Intel's former CEO tried to buy Nvidia almost 2 decades ago

[Daniel Nenni]

An Intel sign in front of company headquarters. - Photo: Justin Sullivan (Getty Images)

Tech pioneer Intel (INTC) has seemingly missed out on the artificial intelligence boom — and part of it can reportedly be traced back to a decision not to buy the chipmaker at the center of it all almost two decades ago.

Intel’s former chief executive Paul Otellini wanted to buy Nvidia in 2005 when the chipmaker was mostly known for making computer graphics chips, which some executives thought had potential for data centers, The New York Times (NYT) reported, citing unnamed people familiar with the matter. However, Intel’s board did not approve of the $20 billion acquisition — which would’ve been the company’s most expensive yet — and Otellini dropped the effort, according to The New York Times.

Instead, the board was reportedly more interested in an in-house graphics project called Larrabee, which was led by now-chief executive Pat Gelsinger.

Almost two decades later, Nvidia (NVDA) has become the second-most valuable public company in the world and continuously exceeds Wall Street’s high expectations. Intel, on the other hand, has seen its shares fall around 53% so far this year and is now worth less than $100 billion — around 30 times less than Nvidia’s $3.4 trillion market cap.

In August, Intel shares fell 27% after it missed revenue expectations with its second-quarter earnings and announced layoffs. The company missed profit expectations partly due to its decision to “more quickly ramp” its Core Ultra artificial intelligence CPUs, or core processing units, that can handle AI applications, Gelsinger said on the company’s earnings call.

And Nvidia wasn’t the only AI darling Intel missed out on.

Over a decade after passing on Nvidia, Intel made another strategic miss by reportedly deciding not to buy a stake in OpenAI, which had not yet kicked off the current AI hype with the release of ChatGPT in November 2022.

Former Intel chief executive Bob Swan didn’t think OpenAI’s generative AI models would come to market soon enough for the investment to be worth it, Reuters reported, citing unnamed people familiar with the matter. The AI startup had been interested in Intel, sources told Reuters (TRI), so it could depend less on Nvidia and build its own infrastructure.

Intel's former CEO tried to buy Nvidia almost 2 decades ago

Tech pioneer Intel (INTC) has seemingly missed out on the artificial intelligence boom — and part of it can reportedly be traced back to a decision not to buy the chipmaker at the center of it all almost two decades ago.

www.yahoo.com

 

[VCT]

If Yahoo had bought Google in the early 2000s, would Google still be the company it is today?

 

[Daniel Nenni]

If Yahoo had bought Google in the early 2000s, would Google still be the company it is today?

It would have been the death of Nvidia for sure, unless Jensen took over the Intel CEO position and cleaned the board.

 

[tedc]

It would have been the death of Nvidia for sure, unless Jensen took over the Intel CEO position and cleaned the board.

Wasn't one of the sticking points that Jensen wanted to be CEO of the combined company (i.e., CEO of Intel)?

 

[samwilde]

When AMD bought ATI it nearly killed the company and they had to divest the whole fab business going fabless.
Because they used all the funds they had saved up to build their next generation fabs on the acquisition.
AMD were also left with next to no money to develop their next generation products. At that point AMD's CPU business was hopelessly behind Intel.

It is a near miracle the company is still around.

 

[vslee]

It is a near miracle the company is still around.

That miracle was Lisa Su.

 

[samwilde]

That miracle was Lisa Su.

Let us not overstate things. The ground for AMD's recovery was laid down before she became CEO.
For example she became CEO in 2014, while Jim Keller had returned to AMD in 2012.

AMD's Zen core was in development before she became the CEO. The trajectory was defined by the previous leadership.

AMD also managed to survive their dry spell thanks to deals like the Playstation 4 and Xbox One design wins. Which brought some much needed cashflow. The Playstation 4 and Xbox One started being sold in 2013.

Both designs used the Jaguar core. Which was also designed before she became CEO. It was derived from Bobcat.

All of this happened when Mark Papermaster was CTO. It was he who hired Jim Keller back. But you can credit AMD's recovery more to their engineering teams than to the management in my opinion.

Lisa Su is a fine CEO, but all of this kind of landed on her lap.

