Locator: 50003BALTRA.
I mentioned "Baltra" once before and then forgot about. Need to update it or provide a stand-alone page for Baltra. Apparently it's still a "thing."
Link to techpowerup, December 16, 2025.
About a year ago, industry insiders let slip early details regarding Apple's mysterious "Baltra" AI chip. At the time, Broadcom was reportedly assisting with the design's network technology. In addition, the "Baltra" project was linked to TSMC 3 nm N3E or N3P node processes. Recent leaks suggest an adjusted gestation schedule; instead of a predicted start of "mass production by 2026," timelines seem to be extended by another year. The reasons behind this rumored delay are not clear, but the North American giant has—officially, as of October 2025—bolstered its Apple Intelligence and Private Cloud Compute-oriented data centers with US-made advanced server equipment, likely running on familiar M2 Ultra processors. Industry moles reckon that M4-based variants are being readied for distribution.
Apple's futuristic in-house "Baltra" AI setup is expected to be inference-oriented, as predicted by Max Weinbach. The tech industry analyst imagined: "basically, I doubt they'll do a massive cluster, but maybe something closer to a GB300 style, with ~64 chips all to all with larger high bandwidth LPDDR memory. Should be significantly cheaper than most current chips, and match the needs." Last month, Mark Gurman claimed that Apple had signed up for a custom Google Gemini-powered LLM. The notorious tipster believes that the claimed $1 billion per year deal will drive next-gen personalized Siri and AI web search functions. As mentioned before, Apple's in-progress "Baltra" data center chip could be designed for efficiency and cost-effectiveness, rather than the raw processing power required for training.
Ai prompt: Is Apple's Baltra still a thing? How has it evolved? Why don't we hear much in the media about Baltra?
Google Gemini:
