Intel

The persistent third-place AI accelerator vendor — Gaudi 3 ships, undersells, and the post-Falcon-Shores pivot to Jaguar Shores is Intel's last credible swing at training silicon.

1. Core Product / Service

Intel's AI accelerator portfolio in 2026:

Gaudi 3 (Habana lineage) — current shipping product, 128 GB HBM2e, ~1.8 PFLOPS FP8, claimed 70% better $/perf than H100 on Llama 3 inference per Intel benchmarks. Available via Dell/HPE/Supermicro systems and IBM Cloud as a service.
Falcon Shores — cancelled as a commercial product. Originally pitched as the CPU+GPU fusion follow-on to Ponte Vecchio + Gaudi; reclassified February 2025 as an internal development platform [3][4].
Jaguar Shores — next-gen data-center AI accelerator, targeted for 2026 launch, designed for LLM training and large-scale inference [3].
Xeon 6 CPUs (Granite Rapids, Sierra Forest) ship with AMX matrix extensions for inference at the edge / smaller models — the only Intel AI silicon actually winning today, though it doesn't compete in the discrete-accelerator market.

Software stack: OneAPI + SynapseAI (Habana) + OpenVINO. Less mature than CUDA, behind ROCm.

2. Target Users & Pain Points

Enterprises priced out of NVIDIA — IBM Cloud Gaudi 3 service, Inflection AI (which switched to Gaudi 3 in 2025), federal/defense customers wanting US-fab silicon.
Cloud customers wanting a third option alongside NVIDIA and AMD; Gaudi 3 is positioned as the budget tier.

The honest pitch: Gaudi 3 is competitive on $/inference for select workloads, but volume is small enough that cloud allocation isn't a meaningful pain solve.

3. Competitive Landscape

Vendor	Discrete-AI revenue 2025	Position
nvidia	~$130B+ DC	Untouchable incumbent
amd	$5B+ AI accel	Credible #2, growing
Intel (Gaudi 3 + Falcon/Jaguar)	<$500M	Missed the $500M Gaudi target [5]
huawei-ascend	$5–10B (China only)	Captive China market
Hyperscaler captive (google-tpu aws-trainium microsoft-maia)	Bigger than Intel combined	Internal use

4. Unique Observations

Intel has missed three consecutive AI cycles. Knights Landing/Mill (Phi) cancelled. Nervana acquired then shuttered. Habana Gaudi/Gaudi 2/Gaudi 3 ship but each generation undersells the prior target. Pat Gelsinger's $500M Gaudi sales target for 2024 was missed [5]; Gaudi 3 in 2025 disappointed [5].
Gaudi line ends with Gaudi 3 — Intel announced the Gaudi family is being wound down in favor of a unified Xe-derived data-center GPU line under "Jaguar Shores" [3]. This is the third architectural reset in eight years; each reset costs Intel another generation against nvidia and amd.
Falcon Shores cancellation is the bigger signal. Falcon was supposed to merge Habana's Gaudi inference DNA with Xe GPU compute. Reclassifying it to internal-only means Intel concedes that none of its current architectures is competitive enough to ship — and Jaguar Shores is effectively a clean-sheet on a longer timeline [3][4].
Foundry strategy is the wildcard. Intel Foundry's 18A node (in production for Panther Lake CPUs in 2025) and 14A node (2027) are the company's actual strategic bet — if Intel can win nvidia or amd foundry business away from tsmc, the AI-accelerator business becomes a sideshow. So far there are no hyperscaler AI-chip wins announced for 18A/14A.
The honest read: Intel's AI accelerator share is structurally capped in the low single digits in 2026. The interesting question is whether Jaguar Shores in 2027 + 18A foundry wins repositions Intel as an infrastructure-tier supplier (CPUs + foundry + niche AI) rather than a head-on nvidia competitor.

5. Financials / Funding

Public: NASDAQ: INTC; market cap ~$80–110B range mid-2026 (a fraction of nvidia or amd)
Total revenue: ~$53B FY2024; data center segment a fraction of NVIDIA's
AI accelerator revenue: <$500M for Gaudi 3 (missed its own target) [5]
Capex: very high — $20B+/year across foundry buildout (Arizona, Ohio, Ireland, Israel)
Government support: ~$8.5B CHIPS Act direct funding + loans for US fab buildout

6. People & Relationships

CEO: Lip-Bu Tan (took over from Gelsinger in 2025)
AI / DC GPU lead: shifted post-Gaudi reorganization
Foundry partner: Intel Foundry (own fabs) — a strategic asset peer to tsmc but generations behind on yield
Customers (Gaudi 3): IBM Cloud, Inflection AI [1], Dell/HPE/Supermicro OEMs, Naver
Key competitors: nvidia amd huawei-ascend google-tpu aws-trainium microsoft-maia
Acquisitions in this lineage: Nervana (2016, shuttered), Habana Labs (2019, $2B → became Gaudi line)