Recommended Free Tools
Pat Gelsinger’s “10,000× too expensive” remark is an estimate about the economics of running AI inference at scale—not evidence that a particular NVIDIA GPU’s price is literally 10,000 times too high. He later said the figure came from his own calculations about search’s energy, compute and cost, without providing the underlying math. His “NVIDIA got lucky” line, meanwhile, describes how the company’s long-running bet on throughput computing later matched AI workloads; it does not establish that luck alone explains NVIDIA’s success.
What Gelsinger meant by “10,000× too expensive”
In a March 2025 report, HotHardware quoted former Intel CEO Pat Gelsinger saying a GPU was “way too expensive” for the goal of deploying AI inference at scale. He put the required improvement at “10,000x” while discussing what would be needed to realize that deployment. HotHardware’s account attributes the remark to Gelsinger’s appearance on the Acquired podcast as an invited guest at NVIDIA’s GTC 2025 conference.
As an Amazon Associate I earn from qualifying purchases.
The wording can sound like a claim about the price tag on a GPU. The available context does not support that interpretation: Gelsinger was talking about the economics of inference, not saying that a named GPU costs 10,000 times more than a directly comparable alternative.
Free tools Windows power users keep installed
One-click scans. No signup required.
How he later explained the estimate
In a 2026 interview with Ian Cutress, Gelsinger described “10,000x” as “sort of a number that I pulled out based on some math of where search was in terms of energy, compute, cost.” He did not supply the calculation in the interview excerpt. The figure is therefore his estimate of the scale of improvement he believes may be needed, not an independently measured GPU-versus-inference-chip ratio. Read the 2026 interview.
#1 Best Overall
- ATTENTION BUYERS: EXPORTING this item outside the U.S. requires strict compliance with the Bureau of Industry and Security (BIS) U.S. Export Administration Regulations (EAR). Non-compliance carries fines up to $1 MILLION PER VIOLATION AND UP TO 20 YEARS IMPRISONMENT for both the buyer and seller. To uphold our regulatory obligations, we strictly follow BIS "Know Your Customer" guidance by screening each order and immediately aborting transactions showing unresolved regulatory red flags. ***
- [Double-Flow-Through Design] The RTX PRO 6000 Blackwell features a double-flow-through cooling design, optimizing efficiency and airflow to sustain peak performance under 600W power loads. | [5th Gen Tensor Cores] Deliver up to 3X the performance of the previous generation and support for FP4 precision for faster AI model processing times with reduced memory usage, enabling local fine-tuning of LLMs and generative AI | [4th Gen Ray Tracing Cores] Double the ray-triangle intersection rate of the previous generation to create photoreal, physically accurate scenes and immersive 3D designs with RTX Mega Geometry, which enables up to 100X more ray-traced triangles.
- [PCIe Gen 5] Support for PCIe Gen 5 provides double the bandwidth of PCIe Gen 4, improving data-transfer speeds from CPU memory and unlocking faster performance for data-intensive tasks like AI, data science, and 3D modeling. | [GDDR7 Memory] With 96 GB of GPU memory and 1.8 TB ps bandwidth, it can tackle massive 3D and AI projects, fine-tune AI models locally, explore large-scale VR environments, and drive larger multi-app workflows.
- [DisplayPort 2.1] Achieve unparalleled visual clarity and performance, driving high resolution displays at up to 8K at 240 Hz and 16K at 60 Hz. Increased bandwidth enables seamless multi-monitor setups while HDR and higher color depth support ensures superior color accuracy for precision work, such as video editing, 3D design, and live broadcasting.
- [NVIDIA Blackwell Streaming Multiprocessor] The new SM features increased processing throughput, and new neural shaders that integrate neural networks inside of programmable shaders | DLSS 4: Multi Frame Generation ensures ultra-smooth frame pacing for lifelike simulations. | [Universal MIG] Divide a single RTX PRO 6000 Blackwell into multiple isolated instances, each with dedicated resources, allowing for concurrent execution of multiple workloads, optimized GPU utilization, and secure isolation of different applications or users. [WARRANTY] 3 YR Manufacturer's Warranty. Bulk OEM Packaging. Retail Packaging is NOT included.
