Intel's Crescent Island GPU Enters Customer Sampling as Rack Densities Hit 46,000 Cores

Intel revealed its GPU D-Series accelerator with up to 480 GB LPDDR5X and Xeon 6+ racks exceeding 46,000 cores, as shares fell 2.8%.

Intel used the Open Compute Project Global Summit to pull back the curtain on its GPU D-Series, the accelerator previously known under the Crescent Island codename, while simultaneously rolling out Xeon-based rack architectures built for AI data centers. The disclosure came as the chipmaker confirmed that sampling with select customers has already begun.

At the heart of the D-Series is a memory configuration of up to 480 GB of LPDDR5X paired with 1.5 TB/s of bandwidth. Intel rates compute throughput at 1.3 petaflops in FP8 format, drawing 350 watts. Alongside the accelerator, the company outlined server systems targeting extreme packing densities: certain Xeon 6+ configurations scale beyond 46,000 cores per rack, with total power draw of 133.45 kW.

Tan Flags Rapid Adoption of Chinese AI Models

While reworking its data center hardware lineup, Intel's leadership is keeping a close watch on the software side of the AI race. CEO Lip-Bu Tan, speaking on a DBS-organized panel, said leading Chinese AI models are being adopted at a rapid clip, driven by their capability and markedly lower cost per token.

The company's industrial commitments in the US remain in place as well. Tan confirmed on October 7 that Intel intends to stay involved in Elon Musk's Terafab project, according to Reuters. Musk has said Tesla and SpaceX would build and operate the complex, with TSMC potentially subleasing parts of the facility. Intel's specific role in the arrangement was left undefined.

Sector-Wide Pressure Weighs on the Stock

Monday's trading saw Intel shares retreat 2.8% to close at EUR 90.86, caught in a broad semiconductor downdraft. Investors are growing increasingly skeptical about whether the enormous sums being poured into AI will translate into earnings growth on a timely basis. The Philadelphia Semiconductor Index dropped 2.6% as the technology sector emerged as the week's weakest market segment, with Nvidia, AMD, Broadcom and Micron all coming under selling pressure.

Market reports pointed to rising oil prices and climbing US Treasury yields as additional drags on risk appetite, with investors awaiting key economic and quarterly data.

Those doubts found a voice in the industry's own executive suites. At the DBS Asia Future Forward Conference, Intel's Lip-Bu Tan and Broadcom CEO Hock Tan addressed the strains of the current investment cycle, noting that uncertain returns on AI projects and elevated financing costs could temper the pace of AI infrastructure buildout.

Wall Street Adjusts Its Targets

Even so, the cautious tone hasn't stopped sell-side analysts from revisiting their models. Mizuho raised its price target on Intel from $92 to $114 on October 6, according to media reports, while keeping a Neutral rating on the shares.

Intel's stock has climbed 184% since the start of the year, a gain that underscores how much optimism is already baked into the valuation. That makes the upcoming earnings release a critical test. The company will report third-quarter 2026 financial results on October 29 after the US market close — an occasion investors are expected to use to measure whether the recent rally can be justified by actual operating performance.