Intel CEO Lip-Bu Tan was reportedly told employees in July 2025 that the company was “too late” to catch up in large-scale AI-model training and was no longer among the “top 10 semiconductor companies.” Those remarks came from a reported internal conversation, not a publicly released Intel transcript. The “too late” comment was about AI training—not AI as a whole—and Intel’s subsequent strategy continued to target inference, agentic AI, edge computing, CPUs, AI PCs and foundry manufacturing.
What Lip-Bu Tan reportedly said
Reports published in July 2025 attributed two blunt assessments to Lip-Bu Tan, who became Intel’s chief executive on March 18, 2025. According to AI Business, Tan said, “On training, I think it is too late for us,” and reportedly told employees that Intel was “not in the top 10 semiconductor companies.”
As an Amazon Associate I earn from qualifying purchases.
Intel has not published a complete transcript or recording of that exchange. The question Tan was answering, the full wording and the surrounding discussion therefore remain unverified outside the reported accounts. The comments should be described as attributed remarks from an internal conversation, not as a formal earnings-call quotation.
“Too late” meant AI training, not every AI market
Large-scale AI training is the most accelerator-intensive part of the market. Training frontier models requires huge clusters, high-bandwidth memory, fast networking, reliable systems integration and software that developers already know how to use. Nvidia’s advantage extends beyond chip performance: its accelerator products, CUDA software ecosystem, hyperscaler relationships, networking portfolio, deployment scale and financial resources reinforce one another.
#1 Best Overall
- 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
That makes catching up a platform and ecosystem problem, not simply a matter of designing a faster processor. Intel’s own 2025 materials acknowledged that its AI strategy needed refinement. Tan’s reported statement is best read as a judgment that Intel could not close Nvidia’s lead in major training workloads quickly enough to justify making that the company’s primary near-term bet.
It does not establish that Intel considered AI impossible or that it was abandoning the field. Training, inference and edge workloads have different technical and commercial requirements.
Training and inference are different opportunities
| Area | Typical demand | Intel’s stated position |
|---|---|---|
| Model training | Large, concentrated accelerator clusters in hyperscale data centers | Tan reportedly viewed the gap as too large to close quickly |
| Inference | Running trained models in cloud, enterprise, PCs, vehicles and edge systems | Intel said it would emphasize this area |
| AI PCs and edge AI | Distributed workloads using CPUs, integrated accelerators and local software | Continued product and platform investment |
In a July 2025 employee message, Intel said it would concentrate on areas where it believed it could differentiate, including AI inference, agentic AI, AI at the edge, its x86 CPU franchise and foundry execution. The company’s message is available at Intel’s newsroom.
What did “not in the top 10” mean?
The reported “top 10” statement has no stated ranking method. It could refer to semiconductor revenue, market value, profitability, AI-chip sales, manufacturing capacity, technology influence or an internal competitive assessment. Those lists produce different results and change over time.
Rank #2
- Delivers 25 TFLOPS of AI inference performance using four MemryX MX3 accelerators in a compact M.2 2280 M-key module.
- Connects through PCI Express Gen 3 and above for straightforward integration into embedded and edge systems with an M.2 M-key slot.
- Includes 42 MB of distributed on-chip weight storage to support low-latency, batch-1 pipelined inference.
- Designed for low-power deployment with a 1 W–12 W operating range and host-managed airflow requirements.
- Supports ONNX, TensorFlow, Keras, and TFLite models through the MemryX software stack on Linux and Windows hosts.
Consequently, it is not accurate to present the remark as proof that Intel was officially ranked outside a universal top 10. A market-capitalization comparison would also require a date and a definition of “semiconductor company”; equipment makers such as Applied Materials, Lam Research and KLA may be included in some lists.
The safer interpretation is that Tan was describing Intel’s perceived competitive relevance to employees. Intel could still be a major CPU supplier and semiconductor manufacturer while its chief executive believed the company had lost ground against faster-growing rivals.
Intel’s alternative AI strategy
Intel’s portfolio remained broader than a single training accelerator. It included Xeon server CPUs, Gaudi accelerators, Arc GPUs, AI-PC processors and software tools. The company’s official plan was to seek defensible positions in markets where Intel could use its installed CPU base, local processing capabilities and manufacturing ambitions.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Inference and agentic AI
Inference runs models after training and can occur in cloud servers, corporate systems, PCs, factories, vehicles and other devices. It is more distributed than frontier-model training, although it remains intensely competitive. Nvidia, AMD, custom cloud silicon, Arm-based processors and specialized startups all compete for these workloads.
Rank #3
- Professional AI & Creator Workstation: AMD Radeon AI PRO R9700 GPU with 32GB GDDR6 is engineered for AI development, professional content creation, and compute-intensive workloads.
- Massive 32GB Memory Capacity: 32GB of GDDR6 memory on a 256-bit bus provides ample bandwidth for large AI models, 8K video editing, and complex 3D rendering.
- Advanced RDNA 4 with AI Accelerators: 64 Compute Units with 3rd Gen Ray Tracing and dedicated 2nd Gen AI Accelerators for groundbreaking AI performance and visual computing.
