AI Infra · Chip Talent Map · United States

This compute war
is Intel veterans fanning out to hunt the 1/7 of NVIDIA's engineers locked behind golden handcuffs

We x-rayed 15 U.S. AI-compute / chip companies and 56,861 current engineering and research professionals, reading their résumés as data. The conclusion is more than "chips are short on people": what's truly scarce are the "dual-stack bridge people" who reach down to the silicon and up to the model alike — even at NVIDIA they're only about one in seven. And the factory supplying them is one nobody wants to admit to: Intel.

Report Date 2026-06-16 Produced By Metix AI Coverage 15 companies · United States
Executive Summary

01 The real strategic asset is the "dual-stack bridge person"

The figures below follow Metix AI database methodology (data as of roughly 2026 H1). The population = current engineering and research professionals at 15 U.S. AI-compute / chip companies (current employees pulled by authoritative company_id, then cleaned to a "hardware design + ML systems" role scope). All figures are aggregate statistics; the report displays no personal information.

56,861
Current chip / AI-infra engineering & research professionals
15 companies · United States · aggregate visible sample
~1 / 7
"dual-stack bridge people" among NVIDIA engineers
14.6%; just 8.4% across the full map (~1/12)
40 months
NVIDIA core median current tenure
44% are past 4 years — deeply vested
9 / 14
companies whose #1 talent feeder is Intel
including NVIDIA, AMD and most challengers
The real strategic asset is the "dual-stack bridge person"

People who understand both physical silicon (RTL / verification / physical design) and ML systems (CUDA / compilers / kernels) are the hard currency of this race. Of the 56,861 across the full map, only 8.4% are identifiable as dual-stack, and even NVIDIA has only 14.6% (~1/7); at legacy chip giants they're nearly extinct (Broadcom 2.2%, Intel 5.0%). Dual-stack density is the thermometer of how AI-native a company really is.

Intel is this war's "involuntary talent factory"

It is, among the 14 other companies, the #1 talent feeder for 9 of them — 13% of NVIDIA, 21% of AMD, 32% of Rivos and 31% of Tenstorrent come from Intel, and of those roughly 89% held technical roles at Intel. The foundation of the compute race is, to a large degree, being laid by senior engineers who flowed out of Intel.

NVIDIA is the only "retention fortress"

Massive inflow, minimal outflow. As a talent source for other companies, NVIDIA generally accounts for only 5–9% (the lone exception is Groq, at 34%).40 monthsMedian tenure + a roughly 10x move in NVDA stock = the hardest golden handcuffs in the industry, locking the 44% of "4-years-plus veterans" in place.

Every challenger's "lineage" is written in its résumés

You can source straight from it: SambaNova = Oracle / Sun (41%), Rivos = the Apple-chip crew (the ones Apple sued), Groq = ex-NVIDIA (34%), AWS's in-house chip team Annapurna = internal Amazon transfers (54%). Read the lineage and you know which company to raid for which kind of engineer.

About this report. Covers 6 chip / GPU giants (NVIDIA, AMD, Broadcom, Marvell, Qualcomm, Intel) + 8 AI-accelerator challengers (Groq, Cerebras, SambaNova, Etched, Tenstorrent, d-Matrix, Lightmatter, Rivos) + AWS's in-house chip team Annapurna. All figures are aggregate statistics; no personal information is shown. The same x-ray can be generated on demand for any target company; full lists and candidate outreach are available through the Metix AI platform.
The Roster

02Stock Map: scale lives at the giants, density at the startups

Start with the pie. These 15 companies hold 56,861 current engineering and research professionals in the U.S., but the distribution is wildly uneven: the 6 chip / GPU giants account for 55,478 (97.6%), while the 8 challengers combined hold just 1,290. The starkest contrast: Intel, with the most people (21,549), sits near the bottom on AI dual-stack density (5.0%).

