Inside the AI Boom: The Server Rack Layer & Who’s Hiring
- maktinta
- Jul 10
- 9 min read
Updated: Jul 22
The parent piece — “The AI Gold Rush: Who’s Selling the Shovels” — sized the AI-server market at roughly $104 billion in 2025 and on track for ~$449 billion by 2030. That number understates how much of the work is being done by companies most readers haven’t heard of. NVIDIA designs the chip and ships a reference rack architecture; somebody else has to actually screw 72 accelerators, 18 NVSwitches, miles of copper, a coolant-distribution unit, and 120 kW worth of power shelves into a single cabinet, test it, and roll it onto a truck.
Five sub-layers in this post: U.S.-brand OEMs, Taiwanese ODMs (who actually build most of the racks), the U.S. and Mexico final-assembly footprint, in-rack liquid-cooling integrators, and the cables / power-shelf / busbar layer that gets every AI engineer’s second-most-frantic phone call. Then careers at the five companies hiring hardest right now.
1. U.S.-Brand OEMs — Dell, Super Micro, HPE, Lenovo
The OEMs are the names enterprises sign POs with. They take an NVIDIA reference design, validate it, integrate it into their support, financing, and channel, and ship it as a warrantied system. Dell entered FY2027 with a $43 billion AI-server backlog after shipping $25 billion in fiscal 2026. Super Micro guided fiscal 2026 revenue to roughly $39 billion, more than 70% of it now GPU/AI platforms, with what it says is the world’s largest liquid-cooled rack production line — roughly 5,000 racks per month. HPE’s AI infrastructure grew 52% YoY in fiscal Q2 2026; Lenovo’s ISG hit $19.2 billion with AI revenue up 84% year-over-year and now over a third of company revenue.
Player | Latest revenue | AI exposure | 12–24 mo. trajectory |
Dell | FY26 ISG $60.8B (+40%); $25B AI ships; $43B AI backlog | AI servers $9.0B in a single qtr | Backlog conversion; Project Helix w/ NVIDIA |
Super Micro | FY26 guide $38.9–40.4B (~65% growth) | ~70% AI; ~5,000 LC racks/mo | Malaysia + Texas capacity ramping |
HPE | AI infra +52% YoY Q2 FY26; Juniper net +~150% | GreenLake + bundled networking | Juniper close ’25 reshaped portfolio |
Lenovo | ISG $19.2B FY; AI rev +84% | Now >1/3 of group revenue | Neptune liquid-cooled ramp; EMEA push |
2. Taiwanese ODMs — Who Actually Builds Most of the Server Racks
The brand on the bezel and the company on the assembly line are usually not the same. Foxconn (Hon Hai) reported Q1 2026 revenue of $66.6 billion — up 29.7% YoY, with management guiding 2026 AI-server share to 55–60% (up from ~52% in 2025). Quanta booked Q1 2026 revenue of ~$25 billion, up 66.6%. Wistron and Wiwynn are running triple-digit YoY growth on GB300 ramps. Inventec and Compal round out the field. Morgan Stanley sees full-year 2026 GB200/GB300 NVL72 rack shipments landing somewhere between 70,000 and 80,000 units, against ~29,000 in 2025.
Player | GB-rack share (est.) | Latest revenue | Trajectory |
Foxconn (Hon Hai) | ~52% 2025 → 55–60% 2026 | Q1 2026 $66.6B (+29.7%); 2026 target ~$350B | 2,000 racks/wk in U.S.; 100% rack capacity 2026 |
Quanta | ~19–30% | Q1 2026 ~$25B (+66.6%) | Triple-digit AI-server growth |
Wistron / Wiwynn | ~20% combined | Triple-digit YoY Jan–Feb 2026 | Juarez + El Paso County investments |
Inventec | ~5–7% | Strong AI-server mix shift | Mexico + Texas footprint expanding |
Compal / MiTAC | <5% combined | Reshaped onto AI server work | Niche; serving select hyperscalers |
3. U.S. and Mexico Final Assembly — Nearshoring the Server Rack
Where these racks are physically built has shifted in 24 months. Foxconn’s Wisconsin and Texas footprint employs roughly 6,500 across 40 U.S. sites. Wiwynn announced a $150 million Texas facility in El Paso County in 2025; Inventec, Pegatron, and Wiwynn together have invested $75–250 million each in Ciudad Juárez over the past three years. Quanta is expanding its Tennessee plant. Super Micro runs a flagship campus in San Jose and has expanded into the Bay Area’s Houston and Texas corridors. The strategic point: as hyperscalers want GB-class racks delivered in the U.S. on tight schedules, the assembly footprint is being nearshored as fast as construction permits allow.
