Tim Sweeney on Unreal Engine, Fortnite, and the App Store Wars

Tim Sweeney, founder and CEO of Epic Games, traces Unreal Engine and Fortnite from a 1991 shareware game to the Nanite/Lumen rendering pipeline and a $400 million creator economy, then makes his case that Apple’s and Google’s app-store fees are a monopoly tax blocking an open, interoperable ‘metaverse’.

Key ideas

1. A game engine’s history is a history of rewrites. Sweeney describes Unreal Engine’s evolution — software rendering, then DirectX 9 shaders, then Nanite’s virtualised geometry and Lumen’s global illumination — as near-total rebuilds each generation, driven by roughly ten-million-fold GPU performance growth over thirty years, with the single-threaded core simulation loop now the last major architectural debt Unreal Engine 6 aims to pay off.
2. The metaverse is an economic and standards problem, not a VR problem. Sweeney defines it as large-scale multiplayer social gaming with a shared economy, not a headset experience — the missing pieces are portable identity and portable cosmetic items across otherwise-siloed games, which he wants industry standards bodies to define the way USD and glTF already standardise 3D asset data.
3. Verse is built to remove concurrency from the programmer’s job. Epic’s new ‘functional logic’ programming language borrows 1980s transactional-memory research so that millions of non-specialist creators can write ordinary-looking code that runs safely in parallel across a constantly-updating shared world, without manually managing thread conflicts.
4. Fortnite’s scaling story was existential, not incidental. Battle Royale shipped as a four-week pivot in 2017; the harder problem was a year-long effort to scale Epic’s backend from 40,000 to 15 million concurrent users on AWS — Sweeney says failure there would have ended the company.
5. The 30% app-store take rate is, in Sweeney’s account, a monopoly rent. Citing figures produced in Epic’s litigation discovery, he argues the all-in cost of running a store like Google Play is roughly 6% of revenue, framing the remainder as a tax that inflates consumer prices and pushes mobile game design toward manipulative monetisation rather than fun — a case he has pursued in court against both Apple and Google while underpricing his own Epic Games Store at 12%.

Content

From shareware to Unreal Engine

Sweeney taught himself to program at 11 on his brother’s IBM PC and estimates 10,000–15,000 self-directed programming hours before ever shipping commercial software. Epic began in 1991 when a text editor he was building turned into ZZT, a shareware game trilogy sold by mail order from his parents’ address; the choice to release the level editor alongside the game — letting players become creators — became, in his telling, Epic’s founding principle. Unreal Engine followed as a near-bankrupting three-and-a-half-year build (1993–1998), begun after Doom’s release left Sweeney ‘depressed’ about competing, until Michael Abrash’s published articles on texture-mapping assembly code showed the technique was learnable. Engine-licensing deals with MicroProse and GT Interactive funded the build, cementing the dual identity — game-maker and tools-maker — that Sweeney credits for Epic surviving every industry downturn since.

Engine architecture, in mechanical detail

Sweeney narrates several specific engineering breakthroughs with enough technical grain to follow the actual reasoning. Real-time constructive solid geometry — adding and subtracting simple volumes rather than hand-assembling every wall and door frame — came from a single 30-hour coding session built on a binary space partitioning tree, broken into 14 geometric cases. Dynamic lighting used ‘lightmapping’: baking a lighting value onto a coarse texture ‘wallpaper’ rather than computing every pixel, since 1990s CPUs had no cycles to spare. Volumetric fog followed from recognising the physics as a line integral — math Sweeney had studied in mechanical engineering without realising he’d ever use it — inspired by a rival hardware demo that, he later learned, had faked the same effect by pre-rendering it in 3D Studio Max.

Thirty years on, Sweeney frames the jump to Unreal Engine 5 as the single biggest leap: Nanite (led by Brian Karis) renders roughly two triangles per screen pixel regardless of source-art complexity by bypassing the GPU’s triangle-rasterisation hardware entirely and working in the pixel shader; Lumen (led by Daniel Wright) calculates how light bounces around a whole scene at multiple scales simultaneously, from kilometre to millimetre, so a scene lights itself correctly without hand-placed light behaviour. Human faces remain the hardest problem in computer graphics, Sweeney argues, because evolution gave humans dedicated neural circuitry for reading faces — any single wrong system (skin subsurface scattering, hair, a micro-expression) is immediately conspicuous, unlike an imperfect rock. MetaHuman, led by Vladimir Mastilović, works from a dataset spanning the full range of human faces so a consumer capture — a single iPhone photo — can be reconstructed against that prior.

