Intel stock surged 20% after hours — and for the first time in years, the bull case actually makes sense. The company posted $13.6 billion in Q1 revenue, beating analyst estimates by 11%, and CEO Lip Bu Tan delivered a narrative on the earnings call that investors have been waiting years to hear: the rise of AI agents is creating a massive new wave of CPU demand, and Intel is uniquely positioned to capture it. Five distinct forces are converging at once, and that's why the stock is going vertical.
Why Did Intel Stock Jump 20% After Hours?
Intel's Q1 2025 earnings weren't just a beat — they came with a completely reframed story. Revenue hit $13.6 billion, up 7% year-over-year, and next quarter is already guiding north of $14 billion. Yes, there was a headline loss of $3.7 billion, but strip out one-off charges tied to Mobileye and derivative payments linked to the US government's 10% stake, and Intel actually earned $1.5 billion — far better than the break-even analysts expected.
0:45
Intel Q1 earnings breakdown — revenue beat vs. analyst estimates and the one-off charges behind the headline loss
Watch at 0:45 →
But more than the numbers, it was the narrative shift that moved the stock. Intel is no longer just a legacy chipmaker desperately clinging to relevance. It's suddenly the center of five major demand stories all pointing in the same direction. The market repriced that immediately.
Do AI Agents Really Need More CPUs?
This is the core thesis — and it's more compelling than it sounds. For the past few years, the AI trade has been almost entirely about GPUs: Nvidia, TSMC, memory suppliers, power equipment, cloud capex. Intel was the most embarrassing missing piece. Why wasn't Intel booming during a computing boom? It made no sense on the surface.
Here's why: training frontier models is still a GPU story. But running agentic AI workflows is a very different workload. Orchestrating tasks across data centers, routing jobs, managing memory, handling inference, coordinating servers — all of that increases demand for the central processor. The "boring old CPU" is suddenly critical infrastructure for the agentic AI wave.
Intel's data center segment posted $5.1 billion in quarterly revenue, crushing the $4.5 billion analysts expected. On the earnings call, Lip Bu Tan made it explicit: the next wave of AI is moving from foundational model training to inference to agentic AI, and that shift is inherently CPU-intensive. A single powerful AI system inferencing on GPUs can spin off enormous amounts of CPU workload — and every vibe-coded app, every automated workflow, every agent running in the background is sitting on CPUs that run constantly for every user.
4:20
CPU-to-GPU ratio shift explained — from 1:8 historically to 1:4 today, with projections toward 8:1
Watch at 4:20 →
What Is the New CPU-to-GPU Ratio for AI?
This is where things get genuinely wild. Historically, data centers ran roughly one CPU for every eight GPUs. On the Q1 earnings call, Lip Bu Tan said that ratio has already shifted to approximately one CPU for every four GPUs. That's a 2x improvement for Intel's addressable market right there.
But the bull case goes further. Evercore ISI analyst Mark Lipacis upgraded Intel from neutral to outperform with a thesis that as AI workloads continue shifting toward inference and agents, the CPU-to-GPU ratio could eventually flip to 8:1 — eight CPUs for every one GPU. That's not a small adjustment. That's a complete inversion of the prior architecture assumption and would represent a massive structural tailwind for Intel's data center business.
Some skepticism is warranted here. Frontier models are getting bigger, tokens are getting more expensive, and there's a strong argument that inference demand will continue to require massive GPU clusters. But even the intermediate case — where the ratio stabilizes at 1:4 instead of reverting to 1:8 — is twice as good for Intel as the prior regime. That alone justifies a significant re-rating.
8:55
Terra Fab scale context — 1 million wafers per month compared to TSMC's total current output
Watch at 8:55 →
What Is Elon Musk's Terra Fab Chip Project?
The wildcard in the Intel story is Terra Fab — Elon Musk's vision for a massively vertically integrated chip manufacturing operation drawing on Tesla, SpaceX, and other Musk companies. The project would need enormous chip volumes for self-driving vehicles, humanoid robots, and potentially space-based AI data centers. Intel is reportedly in the mix to help design, manufacture, and package chips for the project.