Intel
21,549 · dual-stack 5.0%
NVIDIA
12,134 · dual-stack 14.6%
Qualcomm
8,196 · dual-stack 8.1%
AMD
6,139 · dual-stack 11.7%
Broadcom
5,313 · dual-stack 2.2%
Marvell
2,147 · dual-stack 5.2%
8 challengers combined
1,290 · dual-stack 20.7%
Bars = count of current engineering & research professionals (U.S., cleaned by authoritative company_id + role scope). Purple = the GPU duo, light = legacy / networking / mobile chip makers, green = AI-accelerator challengers. Source: Metix AI

Split each company into three buckets — "metal-leaning / model-leaning / dual-stack" — and the picture snaps into focus: legacy chip makers are pure-hardware battalions, while AI startups tilt the scales toward the model side. Below is the full roster (median tenure / median career computed to the report date).

CompanyCampEng & researchMetal %Model %Dual-stack %Median tenureMedian careerPhD %
NVIDIAGPU duo12,13451.531.514.640 mo15.0 yr14.1
AMDGPU duo6,13967.216.711.736 mo15.1 yr10.8
IntelLegacy chip21,54949.98.75.068 mo16.5 yr20.1
QualcommLegacy chip8,19648.115.08.158 mo15.8 yr12.1
BroadcomLegacy chip5,31342.04.82.290 mo21.8 yr9.9
MarvellLegacy chip2,14765.26.25.251 mo19.6 yr11.1
SambaNovaChallenger17961.545.326.848 mo14.7 yr6.7
EtchedChallenger14177.327.021.312 mo12.5 yr8.5
TenstorrentChallenger30484.526.321.116 mo14.9 yr12.5
CerebrasChallenger22249.139.619.824 mo15.4 yr15.8
d-MatrixChallenger8868.259.140.918 mo15.5 yr22.7
LightmatterChallenger15778.317.812.717 mo15.4 yr30.6
RivosChallenger14087.110.09.340 mo13.8 yr8.6
GroqChallenger5947.533.920.330 mo16.7 yr6.8
AWS AnnapurnaCloud in-house9371.059.141.919 mo10.4 yr15.1
Scale and density are out of sync. Challengers' dual-stack density (20.7%) is 2.6x that of the giants (8.0%) — but the giants have 43x more people. Run the absolute numbers: of the roughly 4,758 identifiable dual-stack professionals across the full map, 94% still sit inside the 6 giants, with NVIDIA alone accounting for 37% (~1,770 people). Startups are higher-grade ore, but the pie is small; they can fight at the margin, but not out-muscle the giants on scale.
The Scarce Prize

03Dual-Stack Scarcity: a monotonic "AI-native" gradient

Line up "dual-stack density" by company and you get an almost perfectly monotonic gradient: from Broadcom (2.2%), which only does networking / analog chips, climbing all the way to d-Matrix and AWS Annapurna (~41%), built from the ground up to make chips for large models. The closer a company sits to large models, the more "understands both silicon and models" people its résumés contain.

Share of "dual-stack bridge people" by company (hardware ∩ ML systems)

Broadcom
2.2%
Intel
5.0%
Marvell
5.2%
Qualcomm
8.1%
AMD
11.7%
NVIDIA
14.6%
Cerebras
19.8%
Tenstorrent
21.1%
Etched
21.3%
SambaNova
26.8%
d-Matrix
40.9%
AWS Annapurna
41.9%
Dual-stack = the same person's history (current + past titles + skills) carries both a "model" signal (ML / deep learning / PyTorch / NLP …) and a "metal" signal (RTL / ASIC / physical design / CUDA / compilers / kernels / interconnect …). Source: Metix AI

Three tiers

Legacy chip makers Broadcom / Intel / Marvell / Qualcomm ≈ 2–8%. They are pure-silicon battalions (metal-leaning 42–65%, model-leaning often single digits), with dual-stack all but extinct.

GPU duo AMD / NVIDIA ≈ 12–15%. Living at the "GPU × ML" intersection for years, they carry the thickest dual-stack layer among the giants.

AI-accelerator challengers20–42%. Built for large models, they pull the model-side talent share to two or three times that of the giants.

