The algebraic modeling language that lets you write a whiteboard equation and hand it to a computer that finds the provably best answer.
Here is a thing that is true and slightly annoying about the modern economy: an enormous number of important decisions - which power plants to run tonight, how to route ten thousand shipments, how to weight a portfolio - are not really decisions at all. They are optimization problems. Somebody has to write down what you want, what you are allowed to do, and then compute the best possible answer. AMPL is the tool a lot of those somebodies use to write it down.
AMPL stands for "A Mathematical Programming Language," which is the kind of name you get when the people naming it are researchers rather than a branding agency. And that is exactly what it is. Instead of coding an optimization problem in the fiddly, error-prone matrix format that solvers actually want, you write it the way you would write it in a textbook - variables, an objective, a list of constraints - and AMPL translates. This sounds small. It is not small. It is roughly the difference between doing arithmetic in Roman numerals and doing it in Arabic ones.
The idea came out of AT&T Bell Laboratories, the legendary research shop that also gave the world the transistor, Unix, and the C programming language. In 1985 three people there - Robert Fourer, an operations-research professor from Northwestern; David Gay, a numerical-computing specialist; and Brian Kernighan, who had literally co-written the book on C - started designing a language that would let humans talk to optimization solvers in something close to math.
What makes the founder list quietly remarkable is Kernighan. When the person who helped define how a generation writes software decides that the interesting problem is "how do we describe an optimization model cleanly," you should probably pay attention to the answer. The answer was AMPL, and in 2012 the three of them received the INFORMS Impact Prize for it - an award given not for a clever paper but for changing how an entire field works.
For years AMPL lived inside the corporate structure of Lucent Technologies, Bell Labs' eventual parent. It became a proper independent company - AMPL Optimization Inc., a California corporation - in 2013, with longtime CEO Bill Wells, who had joined in 2010, steering the commercial side. The research pedigree stayed; the go-to-market grew up around it.
Most software promises to replace your judgment. AMPL does the humbler, harder thing: it takes the decision you already need to make and finds the best possible version of it.
The pitch, stripped of jargon: describe your problem once, then let the best available math engine solve it - and keep the freedom to switch engines whenever you want.
Write objectives and constraints in natural mathematical notation for linear, nonlinear, and mixed-integer problems. The model reads the way you'd explain it to a colleague, which means fewer bugs and faster iteration.
AMPL is deliberately solver-agnostic. Point it at Gurobi today, CPLEX tomorrow, BARON or an open-source engine next quarter - same model, no rewrite. That refusal to lock you in is the whole personality.
Python (amplpy), R, C++, C#, Java and MATLAB APIs, plus cloud deployment, a VS Code extension and Google Colab support, carry a model from a prototype notebook to an operational system.
Illustrative view of AMPL's usage footprint across sectors, based on the company's stated 40+ industries. Relative, not to scale.
"I first encountered mathematical optimization at NBER, where I worked for two years in the mid-1970s."
Operations-research professor at Northwestern; the modeling-language architect who first tangled with optimization at NBER in the mid-1970s.
Numerical-computing specialist from the Bell Labs research center; the engine-room mind for how models connect to solvers.
Co-author of the classic C book with Dennis Ritchie; brought language-design instincts from Unix and C to optimization.
Everyone is busy asking whether AI will make decisions. Fewer people ask the more concrete question: once you've decided what "best" means, what actually computes it? That layer - prescriptive, provable, unglamorous - is where AMPL lives.
The fashionable framing is "decision intelligence" or "DecisionOps," and AMPL is happy to use those words. But the underlying trick is old and durable. A predictive model tells you what will probably happen. An optimization model tells you what you should do about it, subject to the constraints of the real world - budgets, capacities, physics, regulation. AMPL is machinery for the second kind of question.
Consider the flagship example the company points to: Hitachi's GridView platform, used by electricity-grid operators, runs on AMPL. Modeling a power market means juggling millions of variables - every generator, every transmission line, every hour - and the answer isn't a nice-to-have. It's whether the grid balances. When the cost of "roughly right" is a blackout, you want a model you can trust and a solver you can swap for a faster one.
That solver-agnosticism is worth dwelling on, because it's genuinely countercultural. Most infrastructure companies would love to sell you the model and the engine and the lock-in. AMPL sells you the model and then actively helps you shop for engines - commercial ones like Gurobi, CPLEX and Xpress, and free ones too. In a software era obsessed with capturing customers, deciding not to trap them is a strategy, and arguably a moat: the thing customers trust is precisely that AMPL won't box them in.
There's also a lovely lesson for anyone building infrastructure. AMPL spent decades as a research tool before it became a commercial company. It didn't pivot, didn't chase a hype cycle, didn't rebrand itself every eighteen months. It refined one good idea until the idea became load-bearing for other people's businesses. That's a slower path than most startups will tolerate - and it's why AMPL is still here at 40, quietly under the floorboards of the economy.
B2B licensing and subscriptions: commercial licenses of the language and solver bundles, cloud/deployment options, support, training and consulting - with free community and academic licenses fueling adoption.
100+ corporate clients across 40+ industries plus a large academic base - energy, logistics, finance, manufacturing and transportation - alongside partners like Nextmv, DecisionBrain and Hitachi Energy.
AMPL Optimization builds the algebraic modeling language of the same name - a tool that lets engineers and analysts describe huge, messy real-world decision problems in clean mathematical notation and then hand them to any of dozens of solvers. Born at Bell Labs in the 1980s and now an independent California company, AMPL is used by more than 100 corporations across 40-plus industries to run optimization in production, from power grids and supply chains to portfolios and airline schedules.
Last updated: