Intel’s Gaudi 3 blows past datacenter bottlenecks in AI inferencing revolution

Intel’s stock is trading at $22.92, down 1.6% today, with a market cap just under $100 billion. Volume is surging past 66 million shares, well above its 90 million average. Meanwhile, NVIDIA is sitting at $170.70, up 4%, with a $4.16 trillion valuation and 230 million shares traded. The divergence is grotesque.

Intel’s Gaudi 3 baseboard, priced at $125,000 for eight accelerators, is delivering 2.9x better price-to-performance than NVIDIA’s $200,000 H100 HGX baseboard2. That’s not marketing fluff. That’s benchmarked throughput per dollar. And it’s not just training. Gaudi 3 is showing up to 2x faster inference on popular LLMs compared to H100. The kicker? Gaudi 3 uses Ethernet for interconnect, not proprietary NVLink, which slashes networking costs and complexity4.

NVIDIA’s moat is CUDA, but CUDA is just a C++ wrapper. PyTorch, TensorFlow, JAX—they all support multiple backends. The software stack is abstracted. Developers don’t care what silicon they’re running on. The illusion of lock-in is breaking.

The market is shifting from training to inference. Training is Ferrari compute. Inference is Civic efficiency. And inference is where the real money is. The AI inference market is projected to hit $106 billion in 2025 and $255 billion by 20305. Intel is positioned to dominate that space. NVIDIA is optimized for training. That’s yesterday’s war.

DeepSeek’s R1 model trained on 2,048 H800s and matched performance against competitors using 10,000+ H100s. NVIDIA lost $589 billion in market cap in one day8. That wasn’t panic. That was clarity. The market saw what efficient compute looks like.

Intel’s Gaudi 3 is trending across engineering forums and enterprise IT circles. It’s not meme-stock material yet, but it’s gaining traction. If it breaks $24 with volume confirmation, the next tradable level is $27. Above that, $30 is in play. Below $22.50, it’s dead money short term.

NVIDIA’s chart is vertical. But inference workloads don’t need vertical. They need horizontal scale at low cost. That’s Intel’s game. And the market is waking up.

Intel’s valuation is a joke. The setup is asymmetric. The risk is priced in. The upside isn’t.

NOTE: This is not financial advice. Please conduct your own due diligence.

Sources:

https://www.tomshardware.com/pc-components/cpus/intels-gaudi-3-will-cost-half-the-price-of-nvidias-h100

https://www.nextplatform.com/2024/06/13/stacking-up-intel-gaudi-against-nvidia-gpus-for-ai

https://www.techradar.com/pro/intel-discloses-list-prices-of-its-gaudi-3-and-gaudi-2-ai-accelerators-and-were-in-for-a-shock-rivals-to-iconic-nvidias-h100-gpu-have-a-much-better-performance-per-dollar-ratio-but-will-it-matter

https://finance.yahoo.com/news/ai-inference-market-forecast-report-082100788.html

https://www.bernstein.com/our-insights/insights/2025/articles/seeking-clarity-on-deepseek-ai.html

https://www.guinnessgi.com/insights/how-has-deepseek-affected-ai-market-investors

https://www.crn.com/news/components-peripherals/2024/intel-reveals-8-chip-gaudi-3-platform-price-upending-industry-norm-of-secrecy

https://www.monexa.ai/blog/nvidia-s-fy2025-surge-ai-moat-strategic-growth-fue-NVDA-2025-06-27

https://www.forbes.com/sites/karlfreund/2025/03/12/big-ai-inference-has-become-a-big-deal-and-a-bigger-business