Site | Operator | Function | Status |
Houston / Wisconsin | Foxconn | AI server rack final assembly | ~2,000 racks/wk goal; both ramping |
El Paso County, TX | Wiwynn | AI server mfg + test | $150M facility (announced 2025) |
Ciudad Juárez, Mexico | Wiwynn / Inventec / Pegatron | Servers for U.S. DCs | $75–250M each over past 3 yrs |
Nashville-area, TN | Quanta | Server assembly + integration | Expanding for hyperscaler accounts |
San Jose / Bay Area | Super Micro | HQ + flagship LC rack production | ~5,000 LC racks/mo across plants |
4. In-Rack Liquid-Cooling Integrators — CDUs and Cold Plates
Past about 75 kW per rack, air cooling stops being a serious option. Past 100 kW, it isn’t available at all. The companies that bridge a chiller plant to a 120 kW Blackwell rack are CoolIT, Motivair, Vertiv (XDU), nVent, and Asetek. CoolIT was acquired by KKR; Motivair was acquired by Schneider Electric for $1.13 billion in 2024 specifically for the AI-data-center play. These are not optional accessories; they are the difference between a Blackwell rack functioning and not. Many are integrated at the ODM stage, which is why this layer is half-server, half-cooling.
Player | Role | Latest | Trajectory |
Vertiv (XDU) | CDUs + Liebert thermal | $15B+ backlog; ~30% YoY rev growth (Q1 2026) | Expanding CDU capacity globally |
nVent | CDU, DLC Cooling and RDHx | $4.33 B with ~26% growth projected for 2026 | $2.6B in Backlog, Expanding customer base |
Motivair | CDUs, RDHx, direct-to-chip | Acquired by Schneider Electric ($1.13B, 2024) | Schneider-scale distribution |
CoolIT Systems | Cold plates + CDUs | Acquired by KKR | Volume direct-to-chip ramp |
Asetek | Direct-to-chip + Rack CDU | Public; smaller scale | Inverse Liquid Cooling (ILC) push |
5. Cables, Power Shelves, Busbars — The Layer Nobody Photographs
Inside an NVL72 rack is something like two and a half kilometers of cable. Amphenol, TE Connectivity, Volex, and Molex (now Koch) supply the high-speed copper and connectors. Server Tech, Vertiv Geist, and Raritan supply the rack PDUs. Starline Track Busway moves power overhead at the row level. None of these companies are AI stories on the surface; all of them are growing on AI dynamics. Amphenol’s communications-solutions segment, where high-speed copper sits, has been running well above corporate average; the supply of NVLink copper cables has been a periodic bottleneck on GB-class rack shipments.
Player | What they supply | AI tailwind | Trajectory |
Amphenol | High-speed copper, connectors | NVLink copper for NVL72 | Communications-solutions segment leading group |
TE Connectivity | Connectors + high-speed cable | AI server interconnect | Capacity expanding for hyperscalers |
Volex / Molex | Cable assemblies, custom harness | Backplane + power harnesses | Volex public; Molex private (Koch) |
Server Tech / Geist / Raritan | Rack PDUs | 120+ kW racks need new PDU classes | Higher amperage SKUs expanding |
Starline | Busway overhead power | Row-level distribution at AI scale | Lead times stretched |
Careers: Five Companies, Five Buckets, One Taxonomy for Server Racks
Same fixed buckets as post #1: Engineering · Operations & SC · Finance & Accounting · Sales & BD · Corporate. The five companies below — Dell, Super Micro, Foxconn (U.S.), Quanta, and HPE — are where most of the active hiring sits right now. Counts are directional and reflect mid-June 2026 snapshots across each company’s ATS plus LinkedIn, Indeed, Glassdoor, and ZipRecruiter.
Dell Technologies — ~150 on Glassdoor, several hundred U.S. on LinkedIn
Dell’s public face is the AI server backlog. Its hiring footprint reflects an enterprise sales motion, not a research lab. Solutions engineering, account management, services delivery, and supply-chain ops all hire at scale. Round Rock (Austin), Hopkinton, and Bangalore are the gravitational centers; Dell also has a long-running Finance Development Program that recruits BS-Accounting/Finance grads.
