything, but it shows NVIDIA is thinking about deployment speed at scale.
Intel Tries a Different Angle with Crescent Island
Intel's Computex strategy was quieter but honestly more interesting to me from an engineering standpoint. The Crescent Island GPU skips HBM entirely and runs on LPDDR5X — cheap, widely available memory that does not compete with the HBM shortage nightmare.
That is the kind of pragmatic tradeoff you do not see often enough in AI hardware. HBM is the bottleneck of the moment. Every foundry is fighting over it. Intel decided to build an inference GPU that does not need it. The tradeoff is lower memory bandwidth, but for inference workloads — especially at the edge or in cost-sensitive deployments — that is an acceptable compromise. I would not bet my data center on it, but for a second source option in a market starved for AI compute? It has a place.
Hygon Casually Dropped a 128-Core CPU Nobody Asked About
Chinese chipmaker Hygon announced the C86-5G server CPU line: 128 cores, 512 threads using SMT4, 104 PCIe 5.0 lanes, AVX512, and native AI acceleration for INT8 and BF16. They are already in mass production, shipping in standard rack servers and even liquid-immersion cabinets that pack up to 80,000 cores per cluster.
Hygon also introduced a DCU GPU accelerator targeting NVIDIA's A100 — HBM memory, high-speed interconnects, the whole package. And a full networking stack with 400Gb/s links and sub-microsecond latency, plus an 800Gb/s switch.
Now, the catch: Hygon products are essentially China-only. Export restrictions, domestic procurement mandates, and the geopolitical wall mean most of us will never spec these into a build. But the fact that this ecosystem exists at all tells you where the server market is heading over the next five years. Two parallel supply chains. That is going to influence pricing and availability on the Western side more than people realize.
Unverified industry rumor, for reference only — details sourced from vendor announcements and WCCFTech.
Memory Prices Are Not Coming Down Anytime Soon
I saved this one because it directly hits anyone building a PC right now.
The memory market has flipped completely. TrendForce numbers are brutal: DRAM prices surged 172% year-over-year by Q3 2025. Q1 2026 saw conventional DRAM contract prices climb another 55-60%. PC DDR4/DDR5 blended pricing jumped an estimated 105-110% in a single quarter. 1TB consumer SSDs that cost around $45 in late 2024 are now hovering near $90. Some DDR5 kits saw month-over-month spikes of up to 100%.
This is not a normal cycle. The cause is structural: AI data center demand has consumed wafer capacity. HBM gets priority, server DRAM comes second, and consumer memory is left scraping the bottom of the allocation. SK Hynix already confirmed its entire 2026 RAM production is booked. Micron's CEO said tightness will continue into 2027.
For anyone planning a build: DDR5 and SSD deals are going to be rare through at least Q4 2026. If you see a decent price on a 2TB NVMe drive today, I would not wait. QLC SSDs might offer some relief on the low end as they scale up — but that is a cold comfort when you are trying to build a gaming rig on a budget.
The Midrange GPU Battle: RTX 5060 vs RX 9060 XT
I have been watching the sub-$300 segment heat up. The RTX 5060 and RX 9060 XT are both on shelves now, and the comparisons are interesting. UserBenchmark and NanoReview both show the RX 9060 XT pulling ahead in rasterization — especially the 16GB variant — while the RTX 5060 retains its lead in ray tracing and upscaling quality through DLSS 4.
Honestly, at this price point, the decision comes down to what you play. Competitive shooters at 1080p? The AMD card gives you more raw frames per dollar. Single-player titles with ray tracing? NVIDIA has the edge. Both are good cards. Neither is a slam dunk. The 8GB VRAM on the RTX 5060 is starting to feel tight for 2026 titles, and that is worth keeping in mind if you want the card to last three years.
Quick Add-On Note on Laptops
Not much new in the traditional gaming laptop space this week — the Computex spotlight was almost entirely on the RTX Spark Arm machines. ASUS and Lenovo showed some refreshed AMD Strix Point and Intel Arrow Lake HX laptops, but spec bumps rather than real redesigns. If you are shopping for a gaming laptop right now, the usual advice holds: wait for the Arm models to land in fall so you can compare real-world benchmarks before committing.
That is the state of hardware heading into July. If you are building a new rig, lock in storage and memory prices when you see a window. The GPU market is as competitive as it has been in years on the midrange, which is genuinely good news. And the Arm PC transition is happening — slower than the hype suggests, but faster than skeptics expected.
By the way, if you ever need to run the numbers on a big purchase — or just want to sanity-check your monthly budget — Decision Calculator handles loan payments and affordability pretty cleanly. No frills, just the math.
What are you watching more closely in the second half: the RTX Spark laptops or the midrange GPU battle? I am leaning toward Spark because it changes more than just frame rates — but I would love to hear where your head is at.
Top comments (0)