The scale being discussed is almost incomprehensible. Musk is reportedly targeting 100,000 wafers per month initially, eventually scaling to 1 million wafers per month. For context: 1 million wafers a month is roughly 70% of TSMC's total current monthly output across all fabs. TSMC's largest individual fabs run about 100,000 wafers per month — so you're talking about the equivalent of ten leading-edge TSMC fabs just for Terra Fab.
The Wall Street Journal has flagged a more cautious view on timelines, and TSMC's own CEO has noted that fabs take two to three years to build and another one to two years to ramp to production. This is not a 2025 story. But as a long-term demand signal pointing directly at Intel's manufacturing capacity, it's a significant part of why investors are willing to underwrite a messy, expensive chip turnaround story right now.
Why Does the US Government Own a Stake in Intel?
The national security dimension of the Intel story has been building for a few years, but it's now becoming impossible to ignore. The US government took a roughly 10% stake in Intel as part of a broader strategy to establish domestic leading-edge chip manufacturing capability. The question was always: what does the government actually do with that stake beyond the symbolic vote of confidence?
The bull case — articulated early by analysts and strategists including Ben Thompson — is that government backing creates a forcing function for the hyperscalers and AI labs. If the administration has skin in the game, there's a plausible path where a dinner with all the major AI lab leads and hyperscaler CEOs ends with implicit or explicit commitments to source from Intel's domestic fabs. That demand certainty is what makes the economics of building a leading-edge fab viable. Without guaranteed volume, you can't justify the capital expenditure. With it, the whole thing becomes fundable.
Whether or not that plays out, the national security framing has shifted how investors think about Intel's risk profile. This isn't just a semiconductor company trying to claw back market share — it's potentially a strategic national asset with government backstop. That changes the valuation conversation entirely.
Is Intel Finally a Good Investment in 2025?
Intel famously missed mobile, which handed TSMC a volume advantage that compounded for a decade. Volume was destiny. You can't build the leading-edge fab without the customers, and you can't get the customers without the leading-edge fab. Intel got stuck in that loop while TSMC ran away with it.
Now, five demand signals are converging simultaneously: AI agents driving CPU demand, advanced packaging requirements pushing CPU-to-GPU ratios higher, a US government mandate for domestic chip manufacturing, Musk's Terra Fab ambitions requiring impossible amounts of silicon, and hyperscalers actively seeking supply diversification. None of these individually would be enough to turn the ship. Together, they've created a new narrative — and markets price narratives fast.
Jim Cramer put it simply: in 13 months, Lip Bu Tan took Intel from a bailout candidate to one of the most talked-about companies in the chip industry. The Big Three of CPUs — Intel, AMD, and ARM — collectively cannot produce enough to meet agentic AI demand projections. That supply constraint, combined with surging demand, means prices are going up. And Intel, finally, is on the right side of that equation.
There are still real risks: the losses under the hood, the complexity of fab buildouts, execution uncertainty on Terra Fab, and the question of whether the CPU-to-GPU ratio really does invert the way the bulls are projecting. But for the first time in years, the Intel story has a credible path forward — and the market is pricing that in aggressively.
What Do Cursor's Negative Gross Margins Mean?
Separate from the Intel story, The Information reported that Cursor — now on the cusp of a $60 billion acquisition — had negative 23% gross margins earlier this year. That's a striking number for a company generating significant revenue. But the more interesting question is what happens to those margins once inference shifts away from third-party models toward Cursor-trained or XAI-trained models running on dedicated compute like Colossus 2.
The key data point: Cursor users are staying subscribers even as the underlying inference model rotates. They're buying the product experience, not a specific model. That stickiness is what matters for long-term margin expansion. Negative gross margins today, under a scenario where inference costs drop dramatically on owned infrastructure, could look very different in 18 months. It's a bet on compute economics improving faster than competition erodes pricing power — which, given the current trajectory of the AI industry, isn't an unreasonable bet to make.