This is a "floor," but the gradient is real

Dual-stack is identified by keywords from self-reported titles + skills, so it tends to undercount (many people never fill in all their skills). But even under a stricter "physical silicon ∩ model" scope (dropping CUDA / compilers from the metal side), the gradient still holds monotonically: Broadcom 1.7% → NVIDIA 8.4% → d-Matrix 34%.

In other words: the absolute numbers go higher, the relative ranking doesn't move. On the question of who is scarcer, the conclusion is robust.

The Intel Foundry & The Bloodlines

04The Intel Factory, and every company's "lineage"

Where do the people come from? The answer is surprisingly consistent: Intel. It is the #1 talent feeder for 9 of the other 14 companies — and the more hardcore the silicon startup, the higher Intel's share. AI may be the war Intel lost, but many of the soldiers fighting it were sent out by Intel.

Share of "ex-Intel" employees by company (of that company's eng & research talent)

Rivos
32%
Tenstorrent
31%
d-Matrix
24%
Lightmatter
24%
AMD
21%
Cerebras
17%
NVIDIA
13%
Marvell
10%
Qualcomm
8%
"Ex-Intel" = among that company's current eng & research professionals, those whose history includes a formal Intel role (deduplicated by person). Of them, roughly 89% held technical roles at Intel — real engineers, not short internships. Source: Metix AI

Beyond Intel, the remaining 5 "non-Intel-lineage" companies each have their own origin — and every one can be verified in the public record. Read the lineage and you know where to raid for which kind of engineer:

CompanyTop lineageSharePublic verification
GroqEx-NVIDIA34%The founder came from Google's first-gen TPU, but the engineering bench is NVIDIA-bred — "founder lineage ≠ team lineage."
SambaNovaOracle + Sun24% + 17%Co-founder Rodrigo Liang comes from the SPARC-processor lineage at Oracle / Sun, and the whole team carries database-hardware DNA.
RivosApple20%Sued by Apple in 2022 for hiring its chip team and stealing SoC secrets, settled in early 2024 — the data precisely confirms that cohort of Apple-silicon people.
EtchedApple + Intel16% + 13%Founded in 2022 by Harvard dropouts to build a Transformer-specific ASIC (Sohu), filling out its hardware bench by hiring Apple / Intel silicon veterans.
AWS AnnapurnaInternal Amazon54%The Trainium / Inferentia team is driven mainly by internal transfers — a cloud provider's in-house chips are "grown from within," not hired from the market.
Mind the difference between "market hiring" and "M&A." Some giants' "talent sources" are really acquisitions: Broadcom's #1 source is VMware (11%, acquisition completed in 2023), and AMD's sources include Xilinx (acquired in 2022). These people were "bought in," not hired from the market; treat them separately when reading flows, or you'll mistake M&A for hiring prowess.
The Golden Handcuffs

05NVIDIA's Golden Handcuffs: lots in, little out

Intel's people scatter outward; NVIDIA barely leaks anyone — as a talent source for other companies it generally accounts for only 5–9% (the lone exception being Groq). The reason is written in the tenure: the median current tenure of NVIDIA's U.S. core is 40 months, and 44% have already passed 4 years. Layer on NVDA's roughly 10x run across 2023–2025 — and these are the hardest golden handcuffs in the industry today.

Current-tenure distribution of NVIDIA's eng & research talent (n≈11,894)

<12 months
13%
12–24 months
18%
24–48 months
25%
48+ months
44%
Tenure = current NVIDIA role start date → report date. Bars show each band's share (totaling 100%). Source: Metix AI

The mobility window is the 31% under 24 months

Roughly 3,700 people have been on board under two years, their initial RSUs far from fully vested — they forfeit the least paper gains by jumping, making them the most realistic mobility window in the NVIDIA camp. The further you go past 48 months, the tighter the handcuffs: that 44% (~5,200 people) hold deeply in-the-money vested stock and are nearly impossible to pry loose.

