Meta Is Building an In-House AI Chip and Apple and Meta Have Discussed an AI Partnership

Meta Is Building an In-House AI Chip and Apple and Meta Have Discussed an AI Partnership

Two separate stories hit the wire this week that, taken together, describe a platform company play that is different from anything Meta has attempted before — and that puts it in direct conversation with Apple, a company Meta has historically competed with rather than collaborated with.

Meta's In-House AI Chip Effort

The first: Meta is reportedly preparing to begin manufacturing its own in-house AI chip, the Meta AI chip, in September 2026, according to multiple reports including coverage from Facebook's own business channels and analyst commentary from TD Cowen. This Meta AI chip would be Meta's first fully in-house AI silicon — a step beyond the Meta Training and Inference Accelerator (MTIA) family it has been developing internally, and a direct move into the custom-AI-chip space that NVIDIA, Google (TPU), Amazon (Trainium/Inferentia), and Microsoft (Maia) have all occupied to varying degrees.

Apple and Meta AI Partnership Talks

The second: Apple and Meta have held discussions about potentially integrating Meta's generative AI model into Apple Intelligence, according to reporting from the Wall Street Journal. The two companies are longtime rivals in consumer technology — Apple's App Store policies have repeatedly put it in conflict with Meta, and the two have not historically been partnership material. An AI integration would be a notable thaw.

Why These Stories Belong Together

Meta's in-house chip strategy only makes strategic sense if Meta is serious about owning the full stack from model to silicon — which is what you would build if you expected to serve AI to hundreds of millions of users through your own products (Facebook, Instagram, WhatsApp, Ray-Ban Meta glasses) and did not want to be fully dependent on NVIDIA's roadmap and pricing. Apple Intelligence, by contrast, is the operating system-level AI play — AI embedded in the iPhone, iPad, and Mac experience. If Meta's model were to run inside Apple Intelligence, it would give Meta's AI a distribution channel it has never had: the default assistant experience on the most popular smartphone in the premium segment.

The Apple-Meta talks are the more immediately consequential of the two stories. An integration would represent a significant reversal from Apple's current posture. Apple Intelligence currently leans on its own models plus OpenAI's ChatGPT as an optional external integration. Adding Meta's model as another option — or replacing or augmenting the OpenAI integration — would be a meaningful shift in the competitive landscape of on-device and cloud AI assistants.

The honest caveat on both stories: the chip manufacturing timeline and the Apple-Meta partnership are both reported, not confirmed by the companies in a formal announcement. The chip story is circulating in analyst notes and business press; the Apple-Meta story is a WSJ report. Both should be treated as in-progress developments rather than completed deals.

The Broader Infrastructure Picture

Context for the broader AI infrastructure picture: this month has seen infrastructure commitments at every level of the stack. NVIDIA is committing up to $100 billion to OpenAI's 10-gigawatt deployment (covered in this morning's NVIDIA OpenAI partnership post). Anthropic has locked in 14.8 GW of compute across its agreements. Meta is now signaling that it wants its own silicon — not just to buy NVIDIA's, not just to rent from a cloud provider, but to manufacture. And Apple, which has been relatively quiet on the AI infrastructure story, is apparently in active discussions about which models it wants inside its own AI stack.

The through-line: the AI infrastructure buildout is no longer a story about NVIDIA selling chips to labs. It is a story about every major technology platform — chip companies, AI labs, social platforms, device manufacturers — making simultaneous, irreversible commitments to own more of the AI stack than they did eighteen months ago. Meta's in-house AI chip and the Apple-Meta talks are two more pieces of that pattern.

Related AIPress coverage: NVIDIA and OpenAI Announce a $100 Billion Partnership to Deploy 10 Gigawatts of AI Infrastructure — https://aipress.blog/post/nvidia-openai-100-billion-10-gigawatts-partnership-september-2026