Why he distinguishes training from inference
Training is the process of fitting a model; inference is the work of using a trained model to generate outputs, such as answers to user prompts. Gelsinger said GPUs are well suited to training and can help bridge training into inference, but are “not an optimized inference chip.” That is an argument about specialization: a processor designed around the demands of inference might improve the cost or efficiency of serving AI workloads.
The 2025 report did not identify a specific replacement architecture endorsed by Gelsinger. Its references to NPUs and ASICs were the reporter’s possible interpretations, not a confirmed recommendation from him. Nor do the cited accounts provide product-level figures to show that any one alternative delivers a particular cost or efficiency gain.
Rank #2
- NVIDIA Volta GV100 Architecture — 4,608 CUDA Cores, 640 1st-Gen Tensor Cores delivering 14 TFLOPS FP32 and 112 TFLOPS deep learning performance for AI training, inference, HPC, and scientific computing workloads
- 32GB HBM2 ECC Memory — 900 GB/s Bandwidth — High-bandwidth memory on a 4096-bit bus with ECC error correction provides the memory capacity and throughput required for the largest AI models, simulations, and datasets
- PCIe 3.0 x16 Interface — 250W TDP — Standard PCIe Gen3 connectivity with passive cooling designed for enterprise rack server deployment in HPE ProLiant, Dell PowerEdge, and Supermicro platforms with adequate chassis airflow
- NVLink — Scale to 96GB Unified Memory — Connect two V100 GPUs via NVLink at 300 GB/s bi-directional bandwidth to scale GPU memory from 32GB to 96GB for larger AI training and HPC workloads
- Multi-Precision Computing — Supports FP64 (7 TFLOPS), FP32 (14 TFLOPS), FP16 (112 TFLOPS) and INT8 precision modes for flexible deployment across training, inference, and scientific simulation workloads
What “NVIDIA got lucky” does—and does not—say
HotHardware’s headline draws on Gelsinger’s discussion of his earlier conversations with NVIDIA CEO Jensen Huang about throughput computing. Gelsinger’s point, as reported, was that NVIDIA pursued a GPU strategy whose capabilities later aligned with AI’s computing demands. The phrase “got lucky” is a pointed shorthand for that favorable fit between a long-term strategic direction and a major workload shift; the report does not establish that chance was the sole cause of NVIDIA’s position.
The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →HotHardware is secondary reporting, and the original podcast recording or transcript was not independently verified for this account. The quotation should be understood as HotHardware’s attribution of Gelsinger’s words, rather than as a separately checked transcript.
Rank #3
- Professional GPU with Blackwell Architecture
- Blackwell Architecture
- 24GB GDDR7 with PCIe 5.0 & Ray Tracing
- AI Workstation
How to evaluate claims about inference hardware
A claim that one accelerator is more economical than another is meaningful only when the workload and comparison baseline are clear. Gelsinger named several performance measures in the 2026 interview; a practical comparison should also account for energy and cost under the same operating conditions.
- Throughput: total work completed over a stated period; for text generation, report tokens per second and clarify whether the figure is per user, device or system.
- Latency: how long a response takes, including the relevant measurement point, such as time to first token or completion time.
- Performance per watt: output relative to power consumed, measured on the same workload and with a clearly stated measurement boundary.
- Total cost: the costs included in the comparison and the period or amount of work over which they are counted.
- Deployment constraints: model support, software compatibility and other practical requirements that affect whether a chip can serve the workload.
Gelsinger specifically named tokens per second, tokens per second per watt, aggregate throughput and latency as measures for judging architecture performance. The cited sources do not supply results comparing particular GPUs with particular inference chips on these measures, so they cannot establish a hardware winner or validate the 10,000× figure as a benchmark.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