Startups play it exactly the other way

Etched (median tenure 12 mo), Tenstorrent (16 mo) and Lightmatter (17 mo) are all very "young" — they use pre-IPO equity to run the reverse play: pry people out before they fully vest at a giant, betting pre-IPO options against the giant's already-realized certainty.

An honest note on methodology. The database has no compensation figures. This section uses "current tenure × public stock performance" as a visible proxy for the comp-lock structure (vesting / golden handcuffs), not for pay itself. The NVDA move is public-market information used to explain lock-in strength, not data from this database.
The Recruiter Playbook

06Hunter Profile & Field Playbook

Translate the structure above into executable sourcing moves. Remember three things first: this is a war fought with veterans, you source by metal / model tags split into lanes, and timing decides whether you can pry someone loose.

All veterans, no new grads

The full-map median career is 13–22 years; even the youngest, Etched, sits at 12.5 years, and Broadcom is as high as 21.8. The compute war is fought with senior engineers — don't bring a "new-grad hiring" budget and pitch to this market.

Source by metal / model tags

Want pure silicon design (RTL / physical design / verification) → go to Rivos (metal-leaning 87%), Tenstorrent (85%), Lightmatter (78%), Etched (77%).
Want Compilers / ML systems → go to AWS Annapurna, d-Matrix (model-leaning 59%), SambaNova (45%), Cerebras (40%).

PhD density shapes the pitch

Want a research foundation (photonics / analog / compilers) → Lightmatter (PhD 30.6%), d-Matrix (22.7%), Intel (20.1%).
Want delivery-focused engineers where a PhD isn't necessary → SambaNova (6.7%), Groq (6.8%), Etched (8.5%).

The battle map in one line. If you want the scarce dual-stack bridge people, the pool actually sits inside NVIDIA / Intel / AMD (94% are at the giants) — to raid NVIDIA target the <24-months layer, to raid Intel go after any tenure (it's already leaking people); to fill hardware go to the Rivos / Tenstorrent group, to fill the model side go to the Annapurna / d-Matrix group; use pre-IPO equity and move before the other side fully vests. Read the lineage, watch the tenure, split by tags — the rest is speed.
FAQ

Questions this report answers

How many U.S. AI-compute / chip professionals are in this x-ray?
We x-rayed 15 U.S. AI-compute / chip companies and 56,861 current engineering and research professionals, reading their résumés as data.
What population definition produced the 56,861 count?
The figures below follow Metix AI database methodology (data as of roughly 2026 H1). The population = current engineering and research professionals at 15 U.S. AI-compute / chip companies (current employees pulled by authoritative company_id, then cleaned to a "hardware design + ML systems" role scope).
How should this report be cited?
Metix AI Talent Intelligence, 2026-06-16. AI Compute & Chip Talent Map 2026 | Metix AI. https://metix.ai/reports/mapping/ai-infra-chip-talent-2026

Want the full list of a specific slice of these 56,861 people?

This report is an aggregate x-ray; to get down to named individuals, Metix AI can filter a contactable candidate list by "dual-stack / metal-leaning / model-leaning + company + tenure window," and can generate the same talent x-ray for any target company.

Start a Mira trial
Aggregate report · no personal information shown · provided by Metix AI · Mira
Methodology note: This report is based on the Metix AI global talent database. The population is current engineering and research professionals at 15 U.S. AI-compute / chip companies (current employees pulled by authoritative company_id, cleaned to a hardware + ML systems role scope), with data as of roughly 2026 H1. "Dual-stack" is a keyword-identification scope based on self-reported titles + skills and tends to undercount; among cloud in-house chips, only AWS Annapurna can be identified as a standalone entity — Google TPU / Microsoft Maia / Meta MTIA cannot be carved out from their parents and serve only as flow context. Figures are an aggregate scope over the visible sample, for reference only; the report displays no personal names, contact details or sensitive attributes.
Metix AI · Mira | AI Compute & Chip Talent Map 2026 | 2026-06-16 Talent analytics powered by Metix AI
中文版