Fortnite: origin, scaling, and the creator economy

Fortnite began as a 2011 one-week internal prototype (build forts by day, defend them by night against zombies), pivoted toward a Pixar-like stylised art direction, and shipped as Save the World in 2017 to modest success. Battle Royale itself was a four-week pivot by a 30-person ‘war room’ team responding to PUBG’s breakout — reusing seven years of accumulated Fortnite content but built almost entirely from scratch as a game mode. The harder technical story, in Sweeney’s account, was the year Epic’s online team spent scaling backend infrastructure from 40,000 to 15 million concurrent users on AWS; he credits unsung Amazon engineers who reportedly drop-shipped emergency server hardware into Brazil mid-weekend to keep pace with demand, and says failure at this stage would have ended the company.

Fortnite’s business model separates ‘engagement’ — free game modes, including tens of thousands of user-created islands — from monetisation via a cosmetic item shop, with creator revenue shared according to measured engagement across modes. Sweeney credits this mechanism for a creator economy exceeding $400 million and growing. He is explicit that neither VR nor cryptocurrency/NFTs are necessary to a metaverse economy; the actual gap, he argues, is industry standards for portable identity and portable cosmetic items across otherwise-siloed games (Fortnite, Roblox, Call of Duty), extending the same standardisation that already lets Pixar’s USD format and glTF interchange 3D scene and asset data across engines. See Metaverse.

Verse and the concurrency problem

Verse, Epic’s new ‘functional logic’ programming language, is built to let millions of non-specialist creators write safe concurrent code for a metaverse that never stops updating. Its core design choice is that an expression can produce zero, one, or many values rather than a traditional true/false boolean, making success, failure, and iteration first-class language concepts. Its concurrency model draws on 1980s transactional-memory research: object updates run speculatively in isolated buffers, and only updates with no read/write conflicts with other updates commit to shared state — removing manual thread-safety from the individual programmer’s job and relocating that difficulty to the language designers instead. Verse currently ships incrementally inside Unreal Editor for Fortnite; the general-purpose version aimed at all Unreal Engine licensees is targeted for Unreal Engine 6, several years out, which Sweeney also frames as the release that finally replaces Unreal’s single-threaded core simulation loop — a simplicity trade-off made in the 1990s that has become the engine’s biggest remaining architectural debt.

The app-store fight

Sweeney’s central antitrust argument is that Apple’s and Google’s roughly 30% app-store take rate vastly exceeds the actual cost of running a store — he cites a figure of roughly 6% of revenue, produced in discovery during Epic’s litigation against Google, as the all-in cost of operating and maintaining the Play Store. He argues the remainder functions as monopoly rent: it inflates consumer prices, and it pushes mobile game design toward manipulative pay-to-win and loot-box mechanics because a fun, non-predatory game’s profit margin often can’t absorb a 30% platform tax on top of user-acquisition costs he puts as high as 70% of revenue in aggregate. He frames Apple’s 2020 response to Epic’s in-app-payment challenge — pulling Fortnite from the App Store — as a deliberate deterrent message to every other developer, contrasted with Sony’s reversal on cross-platform play after private negotiation rather than litigation (a relationship Sweeney says has since deepened). Epic’s own store undercuts the standard rate (12% versus 30%) and has used paid exclusivity deals to attract supply; Sweeney defends this as ordinary price/incentive competition — developers freely choosing a deal — distinct from Apple and Google’s structural exclusion of any rival payment system or storefront. See App Store Economics.

On AI, the industry, and the future

Sweeney is measured on generative AI’s near-term role in game production: he expects it to multiply human creative output — enhancing scene-graph content and post-processing — rather than replace artists, arguing AI still lacks any real model of an entire scene, plot, or world the way a game engine does natively. On the industry, he reads recent AAA struggles as a ‘fun’ problem, not a mysterious one, and argues Metcalfe’s Law now dominates game design: players converge on whichever game reaches the largest share of their real friends, which explains Fortnite’s and Roblox’s continued growth and predicts a bleak outlook for single-player games without a large addressable audience. He singles out Rockstar’s long GTA development cycles as the correct trade-off — ‘a bad game is bad forever; a late good game eventually is released and is good’ — and closes by contrasting the toxicity he associates with text-based social media against the consistently positive dynamics he observes in Fortnite’s voice-chat squads with strangers, taking it as evidence that humans connecting through shared activity are naturally empathetic.

Related

• Tim Sweeney — guest
• Lex Fridman — host
• Metaverse — concept created from this source
• App Store Economics — concept created from this source
• Mark Zuckerberg on the Metaverse, Codec Avatars, and the Future of Presence — a VR-and-presence-centred account of ‘the metaverse’ on the same show, a contrasting angle to Sweeney’s standards-and-economics framing