Every custom chip Meta deploys is a small reduction in Nvidia dependence
In the long human effort to master the tools of one's own making, Meta has reached a notable threshold: the company will begin producing its own artificial intelligence chip, called Iris, in September 2026, designed alongside Broadcom and forged by Taiwan Semiconductor. At a moment when the cost of computing ambition runs into the hundreds of billions, this is Meta's answer to a question every great builder eventually faces — not whether to spend, but whether to spend wisely. The move does not reduce the scale of what Meta is constructing, but it does suggest the company intends to own more of the means by which that construction proceeds.
- Meta's stock surged six percent on news that its custom AI chip Iris passed testing in just six weeks with no significant problems — a rare sign of execution discipline in a notoriously difficult engineering domain.
- The company is releasing a new chip generation every six months through 2027, roughly twice the industry pace, compressing years of competitive catch-up into a single aggressive product cycle.
- With plans to double computing capacity from 7 to 14 gigawatts by 2027, Meta is not retreating from its $145 billion AI infrastructure commitment — it is trying to extract more from every dollar of it.
- Custom chips won't displace Nvidia GPUs but will work alongside them, quietly eroding Nvidia's pricing leverage with each additional unit of in-house silicon Meta deploys.
- Underpinning all of it is a business growing at thirty-three percent annually with a forty-one percent operating margin — the rare financial foundation capable of sustaining this kind of industrial ambition.
Meta's stock climbed six percent after the company confirmed it will begin manufacturing its custom AI chip, Iris, in September 2026. Designed with Broadcom and built by TSMC, the chip completed testing in roughly six weeks without major issues — a promising early signal for a program that carries enormous strategic weight.
What sets this effort apart is the pace. Meta plans to release a new chip generation approximately every six months through 2027, forming a four-generation family and moving at roughly double the industry's standard cadence. The company also intends to bring seven gigawatts of computing capacity online this year and double that figure to fourteen gigawatts by 2027. Iris won't replace the Nvidia and AMD processors Meta continues to purchase — it will work alongside them, reducing the cost per unit of computing power and gradually loosening the company's dependence on outside suppliers.
The financial foundation supporting this build-out is formidable. In the first quarter, Meta posted revenue of $56.3 billion — up thirty-three percent year over year — alongside a forty-one percent operating margin and earnings per share of $10.44, a sixty-two percent increase from the prior year. The company raised its 2026 capital expenditure forecast to between $125 billion and $145 billion and spent $19.8 billion on capital projects in the first quarter alone.
Custom chips don't shrink Meta's budget; they change what that budget can buy. If Iris delivers on its promise, the same level of spending could yield meaningfully more computing power — and faster earnings growth than the market currently anticipates. Trading at roughly $672 per share, Meta controls not just the infrastructure it is building but the products that infrastructure serves. The September production date for Iris is less a cost-cutting measure than a statement of intent: Meta means to shape the economics of its own ambition.
Meta's stock jumped six percent on Friday after news broke that the company will begin manufacturing its custom artificial intelligence chip, code-named Iris, starting in September. The move signals something investors have been waiting to hear: that Meta has a credible plan to wring more value from the staggering $145 billion it expects to spend on AI infrastructure this year.
Iris was designed in-house with help from Broadcom and will be manufactured by Taiwan Semiconductor Manufacturing Company. According to an internal memo reviewed by Reuters, the chip completed testing in about six weeks with no major problems identified. What matters most is the pace. Meta is planning to release a new chip roughly every six months through 2027—roughly double the industry standard of one new chip per year. This is a four-generation family of chips, and the company is moving faster than most observers expected.
The scale of what Meta is building is difficult to grasp. The company plans to bring seven gigawatts of computing capacity online this year, then double that to fourteen gigawatts in 2027. These custom chips won't replace the graphics processing units Meta buys from Nvidia and Advanced Micro Devices. Instead, they'll work alongside them, filling gaps and reducing the cost per unit of computing power. Every custom chip Meta deploys is, in effect, a small reduction in how much the company needs to depend on Nvidia's expensive processors.
None of this spending would matter if Meta's core business were struggling. It isn't. In the first quarter, revenue climbed thirty-three percent year over year to $56.3 billion, accelerating from twenty-four percent growth in the previous quarter. The company posted a forty-one percent operating margin and earnings per share of $10.44, up sixty-two percent from the year before. For a company of Meta's size, this kind of growth is unusual. CEO Mark Zuckerberg called it a milestone quarter, pointing to strong momentum across the company's apps and the release of its first model from Meta Superintelligence Labs.
That revenue growth is what funds the build-out. Meta raised its 2026 capital expenditure forecast to between $125 billion and $145 billion, up from a prior range of $115 billion to $135 billion. In the first quarter alone, the company spent $19.8 billion on capital projects. The company also guided for second-quarter revenue between $58 billion and $61 billion.
It's important to understand what custom chips actually mean for Meta's budget. They don't shrink it. The company is not cutting spending; it's doubling its computing capacity while trying to reduce the cost of each unit. If the AI investments fail to produce more user engagement and better advertising economics, in-house silicon won't be enough to protect the company from challenges ahead. But if the chips work as planned, Meta's massive spending could convert into earnings growth faster than the market currently expects.
At roughly $672 per share, Meta trades at about twenty-four times current earnings and nineteen times forward earnings, despite growing revenue thirty-three percent last quarter. Unlike a pure chip supplier, Meta also controls the applications that all this computing power serves. The company isn't just building infrastructure; it's building infrastructure for its own products. If the in-house chips deliver even part of their potential savings, the economics could shift meaningfully in Meta's favor. The September production date for Iris doesn't lower Meta's AI bill, but it does strengthen the case that the company can control the cost of a build-out it was always going to attempt.
Citações Notáveis
We had a milestone quarter with strong momentum across our apps and the release of our first model from Meta Superintelligence Labs— Mark Zuckerberg, CEO, in Q1 earnings release
A Conversa do Hearth Outra perspectiva sobre a história
Why does the market care that Meta is making its own chips? Isn't that just a cost-cutting move?
It's more than cost-cutting—it's about control. Meta is spending $145 billion on AI this year. If they can design chips that do the same work as Nvidia's for less money, they've just found a lever to make that massive spending more efficient. Every dollar saved per unit of computing power compounds across billions of dollars.
But they're not replacing Nvidia chips, right? They're supplementing them.
Exactly. Iris works alongside Nvidia's GPUs, not instead of them. But that's the point. Over time, as Meta's custom chips mature and prove themselves, the company needs less of Nvidia's expensive silicon. It's a slow shift in the balance of power, not a sudden replacement.
The pace seems aggressive—a new chip every six months. Is that realistic?
The testing cycle was only six weeks with no major issues, according to the memo. Meta is clearly confident. And they have the resources to iterate quickly. Whether they hit every deadline is another question, but the fact that they're planning this aggressively tells you they believe it's possible.
What happens if the AI investments don't actually drive more engagement or better ad economics?
Then you have a company that spent hundreds of billions on infrastructure that didn't move the needle on the business. But that's a risk Meta is taking regardless of whether they make their own chips. The custom silicon just makes that bet slightly less expensive.
So the stock looks cheap to you?
At twenty-four times earnings for a company growing revenue thirty-three percent? Yes. And Meta controls both the hardware and the software running on it. That's an advantage most chip companies don't have.