07 — study and Role Map (2026)

Heavy-C roles in 2026 are more concentrated but also better-paid than they were five years ago. The shift is real: general application backends have moved to Go and Rust, general frontends have moved to TypeScript, and general ML has moved to Python. What that leaves for C is the specialist end of the spectrum — kernels, drivers, databases, network daemons, ML inference kernels, embedded, and the compiler stack itself. This file maps who is hiring, what they pay, and how to decode their job descriptions.

Counts and salaries below are verified July 2026. All ranges are base + typical target bonus, USD unless marked.

The global picture — who is hiring heavy C

Twelve companies pay serious money for people who write C or C++ that runs at the hardware boundary. This is not “companies that have C in their stack” (that’s thousands); it is “companies where a strong C engineer is core, not peripheral.”

Company

What they build in C

Level of C

Total comp (senior)

Signal to send

NVIDIA

CUDA runtime, driver, cuDNN, TensorRT, GPU kernels

C++ heavy, C at boundary

$180K–$400K US

CUDA kernel PR, GPU perf writeup

Cloudflare

Edge C/C++ legacy, Rust (Pingora), eBPF

Rust dominant, C for legacy + kernel

$180K–$350K US, remote-friendly

eBPF or Pingora contribution

AWS (Valkey / EBS / Nitro)

Valkey, EBS storage engine, Nitro cards, Firecracker

C, Rust, C++

$200K–$450K US

Valkey PR, Firecracker PR

Redis (Redis Inc.)

Redis core (post-SSPL)

Pure C

$150K–$300K US

Redis PR + engagement

ClickHouse

Column-store C++

C++ heavy with C interop

$180K–$350K US, remote-friendly

ClickHouse PR

Neon

Postgres-compatible serverless storage

Rust + C

$150K–$300K US remote

Postgres extension, Neon PR

TigerBeetle

Zig-first financial DB

Zig, algorithms

$150K–$250K US, remote

Read TIGER_STYLE.md, contribute Zig

Modular

Mojo compiler, MAX runtime

C++, MLIR, LLVM

$200K–$400K US Bay Area

LLVM/MLIR contribution

Ampere / Marvell

ARM64 server CPUs, DPDK, kernel drivers

C, kernel-level

$180K–$350K US

Linux kernel patch, DPDK PR

The Linux Foundation projects (Linux, systemd, glibc, LLVM)

Everything

Pure C or C++

Varies, prestige > pay

LKML patch, systemd PR

Small ML-inference startups (Fireworks, Together, Anyscale-adjacent, Modal Labs, MosaicML descendants, Baseten)

Kernels, runtimes, inference servers

C++/CUDA/C

$150K–$350K US

Speedup writeup

Loophole Labs / Fly.io / Vercel Edge / etc.

Microvms, eBPF, WASM runtimes

C, Rust, Zig

$150K–$300K US remote

eBPF/CRIU/microvm contribution

Cloudflare posted 167 open roles across engineering in 2026. NVIDIA posted 319 open engineering roles, with the Systems Software Engineer New Grad 2026 slot in Hillsboro, OR listed at $124K–$242K (jobs.nvidia.com). Cloudflare Systems Engineer median levels.fyi total comp: $209K. Both companies also hire heavily in Bangalore/Hyderabad at a discount to US comp (typically 40–55% of US for equivalent leveling in Bangalore, per levels.fyi India 2026 data).

The India picture — where a Zoho engineer can actually land

The Indian heavy-C market is small but growing. Below is what is actually hiring in 2026, not just “has C somewhere.”

Company

What they build

Comp range (INR/yr)

Base city

Zoho (internal mobility to Catalyst, Vault, Zeta, low-level teams)

Product infra in C/C++

₹8–12 L fresh, ₹15–25 L senior

Chennai, Renigunta, Tenkasi

Postman (India infra)

API tooling infra, C for perf paths

₹20–45 L senior

Bangalore, remote

Nvidia Bangalore

CUDA, driver, GPU perf

₹25–80 L senior

Bangalore

AMD India

ROCm, compiler, kernel

₹20–60 L senior

Bangalore, Hyderabad

Intel India

Compiler, DPDK, ISA

₹20–55 L senior

Bangalore

Qualcomm India

Modem firmware, DSP, kernel

₹20―55 L senior

Bangalore, Hyderabad

Cisco / Arista / Palo Alto Networks India

Networking firmware, ASIC drivers

₹20–50 L senior

Bangalore

VASBM (Bangalore)

Analog/mixed-signal semi IP

Confidential, competitive

Bangalore

RRP Electronics (Hyderabad)

DSP for ISRO radar

₹12–25 L senior

Hyderabad

Saankhya Labs

Custom silicon, LEO modems for ISRO

₹15–35 L senior

Bangalore

ISRO (direct)

Aerospace, embedded

Government scale, ₹8–25 L

Bangalore, Trivandrum, Sriharikota

Defense electronics (BEL, HAL, DRDO labs)

Weapons systems, avionics

Government scale

Multiple

Bengaluru chip startups (Cerebras India, Groq India-adjacent, Tenstorrent India)

Compiler + inference stack

₹30–80 L senior

Bangalore

Glassdoor India as of July 2026 lists 197 open roles for “C programmer” or “C developer” in the country, up from ~130 in 2024. Bangalore accounts for ~40% of these; Hyderabad ~20%; Chennai and Pune the balance.

