How Intel Spurned OpenAI and Fell Behind the Times 63
An anonymous reader shares a report: For U.S. chip giant Intel, the darling of the computer age before it fell on harder times in the AI era, things might have been quite different. About seven years ago, the company had the chance to buy a stake in OpenAI, then a fledgling non-profit research organization working in a little-known field called generative AI, four people with direct knowledge of those discussions told Reuters.
Over several months in 2017 and 2018, executives at the two companies discussed various options, including Intel buying a 15% stake for $1 billion in cash, three of the people said. They also discussed Intel taking an additional 15% stake in OpenAI if it made hardware for the startup at cost price, two people said. Intel ultimately decided against a deal, partly because then-CEO Bob Swan did not think generative AI models would make it to market in the near future and thus repay the chipmaker's investment, according to three of the sources, who all requested anonymity to discuss confidential matters.
Over several months in 2017 and 2018, executives at the two companies discussed various options, including Intel buying a 15% stake for $1 billion in cash, three of the people said. They also discussed Intel taking an additional 15% stake in OpenAI if it made hardware for the startup at cost price, two people said. Intel ultimately decided against a deal, partly because then-CEO Bob Swan did not think generative AI models would make it to market in the near future and thus repay the chipmaker's investment, according to three of the sources, who all requested anonymity to discuss confidential matters.
Investment (Score:1)
>partly because then-CEO Bob Swan did not think generative AI models would make it to market in the near future and thus repay the chipmaker's investment
He's right, it wouldn't have paid back the investment, even now.
Re:Investment (Score:5, Interesting)
Except that of course, it would've made Intel what Nvidia is today. Selling shovels to the gold miners is how you profit in a gold rush, and selling AI accelerators to model trainers is how you make money in AI rush.
Intel just made their own GPUs as AI boom began, and they could've been the ones optimized for AI rather than Nvidia with CUDA.
Re:Investment (Score:4, Insightful)
Except that of course, it would've made Intel what Nvidia is today. Selling shovels to the gold miners is how you profit in a gold rush, and selling AI accelerators to model trainers is how you make money in AI rush.
Except that, of course, there is no gold there. The spectacular marketing success of pretending that we have AI still leaves the problem that we don't. Intel was prudent to avoid being sucked in, which is more than you can say for Google or Microsoft.
Re:Investment (Score:5, Insightful)
Re: (Score:2)
Re: (Score:2)
There were companies during the dot-com that over-invested in the future (that was not going to happen) and in the end died under a pile of unsustainable debt. Not saying Nvidia is going to have that happen, just that is how someone that should be in a profitable business can sink themselves.
MCI/WorldCom was one, their accounting scandal was them trying to cover up/pretty up the reports to look like they were healthy after they overestimated the need for networking and spend too much money building out a n
Re: (Score:3)
Yet. We don't yet have AI. Maybe we'll never have what people think AI should be, but along the way discoveries will be made which will affect us. Improvements will be made and the processes will advance.
Apparently in your world since it doesn't exist now it's a flop and will never exist. The definition of shortsightedness.
Re: Investment (Score:2)
The success of the mining venture is irrelevant to the shovel maker, as long as someone keeps buying.
And of course, at some point everyone will stop buying, but you still made piles and piles of money.
The billion dollar investment they are talking about would have been quite reasonable for Intel, and could have kept their GPU business relevant and producing at the right scale. Could have been the forcing function to improve their driver quality, etc. to remain competitive with nvidia.
That's a lot of could'v
Re: (Score:3)
The problem being that Bob Swan wasn't a tech guy, he was a finance guy, and had zero clue what Intel really does.
Re:Investment (Score:5, Interesting)
Re: (Score:2)
>Intel's problem is that it can't see the benefit of producing something like CUDA that does not generate license fees.
And if it saw benefit in powering OpenAI's models to be the future of everything compute, it would have.
We seem to be in agreement.
Re: (Score:2)
In 2017, Intel was already 10 years behind on supporting AI. The first release of NVidia's CUDA was in 2007. Intel has been making GPUs since 1998, Intel's problem is that it can't see the benefit of producing something like CUDA that does not generate license fees.
CUDA is not AI, but you're not far off the point. NVIDIA released dedicated AI hardware in 2017 which had been in development years prior in the form of the Tesla V100, along with it CUDA 7.0 launched to support their Tensor cores and represented the first true AI acceleration hardware.