 

[vslee]

Let us not overstate things. The ground for AMD's recovery was laid down before she became CEO.

Let's also not understate things. Yes the building blocks existed in the company when she took the helm. But properly managing them and shepherding success is no easy task.

Lisa Su is a fine CEO, but all of this kind of landed on her lap.

In the same vein, Intel had the building blocks to be both Nvidia and TSMC today. But somehow, neither landed in their lap.

 

[samwilde]

Fine. The two years before the launch of AMD Zen were particularly difficult in financial terms, and she was the CEO back then. She got AMD out of the last hole by doing that deal to license the AMD Zen core to the Chinese at Hygon.

But like I said, Mark Papermaster has been CTO since 2011, yet for some reason he is not credited with AMD's success to nearly the same degree as she is. Despite him being there all the way. Part of the reason is he is kind of subdued, he is one of those people who keeps a low profile yet keeps things moving.

Like I said, you can credit the engineering team, they are the ones who got AMD out of the hole. People like Jim Keller, Mike Clark, and tens or hundreds of others.

 

[blueone]

I'm not buying the NYT article at all. Only unnamed sources, which probably means nobodies who weren't directly involved. A BoD voting against a CEO-backed proposal, from a newly appointed CEO (PSO was appointed CEO in 2005), is the same as a no-confidence vote in the new CEO, so that's difficult to believe. And that the BoD liked Larrabee better? Who on the BoD at the time was qualified to judge Larrabee versus a GPU? No one, so that would mean one of Otellini's own staff lobbied the BoD against him? Larrabee was still prototyping in 2009, so what was available in 2005 to drive a pivotal BoD decision? Then that quip about some executives thought GPUs would have significant datacenter applications... in 2005? This article smells like a steaming pile of

 

[vslee]

But like I said, Mark Papermaster has been CTO since 2011, yet for some reason he is not credited with AMD's success to nearly the same degree as she is. Despite him being there all the way. Part of the reason is he is kind of subdued, he is one of those people who keeps a low profile yet keeps things moving.

There are probably several reasons for that. Subdued might be one, but also CEOs usually get the glory or blame. Unfairly or not.

I do agree at that the CTO's job is important as well. But even with good technology, success is not guaranteed. See again, Intel.
On the flip side, if you do want to split hairs, then was it Papermaster that developed Zen? Or was it Keller who he hired? Or was it the people under Keller who were doing the actual work? You could go way down the tree, but ultimately, it's a team effort. And the leader is responsible for guiding that ship in the end.
But I agree that there have also been many instances where leaders simply kept their heads down and glided to success on the backs of their team. So you are right to question it.

 

[samwilde]

I agree that there have also been many instances where leaders simply kept their heads down and glided to success on the backs of their team. So you are right to question it.

AMD had a good team which worked. Everyone knew that the company was in dire straits and everyone worked to pull through out of the hole. Their financial condition after the ATI purchase was just miserable. AMD is a real survivor company, they have been close to bankruptcy several times in their history, and they managed to get out every time.

After Bulldozer flopped, they took a long time to transition to a new microarchitecture, there was no money for fabs after the ATI purchase either, a lot of people in engineering lost faith in management for realistic reasons.
It was good to bring Jim Keller back to AMD from Apple to kind of shake off the idea of continual tweaks to the failed Bulldozer architecture out of the team and start the new Zen architecture. That kind of attitude had led them nowhere.

When you have a hard working team like that, the best thing a CEO can do is get out of the way, and facilitate the conditions for the team to do the work properly. Which is what happened both with Rory Read and with Lisa Su. They kept the funding alive for Zen core development through some really dire economic conditions internally.

You could say that Intel currently is stuck into a similar mindset with continuous tweak to the same failed P-cores. It is a shame that Jim Keller's Royal Core idea didn't stick at Intel.

 

[samwilde]

I'm not buying the NYT article at all. Only unnamed sources, which probably means nobodies who weren't directly involved. A BoD voting against a CEO-backed proposal, from a newly appointed CEO (PSO was appointed CEO in 2005), is the same as a no-confidence vote in the new CEO, so that's difficult to believe. And that the BoD liked Larrabee better? Who on the BoD at the time was qualified to judge Larrabee versus a GPU? No one, so that would mean one of Otellini's own staff lobbied the BoD against him? Larrabee was still prototyping in 2009, so what was available in 2005 to drive a pivotal BoD decision? Then that quip about some executives thought GPUs would have significant datacenter applications... in 2005? This article smells like a steaming pile of

Intel has been in the graphics business since the i740 in the late 1990s. They kept developing GPUs all the time it is just that they were always low or mid end GPUs. The kind of thing you integrate in the motherboard.

It is easy for people to forget that in the mid 2000s Intel was under regulator scrutiny as it was being prosecuted for acting as a monopoly. So them buying out NVIDIA back then, already the biggest GPU company which had absorbed 3dfx, could have gotten them into a lot of trouble.

I think that is what made Intel give up on the NVIDIA acquisition, government regulators, not Larrabee. Which like you rightly state came much later. Curiously I remember back then the rumors of Intel buying NVIDIA were one of the reasons which led to AMD buying ATI and nearly bankrupting themselves in the process.

Larrabee was a kind of pie in the sky project. I do not think there were these huge expectations about it. But yes it was originally targeted as a high end graphics processor architecture based on x86. That they were doing with Mike Abrash and some other people. But those in the know kind of expected it to be a flop to begin with. Using an x86 based architecture to compete with a dedicated architecture on graphics, was always going to have some overhead, and after Mike Abrash and some other people left, the software side of the graphics stack was dead on arrival. Oh did I mention that the Larabee hardware was designed by the same people which designed the original Itanium? After trying to use VLIW on the general purpose processor market and failing, they decided to use a general purpose processor where they would have been better off using a VLIW architecture.

 

[blueone]

Oh did I mention that the Larabee hardware was designed by the same people which designed the original Itanium? After trying to use VLIW on the general purpose processor market and failing, they decided to use a general purpose processor where they would have been better off using a VLIW architecture.

No, not VLIW, GPUs are better implemented on SIMD systolic arrays, ya know, like Nvidia uses. The Cerebras WSE is also a systolic array apparently. AMD's GPU too. For AI problems VLIW has at least one interesting implementation (Gaudi), but VLIW (even with hints like Itanium EPIC) takes a lot of application affinity to succeed. I'm impressed by what I see in Gaudi, but only customers can eventually tell the real story. For Larrabee it just always seemed like the rule was, "The answer is x86, now what's the problem?".

 

[samwilde]

Intel at some point just went bonkers with x86 in everything. After trying to kill x86 with Itanium and failing they then did a 180 and tried to use x86 to replace XScale, replace GPUs, replace Itanium, replace everything.

 

[blueone]

Intel at some point just went bonkers with x86 in everything. After trying to kill x86 with Itanium and failing they then did a 180 and tried to use x86 to replace XScale, replace GPUs, replace Itanium, replace everything.

And the king of bonker-land is now CEO...

 

[samwilde]

I think Intel's attempts to capture the GPU market with Intel Arc, and the Gaudi AI chips prove the opposite.
I do not even think those efforts were necessarily bad. It is just that both have been long hard sloughs in terms of software stack, and it would take huge continuous investments for Intel to break through in either market. And Intel ran out of cash.

 

[BruceA]

Intel’s culture and arrogance is strong. There is no case of successful integration of an acquisition that flourished.

All this talk is wishful dreaming of what would have been another failed acquisition probably killed Nvidia, I guess that would be a win for Intel from another prospective, LOL

 

[Paul2]

Like I said, you can credit the engineering team, they are the ones who got AMD out of the hole. People like Jim Keller, Mike Clark, and tens or hundreds of others.

The missed mention in the AMD's recovery story is a huge number of South Asian VLSI outsourcers, and a lot of cheap 3rd party ip, which they were previously too proud to use.

 

[samwilde]

The missed mention in the AMD's recovery story is a huge number of South Asian VLSI outsourcers, and a lot of cheap 3rd party ip, which they were previously too proud to use.

AMD used to use some internal tools I think. They switched to regular EDA tools some time after the merger with ATI.
The guys at ATI wouldn't touch AMD's tools with a barge pole, and AMD's engineers knew something had to change so they went into commercial EDA. Which facilitates using 3rd party ip. Which would have previously been next to impossible to incorporate.