Bucket | Open reqs (approx.) | Typical entry-level requirements | Example titles |
Engineering | ~400 | BS/MS CS / EE / CompE; storage, server, networking | Sr. Hardware Engineer; ML Systems SWE; Storage Engineer |
Operations & SC | ~150 | BS Supply Chain / IE; SAP / Kinaxis | Demand Planner; Logistics Manager; NPI Program Manager |
Finance & Accounting | ~80 | BS Accounting / Finance; FDP rotational program | Finance Development Program Analyst; FP&A Analyst, ISG |
Sales & BD | ~250 | BS technical or business; enterprise sales | Account Executive — Public Sector; Solutions Architect — AI |
Corporate | ~80 | HR / Legal / IT / Comms tracks | Senior HR Business Partner; Counsel — Commercial; IT Site Lead |
Surprise: Dell’s Finance Development Program quietly takes ~50–80 finance/accounting new grads per year. If your degree is in accounting and you’ve never seen yourself at a tech company, this is one of the cleanest entry points in the industry.
Super Micro — ~300 open in San Jose, with the most accessible technician roles in the OEM tier
Super Micro’s San Jose campus runs the world’s largest liquid-cooled rack production line. The hiring mix reflects that: heavy on hardware, manufacturing engineering, NPI, and rack-integration technicians, with a sales force scaling fast for new hyperscaler and Tier-2 accounts. Texas and Malaysia campuses are ramping in parallel.
Bucket | Open reqs (approx.) | Typical entry-level requirements | Example titles |
Engineering | ~200 | BS EE / ME / IE; rack design, PI/SI, thermal, NPI | Hardware Test Engineer; Liquid-Cooling Design Engineer; ASIC Validation |
Operations & SC | ~50 | HS to BS; warehouse, materials, planning | Warehouse Specialist; Materials Planner; Production Supervisor |
Finance & Accounting | ~20 | BS Accounting / Finance | Cost Accountant; Sr. Financial Analyst, ISG |
Sales & BD | ~50 | BS technical or business | Account Manager — Hyperscaler; Solutions Engineer — Storage |
Corporate | ~20 | HR / Legal / IT | HR Generalist; IT Service Desk Engineer |
Surprise: a warehouse / technician role at Super Micro pays in the range of $21–$33/hour in San Jose. Not a glamorous title, but the volume is real and the skill ladder into hardware testing is short.
Foxconn U.S. — ~6,500 employees across 40 sites; record-pace U.S. hiring
Foxconn’s U.S. footprint exists today specifically to absorb the GB200/GB300 rack ramp inside U.S. borders. The bulk of openings are operations-floor and manufacturing-engineering — the kind of roles where a community-college AAS or a few years of manufacturing experience opens the door without a four-year degree.
Bucket | Open reqs (approx.) | Typical entry-level requirements | Example titles |
Engineering | ~300 | AS/BS EE/ME/IE; manufacturing, test, automation | Manufacturing Engineer; Test Engineer; Process Engineer |
Operations & SC | ~250 | HS diploma+; forklift, SAP/WMS, planning | Production Operator; Warehouse Lead; Materials Planner |
Finance & Accounting | ~20 | BS Finance | Plant Controller; FP&A Analyst |
Sales & BD | ~15 | BS technical or business | Customer Program Manager |
Corporate | ~40 | HR / EH&S / IT | HR Generalist; EH&S Specialist; IT Site Engineer |
Surprise: Foxconn is now the largest single non-hyperscaler hirer of operations-floor roles in U.S. AI infrastructure — Wisconsin, Texas, Ohio, and Indiana all have openings, most of them not requiring a four-year degree.
Quanta Computer — ~3,000 engineers in Taipei alone; Tennessee plant expanding
Quanta is the second-largest GB-rack assembler and runs into the Taipei engineering-talent gap head-on: roughly 4,200 unfilled AI-hardware design positions across Taipei as of mid-2026, vacancy rates above 35% on senior architecture roles, and average time-to-fill on senior reqs around 87 days. The U.S. side is smaller but the Tennessee plant is expanding for hyperscaler accounts.