- Professional Blower Cooling: Efficient single blower design exhausts heat directly out of the chassis, ideal for multi-GPU workstation and server configurations.
- Enterprise-Grade Thermal Solution: Vapor chamber heatsink with industrial Honeywell PTM7950 thermal interface material ensures reliable cooling under sustained professional loads.
Agentic AI—systems that plan and perform multistep tasks—was another focus identified by Intel. Naming the opportunity did not guarantee product leadership; success still depends on performance, software compatibility, power efficiency, customer adoption and timely execution.
Manufacturing and the x86 franchise
Intel treated AI as one of four strategic priorities in its second-quarter 2025 materials, alongside organization and culture, foundry and the x86 business. That combination matters because Intel was trying to repair both product competitiveness and manufacturing economics. Products determine whether customers buy; manufacturing determines whether Intel can deliver them competitively and operate a viable foundry.
How large were the workforce cuts?
Intel’s own filings provide a clearer measure than headlines claiming a precise worldwide layoff total. In its July 24, 2025 earnings materials, Intel said it had completed most planned headcount actions and expected to finish 2025 with approximately 75,000 core employees through layoffs and attrition. The target represented an approximately 15% reduction in the core workforce.
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 →The later 2025 Form 10-K said the core workforce had been reduced by approximately 15% by the end of fiscal 2025. It recorded about $2.2 billion in restructuring charges for the year, with some actions continuing into 2026. The 15% figure applies to Intel’s core workforce, not necessarily every contractor, subsidiary or role worldwide; the 75,000 figure likewise excluded subsidiaries in contemporary reporting.
Intel’s second-quarter Form 10-Q reported $1.9 billion in restructuring charges for the quarter, including $1.5 billion in cash-based employee severance and exit costs. These figures document a major restructuring but do not provide a single exact count of people laid off globally, because the plan combined layoffs with attrition.
Restructuring went beyond job cuts
Intel paired the workforce reduction with changes to factories, projects and spending:
- Planned projects in Germany and Poland were stopped.
- Costa Rica assembly and test operations were to be consolidated into larger sites in Vietnam and Malaysia.
- Construction in Ohio was slowed so spending would better match demand.
- Resources were redirected toward core client and server businesses, while lower-priority programs were reduced.
- Intel targeted $17 billion in non-GAAP operating expenses and approximately $18 billion in gross capital expenditures for 2025.
These actions addressed a broader financial and organizational problem. It would be misleading to label every affected role an AI layoff: the restructuring also covered management layers, manufacturing capacity, capital projects and product priorities.
Why the pessimism was understandable
Intel was attempting several expensive turnarounds at once: regain leadership in client and server products, build an external foundry business, develop AI-related silicon and fund new manufacturing technology. The pressure was visible in its financial disclosures.
Best Value
- Memory Size: 16 GB GDDR6 ECC.
- Memory Bus Width: 128-bit.
- Memory Bandwidth: 200 GB/s.
- CUDA Cores: 1280.
- Peak Single Precision floating point performance: 18 Tflops (GPU Boost Clocks).
- Intel Products revenue was $49.1 billion in 2025, down $324 million from 2024.
- The company recognized approximately $2.2 billion in restructuring charges in 2025.
- Its data-center and AI results still included charges related to Gaudi accelerator inventory, although those charges were lower than in 2024.
- In the contemporaneous second quarter of 2025, revenue was $12.9 billion, flat year over year, and Intel reported a $2.9 billion quarterly loss, according to the Associated Press.
Those numbers show the scale of the turnaround; they do not by themselves prove that Intel was doomed. Cost reductions can improve focus, but they also risk losing engineering talent and slowing execution if priorities are not managed carefully.
What happened after the July 2025 reports?
Intel’s fiscal 2025 filing confirms that the workforce plan materially progressed: the core workforce was down approximately 15% by year-end, and restructuring charges reached about $2.2 billion. That later filing is stronger evidence of the outcome than early reports estimating “thousands” of cuts.
Intel also continued investing in AI-enabled products. Its 2026 corporate filing describes Core Ultra Series 3 AI-PC products built on Intel 18A and unveiled in January 2026. That development is inconsistent with a claim that Intel exited AI altogether, although it does not demonstrate that the company had caught Nvidia in training or that the turnaround had succeeded.
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minutePC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11As of the available 2026 reporting, the evidence supports a distinction between strategy and results: Intel changed where it intended to compete, but the cited filings do not establish a definitive recovery in AI market share, training leadership or overall semiconductor ranking.
Bottom line
Tan’s reported “too late” comment was a narrow but consequential admission about Intel’s position in large-scale AI training. It was not a declaration that Intel had no future in AI. The company’s response was to pursue inference, agentic and edge AI, AI PCs, Xeon and Arc products, x86 strength and foundry execution while cutting costs and restructuring operations.
The “top 10” claim should remain attributed to Tan because no ranking methodology was supplied. The layoffs were real and substantial, but Intel documented them as an approximately 15% reduction in its core workforce through layoffs and attrition, not as one precisely verified worldwide layoff total.
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.
Recommended Free Tools