Re: (Score:2)
Even having a pre-trained model for sale, it seems, is enough to cash in. Because they can just sell it as a service to people who are sure it's the next big thing and let them try to monetize it. Either way, OpenAI gets a payday even without the users getting value. It's true they make money selling direct to consumer, but I'm not sure to what extent considering their costs.
Re: (Score:3)
OpenAI is burning money. As are pretty much all other AI model builders. This is because of extreme costs involved, in terms of hardware and energy burned to run it.
There are two ideas on how to make AI profitable in the future. The first is extreme optimization of the models to each use case. We're at the early stages of seeing it in fields like phone customer support and image recognition (radiology etc). These are going to increase efficiency of these fields many times over in the next few years, that is
Re: (Score:2)
That assumes that Intel could have developed the technology that Nvidia has developed. Not necessarily the case for a lumbering dinosaur now run by bean counters.
Re: (Score:2)
There are two other options:
OpenAI would be influenced by Intel's hardware more and tailor their models to work better on it.
Intel would be influenced by OpenAI's technology, and develop their completely new GPU architecture to serve those models, without all the legacy GPU stuff that nvidia's AI accelerators come with.
Latter is almost certain, as intel's choice for its discrete GPUs has been to drop a lot of legacy technology, like native support for directx9.
Re: (Score:2)
That's not entirely true. NV has the most desirable combination of hardware and software available for setting up your own LLM training system. Even if Intel owned a piece of OpenAI, it wouldn't mean that the market would be clamoring for Gaudi 3 (or what have you).
Re: (Score:2)
If intel had internal access to OpenAI, it would be pushed to adapt its completely new GPU architecture to their models. Remind yourself of the timing of the offer from the OP:
>About seven years ago
Re: (Score:2)
And which new GPU architecture would that be? Ponte Vecchio? Would even Intel be stupid enough to try something like that?
Re: (Score:2)
Are you seriously asking about naming conventions differential in alternate history of technology going back seven years?
Re: (Score:2)
No, dumbass. Intel ALREADY INVESTED IN TWO AI HARDWARE COMPANIES AND CREATED THEIR OWN GPGPU PRODUCT IN HOUSE. Buying 15% of OpenAI seven years ago wouldn't have changed ANY OF THAT. The hardware would still be the same. They'd still be stuck with Loihi, Gaudi, and Ponte Vecchio/Xe.
Re: (Score:2)
Seriously, all the people who think technology development works like research tree in a game of Civilization need to take a chill pill.
Re: (Score:2)
No, you forget your history. GPUs don't appear magically overnight. Intel started developing ARC back in 2016. They would be no better off now than they were back then if they were onboard with OpenAI. NVIDIA would have still eaten their lunch several times over having already released dedicated AI hardware (Tesla V100 accelerator) the same year that Intel was according to this article talking about AI.
Even if they jumped on board with OpenAI and thrown their considerable resources at it they would be where
Re: (Score:2)
The amount of people who do not understand what "development" even means is astounding.
Truly, Civilization games ruined a lot of people's understanding of reality. Development in real life is nothing like Civilization's development tree, where you need to go through all previous technologies to reach the relevant one.
Re: (Score:2)
Re: (Score:1)
This is what happens when you let bean counters manage complex products.
Re: (Score:1)
Re: (Score:2)
I would say at a stock once valued at $70 falling to $20.
"AI era"? (Score:2)
in the AI era
but we don't have AI yet
Re: (Score:2)
Comment removed (Score:4, Interesting)
Re:Understandable (Score:5, Interesting)
Considering openAI is "supposedly" a nonprofit and losing billions, I can understand the reasoning. Is this even a missed opportunity? What's stopping them from investing in some form of AI development now?
They're behind the curve, by a *LOT*. Even if the bottom doesn't fall out of the LLM craze, and even if someone doesn't come along with an entirely new approach that makes the LLM craze look like a 32 Ford Model T up against a 2024 Demon Dodge Challenger with an 850 HP motor, there's a missed opportunity in their rear view mirror that's worth considering, for the simple fact that they could have had at least a small chunk of the pie that NVIDIA is now baking in its bank accounts. It's a bad look for investors, it's a bad look for execs. And it will leave a foul odor around Intel when it comes to future investments. "They were so out of touch they missed the entire LLM thing."
Which, for the first time, almost makes me cheer for the LLMs to come up with something more successful than fucking up code or telling me I can't ask that question.
Re: (Score:1)
makes the LLM craze look like a 32 Ford Model T up against a 2024 Demon Dodge Challenger with an 850 HP motor
I hate all the pedants on this site as much as the next guy, so I'll let the LLM explain:
"What was the last year of production for the Ford Model T?"
