C++ isn't losing to Rust. The infra it runs is the moat.

Rust wins in any online debate. However, C++ is the one that wins in reality and gets shipped. When I saw ThePrimeagen arguing about Rust versus C++ for GPU and systems dominancy, I realized what's actually happening: It's not a safety language fight, it's a language lockdown fight.

The vote and the paycheck are different things

For the 9th consecutive year, Rust was voted the most "admired" programming language in the 2025 Stack Overflow survey. The survey gave it a 72% approval rating. As much as we would like to believe that love conquers all, the reality is that we need money to survive. As of July 2026, the TIOBE Index ranks C++ as the third most popular language globally, with a market share of 8.03%. An industry analysis from September 2025 cited that enterprise C++ usage was at 90% while Rust was at 60%. Rust may have won the popularity contest, but when it comes to earning, C++ takes the crown. 💰

The GPU stack doesn't care about your feelings

Here's something people are usually hesitant to admit. The infrastructure for every model you've constructed is implemented using C and C++. NVIDIA's CUDA, cuDNN, cuBLAS, and TensorRT are the libraries used to perform the actual tasks on GPUs, and they are based on C/C++. Do you mean to discuss on-device AI? llama.cpp and ggml were actually the gravitational center for inference. Not llama.rs. It's right in the name, it has "cpp" in it. That wasn't a coincidence. It's about which language you can't rip out. An entire industry poured concrete decades ago, and you can't move a foundation with a survey result.

Ecosystem gravity beats safety marketing

The best part of it all is a February 2026 engineering report, where over five years, a team spent hundreds of man-years "trying to fold Rust into a C++ environment."

They failed. But not for the reason you'd guess. Performance-wise, Rust was doing fine. The real problem was what we'd later come to call the "ecosystem gravity" at the FFI boundary:

→ ownership semantics that don't translate cleanly across the wall
→ error handling that needs constant re-plumbing
→ build system integration friction that never stops nagging

This is the entire story summarized in one report. It wasn't that C++ was slow; it was the expense of stepping outside the room that C++ had everything in.

C++ isn't sitting still either

According to the cope theory, C++ is a dying language that's being propped up by its past successes and the unwillingness of other languages to take on its role as the systems-programming language. Profiles introduced the concept of tiered support into the C++ language itself. This is a systematic approach to opting the programmer into certain kinds of breakage in return for compile-time and/or run-time checking. C++ creator Bjarne Stroustrup also cautioned that the language was in danger of "a painful decline" if nothing was done. That's the creator looking the safety argument straight in the face and responding to it in his own way. Having backward compatibility is like an escape hatch. You adopt new safety without giving up the 40 years of code that's holding up the entire internet. Rust asks you to rewrite it. C++ asks you to opt into it.

The AI wildcard nobody planned for

And to top it off, there was a plot twist, GitHub Copilot had reached 20 million users in 2025. And the interesting part is that AI models can generate and debug C++ better than Rust. The reason is that C++ has a much much larger public code corpus to train on. The greater the dependency we have on AI to program, the more support the ancient behemoth receives. The models grew up with C++. They look after it better. Safety marketing had assumed that the future would be a clean rewrite, when in fact, the future turned out to be an autocomplete tool trained on a mountain of legacy C++.

The takeaway

Rust is indeed an amazing language, and I don't want to criticize it. However, being the "most admired" and being the "most deployed" are two separate things. The barrier is not the language itself. It's the toolchains, the GPU stack, the FFI expense, the compiler that you can't afford to replace, and now the AI based on years of experience with it. That's an interesting point. Do you think any new systems language stands a chance with ecosystem gravity being so strong? Or is everyone simply going to keep building costly wrappers around C++ for eternity? What are your thoughts?