Memory Prices Soar, 600W GPU Connector Melts, and China’s C700 Targets H100

The GPU and memory market is becoming increasingly interesting in 2026. Memory prices have surged enough to attract new potential entrants, NVIDIA’s 600W RTX PRO 6000 has suffered another 16-pin connector failure, and Chinese GPU maker MetaX is preparing its next flagship accelerator to compete with NVIDIA’s H100.

These stories may look unrelated, but they all point to the same trend: demand for computing hardware is growing rapidly, while memory, power efficiency and supply-chain control are becoming increasingly important.

Another Semiconductor Company May Enter the Memory Market, but Nobody Knows Who

Memory prices have soared over the past year, bringing enormous profits to the industry's biggest manufacturers. Now, a mysterious semiconductor company is reportedly preparing to enter the memory market, although nobody seems to know exactly who it is.

The global market is dominated by Samsung, SK hynix and Micron, while China's CXMT has been expanding rapidly. Taiwan-based companies such as Nanya Technology also remain active, although with much smaller market shares.

According to semiconductor leaker AogT, another company is preparing to enter the memory business, but its identity is considered too sensitive to reveal. That immediately led to speculation among technology enthusiasts.

Intel was one of the first names mentioned, but the leaker reportedly denied that Intel was the company in question. Intel does have plans related to AI-focused memory technologies, but those products are still several years away.

Other speculation has pointed toward SMIC and YMTC. YMTC would not be particularly surprising because it is already deeply involved in NAND flash memory. SMIC has some flash-related business as well, but there is little public evidence that it plans to become a major DRAM manufacturer.

Another possibility is TSMC. Some observers believe AogT may have connections to TSMC, although that has not been independently confirmed. TSMC certainly has the financial resources and technical expertise to explore another semiconductor segment, while its relationships with major AI chip companies could potentially provide strategic advantages.

There is also speculation involving a major Chinese technology company whose name begins with H. Rumors suggest it has the demand and resources to support a domestic memory business, while another recent report mentioned a related company using a 28nm process.

That detail is difficult to interpret. A 28nm process would be extremely old for modern DRAM production, while using it for a new high-end logic product would also raise questions.

For now, none of these theories has been confirmed.

What is clear is why companies are suddenly interested in memory. AI infrastructure has pushed demand for DRAM and high-bandwidth memory to extraordinary levels, making the business far more attractive than it has been during previous downturns.

The problem is that competing with Samsung, SK hynix and Micron requires enormous investment, advanced manufacturing technology and years of experience. High prices may attract newcomers, but building a competitive memory business is another matter entirely.

600W RTX PRO 6000 Suffers Another Melting 16-Pin Power Connector

While memory manufacturers are enjoying booming demand, NVIDIA's highest-end professional GPUs are facing a very different problem: power delivery.

The RTX PRO 6000 Blackwell Workstation Edition has reportedly suffered another failure involving its 12V-2x6 power connection. This time, the damage occurred on the power-supply side rather than the GPU side.

According to repair records, the connector inside an ASRock Taichi TC-1650T 1650W power supply reportedly melted during operation. The power cable was said to have been properly inserted at both ends before the incident.

The RTX PRO 6000 is a monster by workstation standards, featuring 96GB of memory and a maximum power consumption of 600W. It uses a single 16-pin PCIe CEM5 connector, putting its power requirements in roughly the same class as the GeForce RTX 5090.

What makes this incident particularly interesting is the power supply itself. The ASRock Taichi TC-1650T supports ATX 3.1 and includes TempGuard as well as an NTC temperature sensor intended to monitor the temperature of the 12V-2x6 cable.

In theory, such protection should help identify abnormal heat before the connector reaches the point of physical damage. In this case, however, the connector reportedly continued operating until the socket itself melted.

It is also not the first reported problem involving this GPU and the same power supply. In March, another user reported that a 12V-2x6 cable had burned at the GPU end. PNY reportedly inspected the graphics card and found no obvious fault, while ASRock replaced the damaged cable.

The latest incident ended relatively well because the damage happened inside the power supply and the approximately $16,000 RTX PRO 6000 reportedly escaped serious damage.

Still, repeated incidents involving high-power GPUs and compact 16-pin connectors are not something enthusiasts can simply ignore. Even when the cable is properly installed and the PSU includes temperature monitoring, failures can apparently still occur under extreme loads.

One possible improvement would be independent temperature monitoring at both ends of the connection, which could potentially provide earlier warning when abnormal heat begins to develop.

For 600W-class GPUs, the power connector is no longer just a technical detail. It has become an important part of the overall hardware design.

China’s MetaX C700 GPU Targets NVIDIA H100 Performance

At the same time, China's domestic GPU industry is moving toward a new generation of high-end AI hardware.

Chinese GPU manufacturer MetaX recently revealed new progress on its next flagship accelerator, the MetaX C700. According to the company, the chip's core design and functional verification are close to completion, after which it will move toward tape-out, software adaptation, customer testing and mass-production preparation.

The C700 began development in April 2025 and is designed to deliver major improvements in computing, memory, communication and power efficiency compared with its predecessor. MetaX also says it will support low-precision formats such as FP4, which are increasingly important for large-model inference.

The company sees supply-chain control as one of the C700's major advantages. MetaX says it has domestic solutions covering wafer manufacturing, packaging and high-capacity memory, allowing it to build a more self-controlled supply chain from chip design through testing.

The C700 follows MetaX's strategy of mass-producing one generation while developing the next. Its current C600 entered mass production in May 2026 using a domestic 7nm process, with 144GB of HBM3e memory and around 1,000 TFLOPS of FP8 performance.

MetaX is also pushing its own MetaXLink high-speed interconnect technology. The company says the technology has already been used at thousand-GPU scale and is being prepared for much larger deployments.

According to industry estimates, the C700 could enter tape-out testing in 2026 and begin mass production in the second half of 2027. MetaX says its overall performance target is the level of NVIDIA's H100.

That comparison will ultimately need to be tested through real-world workloads. Raw compute figures are only part of the equation for an AI accelerator. Memory bandwidth, interconnects, drivers, software compatibility and application optimization can all have a major impact on actual performance.

Even so, the C700 is another sign that China's GPU industry is moving beyond simply designing chips on paper. With the C600 already in mass production and the C700 moving through development, MetaX is trying to build a complete product pipeline for high-end AI computing.

The broader hardware market is therefore becoming much more competitive—and much more complicated. Memory shortages are attracting new companies, 600W GPUs are putting increasing pressure on power delivery, and domestic AI accelerators are beginning to challenge established products on their home turf.

For PC users and AI customers alike, the next few years could bring some major changes to the hardware market.