For 35 years, Arm Holdings (NASDAQ:ARM) designed the blueprints and let everyone else build the house. That era ended Tuesday night.
The British semiconductor giant unveiled the AGI CPU at its "Arm Everywhere" event in San Francisco -- the company's first-ever in-house chip, purpose-built for AI inference in data centers. Meta Platforms (NASDAQ:META) is the launch customer. Taiwan Semiconductor (NYSE:TSM) is manufacturing it on its cutting-edge 3-nanometer process. And CEO Rene Haas told the audience the new chip business alone would generate $15 billion in annual revenue by 2031, pushing total company sales to $25 billion, which is six times the $4 billion Arm generated in fiscal 2026.
Shares of Arm Holdings (ARM) surged roughly 16% on Wednesday to around $154, adding more than $20 billion in market value in a single session.
Here's the thing: this isn't just a product launch. It's the most consequential strategic pivot in the semiconductor industry since Nvidia abandoned gaming GPUs to chase data centers. And the market is only beginning to price it in.
Arm's traditional business was elegant but limited. The company designed processor architectures, licensed them to Apple, Qualcomm, Samsung, and dozens of others, then collected royalties on every chip shipped. The model produced 97.5% gross margins and a loyal ecosystem, but it also capped revenue. You can only charge so much for a blueprint.
The AGI CPU changes the math entirely. Instead of collecting a few cents per chip in royalties, Arm is now selling the chip itself. A 136-core processor drawing 300 watts, optimized for the agentic AI workloads that are driving the next wave of data center spending. Meta's head of infrastructure, Santosh Janardhan, said the chip "significantly improves our data center performance density." Mohamed Awad, Arm's head of cloud AI, put the efficiency advantage in concrete terms: "You can get two times the performance-per-watt than you can from an x86 rack."
That performance claim matters because power is the single biggest constraint in AI data center expansion. Meta is spending up to $135 billion on capital expenditures this year alone. When your electricity bill runs into the billions, a chip that delivers twice the compute per watt isn't a nice-to-have -- it's a strategic imperative.
The customer list extends well beyond Meta. OpenAI, Cloudflare, SAP, Cerebras, and SK Telecom have all signed on as early adopters. More than 50 companies, including Nvidia (NASDAQ:NVDA), Broadcom (NASDAQ:AVGO), Google, Microsoft, AWS, Samsung, SK hynix, and Micron, have publicly endorsed the platform expansion. Server systems from Lenovo, Quanta Computer, and ASRock Rack are available now, with broader volume expected in the second half of 2026.
The analyst reaction has been remarkably one-directional. Guggenheim's John DiFucci raised his price target to $240 -- the most bullish call on the Street -- citing Arm's "incredible 97.5% gross profit margin" and the scalability of the new chip business. Evercore ISI went to $227.
HSBC's Frank Lee double-upgraded the stock from Reduce to Buy and more than doubled his target to $205, arguing that Wall Street is "underestimating the game-changing impact" of AI on Arm's business. RBC Capital raised to $175. Raymond James upgraded to Outperform with a $166 target.
Even the skeptics aren't selling. Bank of America maintained Neutral but acknowledged the strategic significance, setting a $140 floor target, essentially where the stock traded yesterday.
The consensus average sits around $152, which means the stock has already blown through the Street's median expectation in a single day. That's the kind of re-rating event that forces model revisions across every institutional desk covering semiconductors.
Arm Holdings (ARM) -- The obvious primary position. At around $154, shares trade at roughly 38 times fiscal 2027 revenue estimates, which looks steep until you consider the trajectory. If Haas delivers on the $25 billion revenue target by 2031, and the company maintains even half its current margin structure, you're looking at $9 in earnings per share, which would put today's price at roughly 17 times 2031 earnings. Guggenheim's $240 target implies 56% upside. The risk is execution: designing a chip is one thing; manufacturing and supporting it at scale is another. But the fact that working test chips already exist and server systems are shipping now suggests the hardest engineering problems have been solved. Next earnings: May 6.