ChatGPT:
The last year of production for the Ford Model T was 1927.
"What model of 1932 Ford was used for drag racing?"
ChatGPT:
The 1932 Ford Model B and its variants, including the Ford V8 (also known as the Deuce Coupe), were popular choices for drag racing.
Re: (Score:2)
makes the LLM craze look like a 32 Ford Model T up against a 2024 Demon Dodge Challenger with an 850 HP motor
I hate all the pedants on this site as much as the next guy, so I'll let the LLM explain:
"What was the last year of production for the Ford Model T?"
ChatGPT:
The last year of production for the Ford Model T was 1927.
"What model of 1932 Ford was used for drag racing?"
ChatGPT:
The 1932 Ford Model B and its variants, including the Ford V8 (also known as the Deuce Coupe), were popular choices for drag racing.
Well, crap. Got too many facts overlapping in my noggin'. Thank you for the correction.
Re: (Score:2)
You're a bit behind the times. The MI series is picking up steam.
Re: (Score:2)
Buying a 15% stake in an alleged not-for-profit wouldn't change that situation one bit. Their hardware and software stack would still suck.
Re: (Score:2)
Buying a 15% stake in an alleged not-for-profit wouldn't change that situation one bit. Their hardware and software stack would still suck.
You're viewing it from a tech standpoint, not an investor standpoint. They don't give a shit if they make decent product. They give a shit if they have "shareholder value." Shareholder value is rarely reflective of actual value. It's just the big boys playing with their imaginary tokens of future achievement.
Re: (Score:2)
LOL
OpenAI is at least nominally a not-for-profit. You really think owning a small chunk of OpenAI would push up their share price? Intel is flailing for reasons having nothing to do with AI.
Re: (Score:2)
LOL
OpenAI is at least nominally a not-for-profit. You really think owning a small chunk of OpenAI would push up their share price? Intel is flailing for reasons having nothing to do with AI.
Yes, 100% true. But it doesn't change the fact that if they were "in on it" they could ride the hype cycle to mediocre status in investor's eyes, instead of irrelevant yesterday junk, as it appears to be now. We're literally talking about completely different views of reality. Investors aren't technical people, usually. They only care about perception, because perception = $$$$. Who gives a shit about reality when there are dollars to be made?
Which seems like the motto of our entire country, or perhaps worl
Re: (Score:2)
I don't buy it. The companies making the most money in the AI gold rush are the ones selling shovels. Something OpenAI can't and isn't doing.
Re: (Score:2)
I don't buy it. The companies making the most money in the AI gold rush are the ones selling shovels. Something OpenAI can't and isn't doing.
But Intel does. Or could be giving the perception that they do, had they been "in on it."
Re: (Score:2)
I doubt it, the timeline doesn't add up. Firstly Intel was already entering the GPU space in this time with the Arc series already a year into development. 5 years later what did they have to show for it on release? Not much, they are still laughable in the space.
Why am I talking about GPUs? To demonstrate that hardware doesn't just appear overnight. This is especially relevant since NVIDIA launched their AI products in 2017. The Tesla V100 had tensor cores and CUDA was updated with AI relevant routines to
Re: (Score:2)
Intel's investors could have made billions riding the hype wave. I think that's all they missed out on. A 15% stake in OpenAI in return for developing and manufacturing chips AT COST seems like it was a good idea to walk away from. Those at-cost chips would be using up capacity in the fab pipeline, or require building out more capacity, which is really expensive and takes time.
Re: (Score:2)
Considering openAI is "supposedly" a nonprofit and losing billions, I can understand the reasoning.
What is this weird idea that a big startup is supposed to be profitable from day one? Amazon wasn't profitable until 2001, and it didn't start pulling in big numbers until 2017.
It's going to take time for the LLM market to grow out, for consumers to get around to subscribing, for businesses to figure out applications.
A big reason OpenAI is losing billions is they're dumping them into R&D, because they and their investors think there's a much bigger payoff at the end.
Is this even a missed opportunity? What's stopping them from investing in some form of AI development now?
OpenAI's valuation is $80B, Intels cu
Re: (Score:2)
All of those AI company (Score:3)
at least don't have to worry about melting Intel Gen 13 & 14 chips.
Not the angle of interest (Score:3, Interesting)
It is clear, from the various ways caches can leak information, the number of serious bugs in instructions, and the voltage defects of recent chips, that Intel's capacity to design and test is inadequate to handle modern chip complexity.
