Intel has started shipping early D-Series GPUs to select customers for testing, the company confirmed this week. The chips are designed specifically for AI inference—the process of running trained machine learning models—and mark Intel's most direct push yet into a data center market currently dominated by NVIDIA and AMD.
The move comes as Intel tries to turn around its foundry and chip design businesses. While the company hasn't disclosed pricing or a full launch date, the sampling phase suggests a commercial release could come within months.
What D-Series actually does
Inference is the boring but essential half of AI. Training gets the headlines—it's how models learn from massive datasets. Inference is what happens when you actually use those models: every ChatGPT response, every photo tag, every fraud alert. It's cheaper to run than training but happens far more often, which adds up at scale.
Intel is betting that inference workloads will be the bigger long-term market. The D-Series is built to handle those jobs with better power efficiency than general-purpose GPUs, according to the company's technical briefings. That matters because data centers are increasingly constrained by electricity, not floor space.
Why NVIDIA and AMD should pay attention
NVIDIA owns the lion's share of AI accelerators today, largely because its CUDA software platform has become the default for developers. AMD has made inroads with its MI series, particularly among cloud providers looking for a second source. Intel's challenge isn't just silicon—it's convincing customers to rewrite code for a new architecture.
The company has been quietly building out its software stack, OneAPI, to ease that transition. But the real test will be whether D-Series can deliver competitive performance per watt and per dollar. Early sampling customers—whose names Intel hasn't disclosed—will be the first to find out.
The data center chessboard
Intel's timing is strategic. AI infrastructure spending is still climbing, but buyers are getting pickier. Cloud providers want options beyond NVIDIA, partly to negotiate better prices and partly to avoid supply chain bottlenecks. If D-Series works as advertised, it could give them a credible third choice.
That said, Intel has a history of promising chips that slip. Its original GPU line, Ponte Vecchio, faced delays and mixed reviews. The company can't afford another stumble in a market this competitive.
What happens next
Intel hasn't said when D-Series will be generally available or how much it will cost. The sampling phase typically lasts six to twelve months before a full launch, meaning a commercial debut could land in late 2025 or early 2026. Until then, the only concrete signal will come from those early customers—if any of them go public with benchmarks. Intel's next earnings call, expected in January, may offer the first hints about whether D-Series is on track or running into the same problems that have plagued the company's other hardware bets.
