NVIDIA Corp. (NASDAQ: NVDA) stock fell 1.35% to $217.80 in Tuesday’s premarket trading as investors assessed its expansion into the custom artificial intelligence chip market. The move signals a strategic shift toward infrastructure control beyond traditional GPUs.

NVIDIA’s NVHBM Push

According to Counterpoint Research analyst Neil Shah, NVIDIA is expanding deeper into the custom AI chip market with NVHBM. This technology moves the memory controller from the accelerator chip into the base die of a high-bandwidth memory stack. It also replaces the standard JEDEC memory bus with NVIDIA’s proprietary die-to-die connection.

Shah noted that this approach gives NVIDIA control over the memory controller, interface, and link protocol. Memory suppliers would manufacture the base die based on NVIDIA’s design. This strategy allows NVIDIA to sell its networking fabric, rack architecture, software, and memory subsystem to companies developing custom accelerators without designing the accelerator itself.

Customer Adoption

Amazon.com Inc.’s (NASDAQ: AMZN) Annapurna Labs is the first disclosed customer. It plans to use NVHBM with its Trainium4 AI chip. MediaTek also adopted NVIDIA’s NVLink Fusion platform, providing another channel for reaching custom chip customers.

However, Shah questioned some of NVIDIA’s performance claims. The company compared NVHBM with HBM4E, which is not yet shipping, and did not disclose its full testing baseline. Only the claimed 67% reduction in physical interface area can be checked using available pin-count data.

Competitive Landscape

NVHBM places NVIDIA directly in a market served by Broadcom Inc. (NASDAQ: AVGO) and Marvell Technology Inc. (NASDAQ: MRVL). Broadcom could benefit from wider adoption of custom AI chips but lacks a comparable base-die offering. Marvell outlined a similar architecture in December 2024 and has joined NVLink Fusion. This leaves Broadcom as the only major custom chip design company without a direct NVIDIA partnership.

Technical Analysis And Outlook

Despite the early decline, NVIDIA remains in a longer-term uptrend. The stock trades 4.5% above its 50-day simple moving average of $208.62 and 11.3% above its 200-day SMA of $195.96. However, it trades about 0.3% below its 20-day SMA of $218.75.

The relative strength index stands at 54.25, suggesting consolidation. Key resistance stands near $228, while support sits near the 200-day SMA. The stock trades at 27.9 times earnings. Citigroup and Mizuho raised their price forecasts to $315 on Aug. 27. JPMorgan increased its forecast to $320.

What the Numbers Show

The divergence between NVIDIA’s technical position and immediate price action highlights investor caution. While the stock remains significantly above its long-term moving averages (50-day and 200-day), trading below the shorter-term 20-day SMA indicates near-term pressure. This suggests that while the long-term structural trend remains bullish, the immediate market reaction to the NVHBM strategy and broader tech sector weakness is causing short-term consolidation.

Nvidia Corp (NASDAQ: NVDA) has revived its Rubin CPX AI accelerator program after market speculation suggested it had been dropped from the roadmap. Analyst Ming-Chi Kuo stated on Monday that industry checks confirm the chip giant plans to begin production in the first quarter of 2027.

The revived Rubin CPX features a substantially redesigned architecture aimed at delivering stronger prefill performance. Prefill is the stage of AI inference where a model reads and processes input before generating a response. Kuo noted that the new design includes significant changes to both GPU specifications and rack architecture compared to the earlier CPX iteration.

Technical Specifications and Architecture

The updated Rubin CPX will feature 168GB of HBM4 high-bandwidth memory per GPU. This sits between the 288GB of HBM4 memory found in Nvidia’s standard Rubin GPUs and the 128GB of GDDR7 memory in the previous CPX design.

Kuo explained that the revived CPX will move into a standalone MGX ETL rack rather than sharing space with Rubin GPUs. Customers can configure systems with 64, 128, 192, or 256 CPX GPUs. Within each 64-GPU module, eight compute trays house eight CPX GPUs each, alongside a switch tray. NVLink will connect the eight GPUs within each tray, while Ethernet handles communication between trays and rack modules.

Integration with Vera Rubin Systems

Rubin CPX is expected to work alongside Nvidia’s Vera Rubin NVL72 systems rather than replacing them. Kuo said Nvidia recommends a 1:1 ratio of CPX to Rubin GPUs. In this configuration, CPX handles the prefill stage and generates the KV cache before transferring that information to Rubin over Ethernet RDMA for the decode stage.

This announcement follows Nvidia’s GTC 2026 event, where the company appeared to remove CPX from its roadmap, instead highlighting Groq 3 LPUs and LPX racks. Nvidia did not immediately respond to requests for comment.

Financial Context

Nvidia reported $96.22 billion in second-quarter revenue in August, marking a 106% increase from a year earlier. This figure surpassed the Street consensus estimate of $92.18 billion. The company also confirmed that Vera Rubin was ramping into full production, with CoreWeave, Nebius, Microsoft Azure, Google Cloud, and Oracle Cloud among its partners.

Nvidia shares closed at $220.50 on Monday, up 1.36%. In after-hours trading, the shares were down 0.10%. According to Benzinga Edge Rankings, Nvidia ranks in the 98th percentile for growth and maintains positive price-trend ratings across short-, medium-, and long-term time frames.