Instead of belly button fluff picking, Intel should look into the possibility of using neural nets (but not generative AI) for detecting where new instructions might have the largest impact, detecting where to concentrate testing, and detecting likely security risks.
(If they're sensible, they already do. But if they were sensible, they'd not be releasing defective chips intended for customers rich enough to afford good lawyers.)
Re: (Score:2)
How exactly is AI going to do what a whole department of engineers who specialize in chip design, cannot?
CPUs have been designed with computer tools for many many years, not directly by engineers. But I don't call it AI.
Re: (Score:3)
Generative AI hallucinates, but I already said I wasn't including that.
I also didn't say the NN should do any of the engineering, so engineers aren't relevant here.
I specified that it would look for areas to focus attention on. Engineers are great, but my own experience working for Intel showed that focus wasn't always there, things tended to be scattergun.
What you want to know is where to put those engineers, where you'll get the most return and close the most defects for the same amount of time. The engin
Re: (Score:2)
Re: (Score:1)
Kinda like missing bitcoin? (Score:5, Interesting)
I am reminded "it's not there yet" by IntelliJ daily. It uses AI to generate code. It suuuuucks. Now maybe IntelliJ did a bad implementation, but when I've played with copilot, I thought the same. It's fine for their pre-trained demos, but once you give it real problems...yeah...not so great. It will frequently suggest code that doesn't compile, but what's scary is it "looks" pretty reasonable. If I were to rely on it, it would slip in all sorts of bugs that look similar enough to the correct answer, but ends up being very wrong, like field mappings.
This is probably common knowledge for the slashdot crowd, but I always remind my non-technical friends, AI is NOT intelligent. It cannot solve a new problem. It can only determine the most probable answer based on past training input. It's just a fancier version of autocomplete. This is fundamentally limiting. If you want to use it for machine vision, that's great. I can potentially automate repetitive tasks. However, few knowledge workers are paid to repeat prior art. Most of us are paid to do something new...write a new contract, write a new application, write a new story. AI cannot do that...it can only mash up existing works and make predictions. There are many valid use cases for that...but it's not going to replace lawyers and engineers....any more than roombas replaced the housekeeping industry (at least that's my prediction).
But regardless, Intel is a chip company. I kind of would prefer they focus on making chips. We don't need them investing too heavily away from their core are aof expertise. We don't need Intel investing in every tech fad. They don't need a gaming studio. They don't need a streaming service. They didn't become a bitcoin broker, etc. We need Intel making good chips. I am not comfortable with TSMC being the only company that can make advanced chips and to my knowledge, it's down to just them and Intel. We need Intel making awesome chips and that duopoly to really compete and push each other.
Re: (Score:1)
The trick is to ride the hype in order to fund upgrades to real products, and then when the bottom of the hype-barrel falls out, you have real products to rely on.
Of course, timing trends and bubbles can be tricky. If it were easy, we'd all be golfing with W. Buffett.
Intel is failing, but not just in AI (Score:5, Insightful)
Intel is an example of what happens to a company when the entire management becomes complacent. When AMD started to recover financially in 2017 with the first generation Ryzen, Threadripper,and EPYC, Intel still had the attitude that AMD couldn't possibly catch up when it came to products. 2018, 2019, and yep, AMD had caught up in IPC. 2020, and the Zen3 based Ryzen were now seen as better for the most part, even though the overclocked "k" chips were still ahead due to clock speed advantage at the top end. Four years later, we have Intel processors degrading, with people not confident that Intel really did figure out the source of the problem.
Graphics. Market share is one thing, but just looking at products, AMD had at least gotten good enough and competitive enough where people SHOULD see Radeon as being worth considering if not for the marketing advantage that NVIDIA has. Intel on the other hand, wasn't even registering as having discrete graphics products for most people, and when it comes to performance, Intel high end competes with AMD low to mid range graphics.
So, the hype area of the past two years, AI. Again, marketing is a big driver to give NVIDIA the spotlight, AMD has very good products with MI300 series, but Intel...claims of having products that are good, but with very few people who have tested Intel AI stuff saying it's worth considering. Intel fell behind, not because it ignored AI, but because management just couldn't comprehend that Intel wasn't the driving force in the industry anymore.
Re: (Score:1)
Re: (Score:2)
The chart you linked to says $54.3B, down from a peak of $79.02B. Intel has never been a $100B net profit company.
Re: (Score:1)
Oh no, fell behind the times! (Score:2)
However will we survive as a species if AI fart videos take five seconds longer to render!?