Bucket | Open reqs (approx.) | Typical entry-level requirements | Example titles |
Engineering | ~600 (TPE) / ~50 (US) | BS/MS EE / CompE / ME; SI/PI, thermal, board design | AI Server Architect; Signal Integrity Engineer; Thermal Engineer |
Operations & SC | ~150 (TPE) / ~30 (US) | BS Supply Chain / IE | NPI Program Manager; Materials Planner |
Finance & Accounting | ~25 | BS Finance / Accounting; bilingual EN/ZH valued | Financial Analyst; Cost Accountant |
Sales & BD | ~25 | BS technical or business | Customer Account Engineer — Hyperscaler |
Corporate | ~20 | HR / Legal / IT | HR Operations Specialist |
Surprise: a senior signal-integrity engineer fluent in Mandarin is currently one of the most valuable single resumes in the industry. Time-to-fill is approaching three months in Taipei.
HPE — Juniper close reshaped the company; hiring across networking + AI infra
HPE’s January 2025 close of Juniper Networks turned it into a combined compute-plus-networking story. AI-infrastructure revenue grew 52% in fiscal Q2 2026, networking nearly 150% on Juniper-acquired lines. The company is hiring across both: AI server engineering and rack solutions in Houston/Spring, networking software in Sunnyvale, and a large enterprise sales footprint everywhere.
Bucket | Open reqs (approx.) | Typical entry-level requirements | Example titles |
Engineering | ~500 | BS/MS CS / EE; Junos/Linux/storage/server | AI Solutions Engineer; ProLiant Server Engineer; Junos SW Engineer |
Operations & SC | ~80 | BS Supply Chain / IE | Demand Planner; NPI Program Manager |
Finance & Accounting | ~60 | BS Accounting / Finance | Senior Financial Analyst; Revenue Accountant |
Sales & BD | ~200 | BS technical or business; enterprise sales | AE — Federal; Solutions Architect — AI |
Corporate | ~70 | HR / Legal / IT | HR BP; Counsel — Commercial; IT Program Manager |
Surprise: HPE is hiring more Junos software engineers than at any point in its history. If you have networking-software experience, the post-Juniper integration is the single most open door in the company right now.
Where Are the Doors for getting into Data Center Server Rack?
The server layer is the most accessible part of AI infrastructure for non-engineers. Foxconn U.S. (Houston, Wisconsin, Ohio, Texas) is hiring operations and manufacturing roles at a rate the OEMs cannot match, and most of those roles open to candidates without a four-year degree. Super Micro’s San Jose campus offers a clean technician-to-test-engineer ladder for AAS-credentialed candidates. For finance and accounting grads, Dell’s Finance Development Program is one of the largest rotational programs in U.S. tech and recruits accounting majors from non-elite schools. For enterprise sales and solutions engineering, Dell and HPE are the dominant employers — both training programs accept non-technical undergrads who can talk to customers. And for anyone whose resume already says signal integrity, thermal engineering, or rack architecture, Taipei is hiring at fill-rates that haven’t been seen since the 1990s — Quanta, Wistron, and Wiwynn are all chasing the same shortlist.
Selected Sources
Dell FY2026 AI-server orders and FY2027 backlog (Dell IR; parent piece)
Super Micro FY2026 outlook and liquid-cooling production (Investing.com; Supermicro IR)
HPE Q2 FY2026 AI infrastructure +52% and Juniper networking (Channel Insider; HPE IR)
Lenovo ISG FY revenue $19.2B; AI rev +84% (Lenovo IR; parent piece)
Foxconn Q1 2026 revenue $66.6B (+29.7%) (Tech-Insider; DigiTimes)
Foxconn 2026 AI-server share 55–60% (DigiTimes; Yahoo Finance)
Quanta Q1 2026 revenue ~$25B (+66.6%) (DigiTimes)
Wistron / Wiwynn triple-digit AI-server growth (DigiTimes)
Morgan Stanley NVL72 shipment forecast for 2026 (Technetbook)
Wiwynn $150M El Paso County investment (Nikkei Asia, June 2025)
Inventec / Pegatron / Wiwynn Juárez investments (Dallas Fed)
Schneider Electric acquisition of Motivair ($1.13B, 2024)
KKR acquisition of CoolIT Systems
Taipei AI server design talent gap (KiTalent)
Foxconn U.S. footprint: 6,500 employees / 40 sites (Foxconn corporate)
Dell careers — Glassdoor; ZipRecruiter; LinkedIn (June 2026)
Super Micro careers — jobs.supermicro.com; ZipRecruiter
Quanta / Wistron / Wiwynn careers — LinkedIn (June 2026)
HPE careers — hpe.com/careers; LinkedIn
Amphenol high-speed copper communications-solutions segment (Amphenol IR)
Comments