PHP is not dying. That is the first thing the Perforce 2026 PHP Landscape Report, released this month, wants you to understand. PHP tied with JavaScript as the most used language in 2026 at 72% of survey participants. PHP powers roughly 74% of all websites. Only 3% of organizations report any plan to retire their PHP applications this year. In raw market terms, PHP is as central to the modern web as it was a decade ago.
What is dying is the pipeline of people who know how to maintain it.
Perforce surveyed more than 700 PHP developers worldwide. Over half of respondents reported 15 or more years of experience with the language. Only 15% had five years of experience or less. For managers and directors, hiring is now the number-one concern — not technical debt, not performance, not security. Hiring. Twenty-four percent of respondents cited "lack of personnel with the right skills and experience" as a leading operational challenge.
Read those numbers carefully, because they describe a slow-motion crisis that most executives will not see until it is already hurting them.
The Generational Hourglass
Every long-lived language eventually faces this moment. COBOL had it first, and it defined most of what we now know about legacy technology risk. Java is starting to face it. PHP, despite its reputation as a "modern" web language among people who were not programming in the 2000s, is now in the early chapters of the same story.
The difference is that PHP's legacy surface is gigantic. WordPress alone accounts for more than 40% of the public web. Add in Magento, Drupal, Laravel-based SaaS, the thousands of bespoke PHP applications quietly running payroll, inventory, government portals, and hospital systems, and you are looking at a codebase measured in petabytes of source. Almost none of it will be rewritten. All of it will need to be maintained.
The people who understand how that code actually works have spent fifteen years figuring it out, quirk by quirk. They remember when register_globals was default-on. They know why the mysql_* functions were replaced. They have feelings about Composer. They also have mortgages, and, increasingly, retirement plans.
The AI Generation Paradox
The obvious counter-argument is that AI will fix this. PHP has 139,000 public repositories on GitHub — a training dataset on which any modern code model has been drilled thoroughly. AI coding assistants report that PHP developers see roughly 40% productivity gains and 30% fewer bugs when they adopt them. GitHub Copilot, the latest GPT-Codex variants, and Laravel-tuned copilots are all comfortable with modern PHP 8 syntax.
But here is the quiet paradox Perforce's data surfaces: AI generation may be making the maintenance problem worse, not better.
Generated code is fast to write and painfully hard to inherit. It tends to produce plausible-looking solutions that subtly violate project conventions, mix idioms across PHP versions, and paper over architectural decisions rather than respecting them. When a senior PHP engineer with fifteen years of experience writes a patch to an old Magento installation, they understand why three layers of indirection exist before the query hits the database. When an AI writes that patch, it optimizes for the local prompt. The global architecture is invisible to it.
A company that leans on AI generation to paper over its PHP skills gap will, in three to five years, discover it has multiplied the amount of code it owns by a factor of five, and reduced the number of humans who understand any of it by about the same factor in the other direction.
The Database Layer Is Worse
The database side of PHP's legacy estate is where this really starts to bite. A typical long-running PHP application has a MySQL schema that has been migrated by seventeen developers over a decade, contains tables whose columns mean different things depending on which feature flag is on, and has foreign key relationships that only one person ever truly understood. That person retires in 2027.
Vector databases, AI-assisted schema recommendations, and natural-language query generation are all wonderful new tools. They do not read old migrations. They do not know which triggers will fire when you rename a column. They do not remember why the orders table has three different status columns. Only humans do, and the humans who do are leaving.
My Take
We have spent two years celebrating the productivity upside of AI-assisted development. The Perforce report is the first piece of 2026 data I have seen that forces us to sit with the downside.
PHP is not going to go extinct because the models are bad. PHP is going to get structurally harder to maintain because we are allowing the language's institutional memory to walk out the door while we wave our new AI tools around. The fix is not more tools. The fix is apprenticeship, documentation, and a deliberate investment in getting the next generation of developers into the seats of the retiring ones — ideally while the retiring ones are still there to explain.
If you are a CTO running a PHP-heavy stack, your most important hiring decision for the rest of this decade is not another senior engineer. It is the junior you are willing to train. If you are a junior developer who keeps hearing that PHP is "legacy," reread these numbers. Seventy-four percent of the web runs on it, half the people maintaining it are about to leave, and nobody is being hired to replace them. That is not legacy. That is leverage.