The chip has a code name and a deadline. Meta plans to deploy a new generation of self-developed AI chips in the first half of 2027, fabricated by TSMC, according to a September 16 report from TrendForce. The company’s official position is that the chips will be more energy-efficient than Nvidia’s offerings. Inside Meta, the effort carries the internal designation “Arke,” and it is aimed at a single number: the cost of running AI in 2027.
The chip is part of a quiet campaign to lower what Meta pays to reason. The company runs one of the largest AI workloads on the planet, serving models to billions of users across its apps, and the inference bill — the cost of answering each query — has become a line item that the company’s finance team watches closely. A chip of its own is the company’s answer to that bill.
The TrendForce report says the new chip will be produced on TSMC’s advanced process and deployed starting in the first half of 2027. A separate report from qz.com filled in the detail: the “Arke” effort is tied to the MTIA 450, the latest in Meta’s line of in-house accelerators, and its goal is to bring the cost of inference down before the next wave of models arrives.
Meta has been building its own silicon for years, and the progress has been uneven. The company’s earlier MTIA chips handled recommendation workloads but not the frontier models that power its chatbots. The 2027 generation is meant to close that gap, taking on inference at a scale that would otherwise flow to Nvidia as revenue.
The business logic is blunt. Nvidia’s accelerators are the industry standard, but they come at a price that has made every large customer think hard about alternatives. Meta is one of the few companies with the scale and the engineering bench to actually build an alternative, and the 2027 target suggests the company has decided the savings justify the effort.
On the same day, Meta turned its attention to the revenue side of the ledger. The company is bundling subscriptions for Instagram, WhatsApp, Facebook, and Meta AI into a single offering called Meta One, according to the reports, charging one price for what had been a set of separate products.
The bundle is a recognition that subscriptions, once an afterthought for a company that lived on advertising, have become a real business. By packaging its most popular paid features together, Meta can raise the average revenue per user while giving customers a reason to stay subscribed across multiple apps rather than churning from one to the next.
Analysts said the two moves describe a company hedging its biggest bet. Meta’s advertising machine is still the source of nearly all its revenue, but the chip push lowers its costs and the bundle raises its non-advertising income. Both lines fall outside advertising, and together they reduce how much the company’s fate depends on any single quarter’s ad market.
The chip effort also has a geopolitical edge. Fabricating with TSMC keeps Meta inside the supply chain that the U.S. government is trying to keep ahead of China, and it signals that Meta is willing to pay for domestic and allied manufacturing even as it designs its own silicon. The company has said little about where the chips will be packaged or deployed.
For TSMC, Meta is another marquee customer in a foundry business that now counts nearly every major AI player. The relationship reinforces the foundry’s position as the industry’s indispensable manufacturer, and it gives TSMC a hedge against any single customer’s fortunes.
Meta has not confirmed the timeline or the code name, and the company rarely discusses its silicon roadmap in public. But the direction is unmistakable: the company that once bought all of its computing from others is now building its own, one chip generation at a time, and bundling the results into a subscription it hopes will outlast the ad cycle.
The test will come in 2027. If Arke delivers the efficiency Meta is promising, the company’s inference costs fall and its margin story improves. If it does not, Meta will have spent years and billions learning what Nvidia already knew. The company is betting the answer will be the former, and it has now written that bet into a chip with a name and a deadline.
The bundle faces its own test, sooner and more public. Subscriptions have never been Meta’s strength, and folding three apps and an AI assistant into one price asks users to value the whole package rather than any single app. The company has done the arithmetic on retention; the market will now do the arithmetic on whether the bundle attracts new paying users or simply repackages the ones Meta already had.
Both bets share a common premise: that Meta’s future lies in owning more of its stack, from the silicon that runs its models to the subscriptions that pay for them. The advertising business that built the company is not going away, but Meta has clearly decided it cannot be the whole story anymore.