Taiwan Semiconductor (TSM) -- Every Arm chip will be manufactured on TSMC's 3nm process. At around $340, TSM is already the world's most valuable semiconductor company at $1.78 trillion, but each new fabless customer entering the silicon market adds incremental demand to TSMC's fabs. Arm isn't replacing an existing TSMC customer; it's additive volume. Wall Street expects $35.4 billion in revenue next quarter. The stock has gained roughly 90% over the past year yet still trades at a reasonable 33 times earnings for a company with 70% foundry market share and zero credible competition at the leading edge. Next earnings: April 17.
Qualcomm (QCOM) -- The contrarian value play. At around $130, Qualcomm (NASDAQ:QCOM) is down 28% from its January peak, beaten down by a memory chip shortage and fears about Apple's modem transition. But here's what the pessimists are missing: Qualcomm is the single largest licensee of Arm architecture on the planet, and its Snapdragon chips are increasingly targeting data center and automotive AI applications -- exactly the markets Arm's AGI CPU validates. CEO Cristiano Amon told the Wall Street Journal that "the biggest opportunity of 2026 is the opportunity that exists on the edge." Automotive revenue jumped 15% last quarter to $1.1 billion. The company just announced a $20 billion buyback at 12 times forward earnings. That's a massive capital return to shareholders from a company generating $44.9 billion in annual revenue. Average analyst target: $167; implying 28% upside. Next earnings: April 28.
iShares Semiconductor ETF (SOXX) -- For investors who want broad exposure without picking individual winners in the Arm architecture expansion, iShares Semiconductor ETF (NASDAQ:SOXX) holds 30 U.S.-listed semiconductor companies including ARM, TSM, Qualcomm, Nvidia, Broadcom, and Marvell. At around $345, the ETF has returned 72% over the past year and provides diversified access to the entire chip value chain, from designers to foundries to equipment makers. The 0.34% expense ratio is reasonable for a targeted sector fund.
Let's be honest about the risks. Arm is entering a business -- physical chip manufacturing and sales -- that requires capabilities it has never needed before: supply chain management; customer support at scale; product liability; and the capital intensity that comes with maintaining a hardware product line. The company spent $71 million building chip labs in Austin, Texas, and grew that team to over 1,000 people. That's real operational complexity for a company that previously thrived on asset-light IP licensing.
There's also the competitive question. Arm is now simultaneously a supplier and competitor to some of its biggest licensees. Qualcomm; Amazon (with its Graviton chips); and Google (with Axion) all build their own Arm-based processors. Will they continue paying royalties as enthusiastically to a company that's also competing for the same data center sockets? Haas has said licensing and chip businesses are complementary but tension is real.
And valuation. At $154; Arm carries roughly a $160 billion market cap on $4 billion in annual revenue. That's 40 times sales. Even if you believe 2031 targets; market is pricing in a lot of execution over five years.
My read: opportunity dwarfs risk. Arm's architecture already runs in 99% of world's smartphones. The data center is next frontier; AGI CPU is beachhead. Dips are buying opportunities.
Three catalysts in next 45 days. First; Arm's fiscal Q4 earnings on May 6 -- this will be first call where management faces live questions about AGI CPU order volumes; production ramp timelines; and impact on existing licensing relationships.
Second; TSMC's earnings on April 17 -- any commentary about demand from new fabless customers or 3nm capacity allocation could confirm scale of Arm's manufacturing commitments.
Third; Qualcomm earnings on April 28 -- watch for commentary on memory shortage resolution and automotive AI chip demand; both of which validate broader Arm architecture thesis for edge and data center computing.
Arm spent 35 years building most widely adopted chip architecture on Earth. Now it's selling chips themselves. If Haas is right about $25 billion in revenue by 2031; Wednesday's 16% rally will look like opening act.