Replit, Cursor and Apple: The brutal economics of AI coding tools
AI coding assistants were supposed to be a nice add‑on to developer tools. Instead, they are quickly turning into one of the most strategically important, capital‑intensive layers in the entire software stack. The rumored 60 billion dollar Cursor–SpaceX deal and Replit’s explosive growth show just how violently this market is reordering itself. In the middle sits Amjad Masad, trying to keep Replit independent while picking a fight with Apple and building what looks less like an IDE and more like a new kind of cloud platform. This is not just a funding story; it is a test of who will control how software gets created in the next decade.
The news in brief
According to TechCrunch’s interview with Replit CEO Amjad Masad at the StrictlyVC event in San Francisco, Replit has gone from 2.8 million dollars of revenue in all of 2024 to aiming for roughly a 1 billion dollar annual run rate just 18 months later. Masad said Replit’s net revenue retention with some customers is approaching 300 percent and that the company has been gross‑margin positive for over a year.
The backdrop is Cursor, a rival AI coding environment reportedly in talks to be acquired by SpaceX for about 60 billion dollars, with TechCrunch noting Cursor has been operating with roughly minus‑23 percent gross margins.
Masad argued that Replit’s economics and full‑stack platform let it plausibly remain independent, though he stopped short of promising never to sell. He also criticized Apple for blocking Replit’s iOS app updates, claiming Apple’s official explanation is inaccurate and hinting he is prepared to challenge the company in court. Finally, Masad floated the idea of Replit investing directly in startups that build on its platform.
Why this matters
On the surface, this is a story about one startup growing very fast while a competitor considers a giant exit. Underneath, it is about whether independent AI tooling companies can survive once the true cost of AI hits their balance sheets.
Cursor allegedly runs at deeply negative gross margins. That is a polite way of saying every incremental user destroys value unless a bigger sponsor pays the GPU bill. SpaceX can justify that if it believes tightly coupling code generation with its own software needs, data and in‑house hardware is worth tens of billions over time.
Replit is staking out a different path. By targeting non‑traditional developers and offering a full pipeline from prompt to deployed app, it captures more value per user: hosting, database, security, and soon, payments. That makes its AI spend look more like a customer acquisition cost for a broader cloud platform than a pure SaaS margin problem.
The winners here are:
- Cloud providers and chip makers, who monetize the massive inference workloads.
- Enterprises that move quickly: tools like Replit can compress months of prototyping into days.
The losers:
- Pure‑play coding assistants without a platform or distribution moat; they risk either being acquired or crushed on margins.
- Developers who prefer open, portable tooling; the more the AI coding stack merges with proprietary clouds, the higher the lock‑in.
Masad’s willingness to challenge Apple also matters. If app‑store gatekeepers can arbitrarily throttle tools that let ordinary users generate apps, then the promise of a billion new software creators is gated not by AI, but by platform politics.
The bigger picture
The Replit–Cursor contrast sits at the intersection of three big trends.
First, AI coding is moving from autocomplete to agency. GitHub Copilot, JetBrains AI, Sourcegraph Cody and others started as assistants embedded in existing workflows. Replit’s agentic experience takes the next step: describe what you want, and an AI orchestrates code, infrastructure and deployment. That is no longer a plugin; it is an operating system for building software.
Second, we are replaying the platform wars of the 2010s, but with GPUs. Heroku, Firebase and others abstracted away infrastructure until the underlying cloud vendors realized that the real power lay in owning the full stack. AI intensifies that logic. Whoever controls the models, the orchestration layer and the developer experience can steer where new applications live and what they depend on.
Cursor inside SpaceX would be the extreme version of vertical integration: bespoke tools for one giant customer, with outside users as a bonus. Replit is trying to be the opposite: a neutral, multi‑model platform that sits on top of Google Cloud and multiple AI labs. That sounds more open, but it is still a bet that Masad can afford to ride on top of foundation‑model providers without being squeezed on price or quality.
Third, this market is consolidating much faster than typical developer tooling. The jump from 2.8 million to a projected billion in revenue is not organic grassroots adoption; it is the result of enterprises realizing that AI‑native tooling is now a board‑level topic. In that environment, large incumbents (clouds, device makers, SaaS giants) will not stand by and let independent platforms own the gateway to all new apps.
The European and regional angle
For European developers and startups, this fight is not abstract. Many teams in Berlin, Barcelona or Ljubljana already rely on US‑based tools like Replit, GitHub Copilot and OpenAI for everyday work. The concentration of both infrastructure and tooling power in a few American companies – and potentially SpaceX – cuts directly across the EU’s push for digital sovereignty.
The Apple conflict is especially relevant in Europe. Under the Digital Markets Act (DMA), Apple is classified as a gatekeeper and must not unfairly discriminate against rival services. If Replit can show that Apple treats its app‑building capabilities differently from comparable apps, EU regulators will be eager to investigate, regardless of what happens in US courts. In the EU, the question is less whether Apple’s guideline interpretation is correct and more whether those guidelines themselves create unjustified barriers.
There is also a missed opportunity. Europe has world‑class research labs and a strong open‑source culture, but almost no scaled AI coding platforms. Companies like Mistral AI work at the model layer; the integrated experience Masad describes – from idea to running app with built‑in payments – is largely imported. That means European SMEs and public institutions are building critical systems on top of stacks that are legally and economically anchored elsewhere.
For cities like Zagreb or Munich that want to host serious startup ecosystems, the risk is clear: if access to AI‑powered creation tools is mediated by a small club of foreign gatekeepers, local innovation becomes dependent on their commercial and political decisions.
Looking ahead
Expect three things over the next 24 months.
First, more M and A. The rumored Cursor price sets a psychological anchor. Any AI coding startup with real traction is now imagining itself as a strategic asset, not a feature. Cloud vendors, large SaaS players and even industrial giants will look at Replit‑like platforms as a chance to own the next developer frontier. Masad says he prefers independence, but sustained 300 percent net revenue retention and high margins will only make his board’s phone ring more often.
Second, the economics of AI coding tools will become brutally transparent. As budgets normalize after the initial AI hype, CIOs will look past magical demos and ask how much code actually ends up in production, how secure it is and what it costs per business outcome. Platforms that, like Replit, tie AI coding directly to hosting, databases and analytics have an advantage: they can show end‑to‑end ROI. Thin assistants that only autocomplete in an IDE will struggle to prove their value once GPU prices and model licensing fees are fully loaded.
Third, regulation will collide with product strategy. The EU AI Act is starting to bite, and tools that generate operational software for banks, hospitals or governments will be scrutinized as high‑risk systems. At the same time, DMA enforcement on app stores will define whether Apple – and Google to a lesser extent – can pick winners in AI creation. If Brussels decides that blocking app‑building tools is a form of self‑preferencing, Apple may be forced to open far more than it would like, at least within the EU.
For Replit, the open questions are sharp: Can it keep enough distance from any single model provider to maintain bargaining power? Will it need its own custom models to protect margins? And if it does launch a fund to invest in customers, will that make it more like a mini‑ecosystem with network effects, or distract it from the core platform battle?
The bottom line
Replit’s trajectory shows that AI coding platforms can be real businesses, not just expensive demos – but only if they own more of the stack than autocomplete. Cursor’s potential sale and Apple’s resistance both point in the same direction: tools that let anyone turn ideas into apps are now politically and strategically sensitive. The open question for readers, especially in Europe, is simple: do we want that power concentrated in a handful of US giants, or are we willing to build and demand alternatives that stay independent – even when the acquisition offers hit forty or sixty billion?
