Intel
Last reviewed
Jun 9, 2026
Sources
35 citations
Review status
Source-backed
Revision
v1 ยท 2,726 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Jun 9, 2026
Sources
35 citations
Review status
Source-backed
Revision
v1 ยท 2,726 words
Add missing citations, update stale details, or suggest a clearer explanation.
Intel Corporation is an American multinational semiconductor company headquartered in Santa Clara, California. Founded on July 18, 1968 by Robert Noyce and Gordon Moore, it commercialized the first microprocessor in 1971 and dominated the personal computer and server eras with its x86 processors. Intel entered the generative AI boom in a weakened state, with no competitive answer to NVIDIA's data center GPUs and a manufacturing arm that had fallen behind TSMC. The result was one of the most dramatic corporate turnaround attempts in technology. In 2025 the United States government converted CHIPS Act awards into a 9.9 percent equity stake [1], NVIDIA invested $5 billion as part of a product alliance [2], SoftBank added $2 billion [3], and the board handed the CEO job to semiconductor veteran Lip-Bu Tan [4]. By mid-2026, with the company's 18A process shipping in volume and a reported multi-million-unit tensor processing unit foundry order from Google, Intel's shares had staged a sharp recovery [5].
Intel designs and manufactures central processing units for PCs (Core and Core Ultra), servers (Xeon), discrete Arc GPUs, and Gaudi AI accelerators, and it operates Intel Foundry, which fabricates chips for Intel itself and for external customers. Its main reporting segments are Client Computing Group, Data Center and AI (DCAI), and Intel Foundry; it also retains majority ownership of Mobileye, the driver-assistance and autonomous-driving company.
Full-year 2025 revenue was $52.9 billion, roughly flat year over year, with fourth-quarter revenue of $13.7 billion exceeding guidance despite industry-wide supply constraints [6]. Under a restructuring launched in 2025, Intel cut its workforce from about 109,000 employees at the end of 2024 toward a target of roughly 75,000 core employees by the end of 2025 [7]. Intel remains the leading supplier of PC and server CPUs, but in AI accelerators it is a marginal player behind NVIDIA and AMD; its current AI strategy centers on inference, AI PCs, and manufacturing other companies' AI silicon.
Intel began as a memory-chip maker and produced the first commercial microprocessor, the 4004, in 1971. The 8086 processor of 1978 established the x86 instruction set that still underpins most PCs and servers. Under Andy Grove the company exited the memory business in the mid-1980s to focus on processors, and co-founder Gordon Moore gave his name to Moore's law, the observation that transistor counts double roughly every two years. Through the "Wintel" era and the rise of cloud computing, Intel's Xeon line came to power the large majority of the world's servers.
The company famously missed the smartphone transition, and its early accelerator efforts (the Larrabee graphics project and the Xeon Phi many-core line) were both abandoned. It then tried to buy its way into AI: Nervana Systems in 2016, computer-vision chipmaker Movidius in 2016, Mobileye for $15.3 billion in 2017 (partially floated in 2022), and Israeli accelerator startup Habana Labs for about $2 billion in December 2019, after which the Nervana processor line was shut down [8].
Pat Gelsinger returned as CEO in early 2021 with the IDM 2.0 strategy: regain process leadership and open Intel's fabs to outside customers. The plan proved enormously expensive just as the AI boom shifted data center spending from CPUs to GPUs. In August 2024 Intel announced more than 15,000 job cuts and suspended its dividend; in December 2024 the board forced Gelsinger out [9]. For 2024 the company reported a net loss of $18.8 billion on revenue of $53.1 billion, its first annual loss since 1986 [10].
Intel's flagship AI chip line is Gaudi, inherited from Habana Labs. Gaudi processors power Amazon EC2 DL1 instances (Gaudi 1) and were followed by Gaudi 2 in 2022. Gaudi 3, unveiled at Intel Vision on April 9, 2024 and built on TSMC's 5 nm process, delivers 1,835 teraFLOPS of FP8 compute and was marketed primarily on price-performance against NVIDIA's H100 [11]. Commercial traction was weak: on the October 2024 earnings call Gelsinger admitted Gaudi would miss even its modest $500 million revenue target for 2024, citing the Gaudi 2 to Gaudi 3 transition and software ease-of-use problems, and Intel took a $300 million accelerator inventory write-down [12]. IBM Cloud became the first cloud service provider to deploy Gaudi 3, with availability rolling out in 2025 [13].
The roadmap has been repeatedly reset. In January 2025 Intel canceled the commercial release of Falcon Shores, the GPU meant to succeed Gaudi, relegating it to an internal test chip and pointing instead to Jaguar Shores, a rack-scale AI system with no announced date [14]. In October 2025, at the OCP Global Summit, Intel unveiled Crescent Island, a data center GPU aimed squarely at inference: it uses the Xe3P architecture with 160 GB of LPDDR5X memory, sidestepping the shortage-prone high-bandwidth memory used by rivals, and targets air-cooled enterprise servers and "tokens-as-a-service" providers [15]. At Computex in June 2026 Intel said partners could configure up to 480 GB of memory, with customer sampling planned for the second half of 2026 [16]. On software, Intel maintains the OpenVINO inference toolkit and backs oneAPI as an open alternative to NVIDIA's CUDA ecosystem.
| Product | Role | Status (June 2026) |
|---|---|---|
| Gaudi 3 | Training and inference accelerator | Shipping; deployed on IBM Cloud [13] |
| Falcon Shores | Data center GPU | Canceled as a product in January 2025; internal test chip [14] |
| Crescent Island | Inference GPU (Xe3P, 160 to 480 GB LPDDR5X) | Customer sampling planned for second half of 2026 [16] |
| Jaguar Shores | Rack-scale AI system | In development; no announced launch date [14] |
Intel coined much of the "AI PC" marketing push. The Core Ultra "Meteor Lake" processors launched on December 14, 2023 were Intel's first client chips with an integrated neural processing unit (11 TOPS) and its first chiplet-based client design; Intel set a goal of shipping AI-capable PCs to more than 100 million people through 2025 [17]. Lunar Lake (Core Ultra 200V, September 2024) raised the NPU to 48 TOPS, clearing the 40-TOPS bar for Microsoft's Copilot+ PC program [18].
At CES on January 5, 2026, Intel launched Core Ultra Series 3 ("Panther Lake"), the first volume product built on its 18A node, combining a 50-TOPS NPU with Xe3 integrated graphics rated at 120 TOPS for up to 180 platform TOPS; systems reached stores on January 27, 2026, with more than 200 designs planned [19].
Even where NVIDIA GPUs win, Intel increasingly supplies the host processor. NVIDIA selected the 64-core Xeon 6776P as the CPU for its DGX B300 Blackwell Ultra systems in May 2025 [20], and Xeon 6 parts were later chosen as host CPUs for DGX Rubin NVL8 systems [21]. In June 2026, at Computex, Intel launched Xeon 6+ "Clearwater Forest," its first 18A server processor, scaling to 288 efficiency cores with 576 MB of L3 cache; Intel claims about 30 percent better per-thread performance than AMD's 192-core EPYC 9965 [22].
Gelsinger's IDM 2.0 bet, continued in modified form under Tan, is that Intel can become a Western alternative to TSMC. The "five nodes in four years" roadmap culminated in Intel 18A, a 1.8 nanometer-class node introducing RibbonFET gate-all-around transistors and PowerVia backside power delivery. In October 2025 Intel declared Fab 52 in Chandler, Arizona fully operational, beginning high-volume 18A production; it is the most advanced process technology developed and manufactured in the United States [23].
Economically, the foundry remains a drag: in 2024 Intel Foundry posted a $13.4 billion operating loss on $17.5 billion of mostly internal revenue [10]. External commitments exist (Microsoft announced a chip design on 18A in 2024, and Amazon Web Services agreed to build an AI fabric chip with Intel), but committed third-party volumes are small [24]. For the next node, 14A, Intel said in early 2026 that it had two prospective anchor customers, with firm decisions expected between the second half of 2026 and the first half of 2027; management has warned that leading-edge investment depends on winning such commitments [25]. Intel targets foundry break-even around 2027 [26]. The US government's 2025 equity deal reinforces the strategy: Washington received a five-year warrant for an additional 5 percent of Intel at $20 per share, exercisable only if Intel's ownership of the foundry falls below 51 percent, effectively discouraging a spin-off [27].
