There was a stretch of the 2000s when Intel was worth around $290 billion and Nvidia was a niche graphics company Intel could have bought with loose change. Two decades later, Nvidia became the most valuable company on Earth, and Intel needed rescue capital.
No competitor did this to Intel. Intel did it to itself, one confident “no” at a time.
Key Takeaways
- Intel said no to building the iPhone’s chip: CEO Paul Otellini passed on Apple before the 2007 launch, and later admitted the volumes came in about 100 times higher than forecast.
- In 2005, Intel’s board rejected buying Nvidia for up to roughly $20 billion: Nvidia went on to become, reportedly, the first company in history to touch a $4 trillion market value in 2025.
- Nvidia launched CUDA in 2006 and funded it through years of skepticism: the 2012 AlexNet breakthrough that ignited deep learning ran on just two of its consumer GPUs.
- Intel’s 10nm manufacturing process slipped from around 2016 to 2019, handing the process-technology crown to TSMC.
- In late 2024 Intel posted a roughly $16.6 billion quarterly loss, the largest in its history, during a year it cut about 15,000 jobs.
- By 2025 the rescue was extraordinary: the U.S. government reportedly converted subsidies into a roughly 10% equity stake, and Nvidia itself invested $5 billion in its old rival.
How Dominant Was Intel at Its Peak?
Dominant enough that for thirty years “computer” effectively meant “Intel”: its x86 processors ran essentially every PC and server that mattered, behind a market value that reached roughly $290 billion. Intel wasn’t just a chipmaker. It was computing.
The pedigree matched the power. Founded in 1968 by Robert Noyce and Gordon Moore (the Moore of Moore’s Law) and hardened by Andy Grove’s “only the paranoid survive” culture, Intel spent decades as both the best chip designer and the best chip manufacturer on the planet, a combination no rival could match. The “Intel Inside” campaign, launched in 1991, turned an invisible component into one of the most valuable brands alive, and its margins were the envy of the industry.
That dominance created the trap. When you own a machine that prints money, every new opportunity gets measured against it, and almost nothing measures up.
Three Noes That Made Nvidia
No to the iPhone. Before 2007, Apple asked Intel to build the processor for a new phone. CEO Paul Otellini ran the numbers on Apple’s projected volumes and passed: the margins looked beneath Intel. He later admitted the volumes turned out to be a hundred times higher than anyone forecast. Mobile computing, the biggest chip market of the era, went to Arm designs, and Intel never got back in. It didn’t just miss mobile. It sold its way out, offloading XScale, its own Arm-based chip unit, to Marvell in 2006 for about $600 million.
“The world would have been a lot different if we’d done it,” Paul Otellini said in 2013, reflecting on the iPhone decision.
No to Nvidia. In 2005, Intel seriously considered acquiring Nvidia for as much as ~$20 billion. Some executives championed it, arguing graphics-style parallel processing could matter to the future of computing. The board decided the price was absurd for a gaming-chip company. This is the same category of miss as Excite refusing to buy Google for $1 million, except Intel’s version cost it the next age of computing.
No to the GPU future itself. Even after passing on Nvidia, Intel could have built the alternative. It tried with the Larrabee graphics project, then cancelled it as a product in 2010 when it underdelivered. Meanwhile, from 2006, Nvidia poured money into CUDA, a software layer that let researchers run general computation on GPUs, and kept funding it through years when Wall Street considered it a waste.
Then deep learning arrived. In 2012, a neural network called AlexNet, trained on just two consumer Nvidia GPUs, shattered the field’s image-recognition records, and every AI lab on the planet discovered their work ran best on exactly one company’s hardware.
How Did Intel Fall So Far Behind?
One compounding miss at a time: losing mobile starved it of volume, fumbling manufacturing cost it its process lead, and having no credible AI chip left it outside the biggest hardware boom in history. While the strategic misses accumulated, the operational crown slipped too:
| Year | Event |
|---|---|
| 2005 | Board passes on acquiring Nvidia; Apple chip deal declined around the same era |
| 2006 | Nvidia launches CUDA; Intel doubles down on CPUs |
| 2010 | Intel cancels Larrabee as a consumer GPU |
| 2015-2019 | 10nm process repeatedly delayed; TSMC and AMD surge ahead |
| 2020-2022 | Apple dumps Intel for its own silicon; AI boom ignites GPU demand |
| 2024-2025 | Historic losses and layoffs; Intel takes major outside backing, including U.S. government support, to fund its comeback bid |
Each miss amplified the others. Missing mobile meant missing the volumes that fund leading-edge factories. Slipping on manufacturing meant even its core CPUs lost their edge. And having no credible AI accelerator meant sitting out the greatest hardware gold rush ever, one that made Nvidia, at its peak, worth more than thirty Intels.
The humiliations stacked up in the stock market too: in 2022, AMD, the perennial also-ran Intel had dominated for fifty years, overtook it in market value, powered by chips manufactured at TSMC.
Could Intel Have Won the AI Era?
It had at least four real chances, and it killed or starved every one of them:
- Larrabee, the x86-based graphics project, cancelled as a consumer product in 2010.
- Xeon Phi, Larrabee’s descendant for supercomputing, wound down by around 2020.
- Nervana, an AI-chip startup bought in 2016 for a reported $400 million, cancelled in 2020.
- Habana Labs, acquired for about $2 billion in 2019, whose Gaudi accelerators reportedly missed even a modest $500 million sales target in 2024, a year when Nvidia’s data-center business was generating tens of billions per quarter.
The common thread was never a lack of money or engineering talent. Each project had to justify itself against the gravitational pull of CPU margins, and each eventually lost the argument. Meanwhile the AI boom repriced the entire stack around Nvidia’s architecture, all the way down to the memory market, where the high-bandwidth chips bolted to Nvidia hardware became the scarcest commodity in electronics.
The Critical Choice
Pick any of the three noes and you can defend it with the spreadsheet Intel was looking at that day. That’s the point. The critical choice wasn’t a single meeting. It was Intel’s decision, renewed year after year, to let the economics of its existing monopoly define what counted as a good idea.
If we have to name one moment: 2005, the Nvidia rejection. Not because buying Nvidia was obviously right (integration might have smothered it) but because the reasoning for the rejection (“parallel graphics chips are a toy market”) was the exact belief that would cost Intel the AI era. The board didn’t just decline an acquisition. It wrote down, in effect, that the future would look like the past. For more empire-scale unravelings, see our corporate collapse files.
Where Things Stand Now
Intel is still enormous, with tens of billions in revenue and chips in hundreds of millions of machines, and its foundry gamble, backed by unprecedented government involvement, is one of the most consequential industrial bets in America. The bill for the lost decade came due fast: a roughly $16.6 billion quarterly loss in late 2024, around 15,000 job cuts, and the departure of CEO Pat Gelsinger that December, with Lip-Bu Tan taking over in early 2025.
Then came the rescue capital: billions in CHIPS Act support, a U.S. government stake reported at roughly 10%, and, in the twist nobody predicted in 2005, a $5 billion investment from Nvidia itself. The company that once decided the direction of computing now navigates by other people’s stars: TSMC’s process lead, Arm’s architectures, and Nvidia’s AI empire. Empires rarely get a second founding. Intel is attempting one anyway.
