# Baltra (AI server chip) by Apple Canonical page: https://technooptimist.io/computing/baltra Frontier: [Computing](https://technooptimist.io/computing.md) · Field: Cloud giants' own chips · Company: [Apple](https://technooptimist.io/companies/apple.md) Stage: Proto (2 of 5: Research → Proto → Pilot → Deploy → Scaling) Updated: last update 2026-07-15, facts checked 2026-10-11 Goal: Give Apple its own chip for running AI in its data centers, so cloud Siri and Apple Intelligence need less NVIDIA hardware rented from others and keep Apple's privacy controls end to end. Status: Reported delayed in July 2026 after a target of mass production this year; Apple has not confirmed it. Next milestone: Baltra-powered Apple data centers in full operation, target 2027 ## Milestones | Status | Milestone | Date | Slip | | --- | --- | --- | --- | | Done | Development with Broadcom first reported | 11 Dec 2024 | | | Missed | Mass production | 2026 | | | Planned | Baltra-powered Apple data centers in full operation | 2027 | | ## Upcoming events - **2027**: Baltra-powered Apple data centers in full operation ## Most important updates - **15 Jul 2026**: Apple's Baltra AI server chip, expected this year, is reportedly delayed as Apple looks at buying AI chip companies. The Information reported that Baltra had been expected to ship in 2026 but has been delayed, with no new date. ([our story](https://technooptimist.io/stories/apple-s-baltra-ai-server-chip-expected-this-year-is-reportedly-delayed-95536875.md) · [original source](https://www.macrumors.com/2026/07/15/apple-looking-to-acquire-ai-chip-companies/)) - **8 Apr 2026**: Apple is reportedly testing glass-core chip substrates from Samsung Electro-Mechanics, a sign it may package AI server chips itself. Korean press reported that Samsung Electro-Mechanics has been sending Apple samples of glass-core substrates, the base a chip is mounted on. ([our story](https://technooptimist.io/stories/apple-is-reportedly-testing-glass-core-chip-substrates-from-samsung-b77d2c14.md) · [original source](https://www.trendforce.com/news/2026/04/08/news-apple-reportedly-sources-glass-substrate-samples-directly-signaling-further-in-house-ai-server-chip-push/)) - **13 Jan 2026**: Analyst Ming-Chi Kuo says Apple will mass-produce its Baltra AI server chips from the second half of 2026. Supply-chain analyst Ming-Chi Kuo said Apple's first dedicated AI server chips would enter mass production in the second half of 2026. ([our story](https://technooptimist.io/stories/analyst-ming-chi-kuo-says-apple-will-mass-produce-its-baltra-ai-server-fb576ac8.md) · [original source](https://www.androidheadlines.com/2026/01/apple-ai-server-chips-baltra-mass-production-2026-kuo.html)) - **11 Dec 2024**: Apple is reportedly developing its first AI server chip, codenamed Baltra, with Broadcom, for production in 2026. The Information reported, citing people with direct knowledge, that Apple is building a chip for its AI servers, codenamed Baltra. ([our story](https://technooptimist.io/stories/apple-is-reportedly-developing-its-first-ai-server-chip-codenamed-0a03e106.md) · [original source](https://decrypt.co/296139/apple-partners-with-broadcom-to-develop-first-ai-server-chips-launch-set-for-2026)) ## Current obstacles - **A phone chip team building a server chip**: Apple's chip team has focused on battery-powered devices; its server chips reportedly could not run a Gemini-scale model, so part of Siri runs on NVIDIA. - **Schedule slip**: Baltra was expected to ship in 2026 but was reported delayed in July 2026, while Apple sounded out AI chip startups to buy. ## Physics limits - **The memory wall**: Serving big models is mostly moving weights from memory to the arithmetic units. Chip maths has grown far faster than memory bandwidth, so inference chips often sit waiting on data. - **A chip can't be bigger than the reticle**: Lithography prints at most about 26 x 33 mm per exposure, so one die can't grow past that. Large models must be split across many chips, and the links between them cost time and power. ## How it works ### Today: Mac chips in servers Apple's Private Cloud Compute runs on its own M-series chips, built for battery devices. For the heaviest jobs Apple uses NVIDIA hardware in Google's cloud. ### Purpose: Built only for answering requests Baltra is reported to be an inference chip: it runs trained models for Siri requests, writing and image tasks, rather than training new models. ### Partners: Apple designs, Broadcom links, TSMC makes Apple's Israel chip team leads the design; Broadcom reportedly supplies the networking that lets many chips work as one; TSMC fabricates. ## About Apple Apple makes the iPhone, Mac and iPad. In AI it ships Apple Intelligence, which runs small models on the device and larger ones on its own Private Cloud Compute servers. Its next-generation models are built on Google's Gemini. - Siri AI, a rebuilt assistant on Gemini-based Apple Foundation Models, shipped as a beta with iOS 27 on Sep 14, 2026. - The personal Siri promised in 2024 slipped to 2026; Apple settled the resulting false-advertising suit for $250M. - John Ternus replaced Tim Cook as CEO on Sep 1, 2026; Cook became executive chairman. - Founded: 1976 - Headquarters: US - Status: public - Listing: AAPL (NASDAQ) - Valuation: $5T (market cap, 2026-07) - Website: https://www.apple.com/newsroom/ - People: John Ternus (CEO (from Sep 1, 2026)); Tim Cook (Executive Chairman); Craig Federighi (SVP, Software Engineering) - Partners, customers, investors: Google (supplier) Company page: https://technooptimist.io/companies/apple.md ## Related programs More from Apple: - [Siri AI](https://technooptimist.io/ai/siri-ai.md) (Deploy) Everything in Computing: https://technooptimist.io/computing.md --- The Techno Optimist tracks every frontier of technology, program by program: https://technooptimist.io/ (a Markdown version of any page: add .md to its URL; site map for agents: https://technooptimist.io/llms.txt). Data API (JSON, realtime stream): https://technooptimist.io/developers. People read 5 pages free, then Reader $10/month; these Markdown pages are open to agents.