Momentum improved markedly in 2026. On June 8 and 9, 2026, multiple outlets reported that Google had placed an order for more than 3 million next-generation TPUs to be manufactured by Intel in 2028, and that NVIDIA was evaluating 18A and Intel's advanced packaging for a future multi-die GPU; Intel shares jumped about 9 to 11 percent on the reports, which neither Intel nor Google has formally confirmed [5].
After Gelsinger's forced retirement on December 1, 2024, CFO David Zinsner and products chief Michelle Johnston Holthaus served as interim co-CEOs [9]. On March 12, 2025 Intel named Lip-Bu Tan chief executive, effective March 18, 2025. Tan ran Cadence Design Systems from 2009 to 2021, founded the venture firm Walden International in 1987, and had quit Intel's board in August 2024 [4].
Tan moved quickly: canceling planned fabs in Germany and Poland, slowing the Ohio buildout, consolidating assembly and test operations, and cutting management layers by about half, with $1.9 billion in restructuring charges in the second quarter of 2025 alone [7][28]. Intel also sold 51 percent of its Altera FPGA unit to Silver Lake for $4.46 billion, valuing Altera at $8.75 billion; the deal closed on September 15, 2025 [29]. Tan has described the strategy as financial discipline plus a renewed AI roadmap focused on inference and agentic workloads.
Intel's 2025 capital story was extraordinary. In early August 2025 President Donald Trump publicly demanded Tan's resignation over his past investments in Chinese chip firms, then reversed course after a White House meeting; two weeks later the equity deal was announced [30].
| Investor | Announced | Amount | Price per share | Approximate stake |
|---|---|---|---|---|
| SoftBank Group | August 18, 2025 | $2.0 billion | $23.00 | about 2 percent [3] |
| US government | August 22, 2025 | $8.9 billion | $20.47 | 9.9 percent [1] |
| NVIDIA | September 18, 2025 (closed December 29, 2025) | $5.0 billion | $23.28 | about 4 percent [2][32] |
The government's $8.9 billion purchase of 433.3 million shares was funded by $5.7 billion in awarded but unpaid CHIPS Act grants and $3.2 billion from the Secure Enclave defense chip program, on top of $2.2 billion in grants already received; Intel says the stake is passive, with no board seat [1][31]. The NVIDIA agreement goes beyond money: Intel will build custom x86 CPUs for NVIDIA's AI infrastructure platforms and x86 system-on-chips integrating NVIDIA RTX GPU chiplets for PCs, with the architectures linked by NVLink [2]. The FTC cleared the investment on December 18, 2025, and it closed on December 29, 2025 [32]. Critics across the political spectrum questioned the precedent of direct federal ownership in a publicly traded chipmaker [30].
Intel reported full-year 2025 revenue of $52.9 billion on January 22, 2026, its fifth consecutive quarter of beating its own forecasts [6]. First-quarter 2026 results, announced April 23, 2026, marked the clearest inflection: revenue of $13.6 billion beat guidance by about $1.4 billion, DCAI revenue rose 22 percent year over year to $5.1 billion with a $1.5 billion operating profit, and non-GAAP EPS more than doubled to $0.29 [33]. The stock surged past its dot-com-era record high from 2000 [34], and by late April 2026 the government's stake was worth about $36 billion, a roughly 300 percent paper gain [35].
In June 2026 Intel launched Clearwater Forest at Computex [22] and benefited from the reported Google TPU order and NVIDIA's evaluation of 18A [5]. The competitive gap nonetheless remains wide: NVIDIA still dominates AI training silicon, AMD holds the clear number-two position in data center GPUs, and Intel's near-term accelerator hopes rest on Crescent Island sampling in late 2026 [16] and on Jaguar Shores, which still has no launch date [14]. Intel's AI relevance in 2026 rests less on beating NVIDIA's GPUs than on surrounding them: host CPUs in AI servers, NPUs in hundreds of millions of PCs, and a US-based foundry that rivals and hyperscalers are beginning to take seriously.