Zoho-specific note: Zoho has internal teams at Chennai/Renigunta/Tenkasi campuses that write C for the Catalyst platform, Zeta cash card, and internal storage/DB layers. Internal transfer is the highest-EV move for a current Zoho engineer — no compensation reset, no study process to survive, and “applied ML → systems” is a rare-enough trajectory internally that you become memorable.

Decoding the JD — what the words mean

Recruiter language obscures what a role actually is. Translate these terms honestly:

JD phrase

What it usually means

Ask about

“Systems engineer”

Anything from Linux kernel to Kubernetes glue

Which kernel? Which layer?

“Platform engineer”

Usually SRE-adjacent, Go/Python

How much C do you actually write per week?

“Performance engineer”

Real C or C++ perf work

Which perf tools daily?

“C/C++ developer”

Legacy maintenance, often

Ratio of new code to maintenance

“Low-level engineer”

Usually driver / firmware / kernel

Kernel? Firmware? Assembly?

“ML infrastructure engineer”

Ranges from Python glue to CUDA kernels

Do I write CUDA? How often?

“Inference optimization engineer”

High-signal C++/CUDA role

Which kernels? Which model families?

“Compiler engineer”

LLVM/GCC/MLIR contribution role

Which project upstream?

“Embedded engineer”

Firmware in C, sometimes RTOS

Which MCU? Which RTOS?

Every role above rewards different signals. Systems wants Linux internals, gdb fluency, and a war story. Perf wants profiler mastery (perf, VTune, Instruments) and a measurable speedup you produced. ML inference wants either a CUDA speedup or a llama.cpp PR. Compiler wants an LLVM/GCC patch or a compelling pass writeup. Aim your PR (from 05_contributing_to_oss_c.md) at the type of role you actually want.

The study loop, decoded

With some variation, heavy-C study loops have four to six stages:

  1. Recruiter screen — 30 minutes, culture and comp range. Do not undersell here; ask their band.

  2. Coding screen — typically a systems-flavored problem (parse a binary format, implement a bounded queue, write a memory pool). C or C++ allowed.

  3. Technical deep-dive — the study partner picks something on your resume and asks about it for 60 minutes. This is where your open-source PR earns its keep.

  4. Systems design — usually “design a rate limiter,” “design a KV store,” “design an inference server.” Not the FAANG web-scale flavor; the actual systems flavor: pick a memory model, discuss cache behavior, discuss what breaks under load.

  5. Debugging round (increasingly common at Cloudflare, NVIDIA, Valkey team) — they hand you a broken C program and you fix it live. Practice this on old bugs in your own repos.

  6. Behavioral / final — usually with a director. Don’t wing this; prep three war stories with numbers.

What senior study partners actually test for

Across all the loops, three signals dominate hiring decisions. Optimize for these.

  1. Debug reflex. When they show you a bug, do you jump to gdb / printf / strace / perf immediately, or do you stare at the code? The right answer is instrument-then-look.

  2. Vocabulary. Can you say “L2 cache miss,” “page fault,” “data race,” “strict aliasing,” “undefined behavior at signed overflow,” “symbol versioning” with confidence and correctness?

  3. Follow-through evidence. “I got a PR merged and here is the review link” beats “I studied X for 6 months” every time.

Levels and target comp for a strong new senior C hire (US, 2026)

Use this table for compensation negotiation. Values from levels.fyi, Glassdoor, and Hacker News “who is hiring” threads Jan–May 2026.

Level

Years

Base

Bonus

Equity/yr

Total

L4 (senior new-grad heavy)

0–2

$130–160K

10–15%

$30–80K

$160–260K

L5 (senior)

3–6

$170–220K

15–20%

$80–160K

$250–400K

L6 (staff)

6–9

$220–300K

20–25%

$150–350K

$350–700K

L7 (senior staff / principal)

9+

$280–400K

25%+

$300K+

$500K–$1M+

India equivalents are typically 35–55% of the US number for the same level at the same company (per levels.fyi India 2026 data). Zoho and defense electronics run separately and lower; Nvidia/Cloudflare/AMD India are close to that percentage.

What most people get wrong about this

They optimize their resume for FAANG when they want a systems role. FAANG’s C headcount is small and concentrated in a few teams (Meta’s PyTorch runtime, Google’s Fuchsia, Apple’s kernel). If you want heavy C, the best offers are outside FAANG — Cloudflare, NVIDIA, Modular, TigerBeetle, ClickHouse, Neon, and the crop of ML-inference startups. Aim there and your open-source signal earns exponentially more than at a general-purpose FAANG loop.

Second thing they get wrong: they think study prep = LeetCode. It is not. study prep for these roles is (a) merge a PR, (b) build a fast thing, (c) write about it, (d) rehearse the debugging round with a friend. LeetCode is 10% of the work at most.

Return to README.md · Next: 08_war_stories_the_hard_lessons.md