HPC News Bytes – 20260302

Welcome to HPC NewsBites, a weekly show about important news in the world of supercomputing,

AI, and other advanced technologies.

Hi everyone, welcome to HPC NewsBites.

I'm Doug Black with a cold of Inside HPC, and with me is Sheehan Khan of OrionX.net.

The biggest news of the week, of course, is the military action taken against Iran and

Shaheen and I are already planning at HPC podcast episodes focused on advanced technology-driven

warfare.

But for today, a big story is the battle between Anthropic and the Department of Defense over

the use of Anthropic AI.

In a nutshell, Anthropic signed a $200 million deal with DOD last July and has recently

been embroiled in a battle over how its AI models are used, specifically Anthropic

objects to mass domestic surveillance and autonomous weapons use cases.

The Defense Department says mass domestic surveillance will not be pursued because it's

already been declared illegal by Congress and that rules are in place regarding autonomous

weapons.

A DOD spokesperson added Anthropic doesn't make the rules.

Questions have been raised about the timing of Anthropics demands, why raise them now

instead of before they signed the DOD contract, also do vendors establish use case rules.

This also raises larger issues such as placing restrictions on our use of AI when and if

we know our adversaries are doing those very things.

Of course, no one favors mass domestic surveillance, but if our adversaries are developing and

using autonomous weapons, should we prohibit ourselves from doing the same?

In addition to the customer vendor discussions and negotiations, there are larger and very

valid issues here.

The big one is the impact of technology on laws.

This has two major dimensions.

The first is digitization, which adds information to the equation.

Laws that were designed to govern things need to evolve or otherwise get upgraded or rewritten

to include information.

Bits versus atoms, so to say.

The second is the consumerization of technology.

Some decades ago, advanced technology would start in government labs and the military,

and it would then get sanitized and adopted by the enterprise, and then it would trickle

down to consumers.

For the past couple of decades, the flow has been getting reversed.

More and more, the consumer uses advanced technologies first and companies and governments

struggle to keep up.

So laws have evolved, but many of them were originally written before they were phones

or pocket cameras.

On privacy, it's been the fourth amendment in the United States that provides constitutional

protections for individuals by safeguarding their privacy while enabling the state to

do its part to protect people.

Over the years, it has been interpreted in broader ways, notably by the recognition that

its focus is the protection of people and therefore their digital presence as well,

not just physical things.

On autonomous weapon systems, that's a whole new concept.

There are some international laws that hold a state responsible and accountable.

There's a NATO framework for responsible use, and there are various non-binding declarations

and also calls for treaties.

In the U.S., there's a DOD directive that allows the development of autonomous weapons,

but requires rigorous testing and human judgment.

But there are no specific dedicated federal laws passed by Congress that regulate the development

and use of autonomous weapons in the United States.

That's really where the discussion belongs.

But there is global competition towards these weapons as you alluded to, and that makes

it very, very difficult for a country to wait and deliberate while others are racing

ahead and may not even follow international laws or honor their treaties anyway.

This kind of thought process has led many to advocate for global regulation of AI that

is modeled after how nuclear weapons are regulated.

On the product side, it is interesting that for technology as complex as AI, there are

in fact so many choices in the U.S. and globally.

And finally, whether or not a product is able to be compliant with policies or directives

or laws or what have you, is another aspect, and then there's may not sign up to that,

and especially when much of AI remains probabilistic versus deterministic.

Intel's road to an AI chip has been rocky from the start.

The latest Go-No-Go occurrence was the scutling of a reported $1.6 billion acquisition

by Intel of AI chip startup, Sam Bonova.

But by no means has the relationship between the companies fallen apart.

The latest development is that Sam Bonova has agreed to adopt Intel's server chips and

graphics cards in a multi-year collaboration.

The companies announced this on Tuesday.

Intel is also taking part in a $350 million funding round, adding to its earlier investment

in Sam Bonova in 2019.

We should also note that Intel CEO Lippu-Tam, known around the company as LBT, also

is chairman of Sam Bonova.

So there are multiple ties between the companies.

To be sure, Intel is developing its own AI GPU.

And as we have discussed here, they really cannot sit it out.

While the Sam Bonova acquisition could have helped kickstart Intel into the AI fray, the

partnership between the two companies seems more focused on strengthening Intel's CPU

business within AI.

Sam Bonova counts hugging face, meta, and major AI labs as customers.

They say their new SN50 chip delivers higher performance than the Blackwell 200 system

for a comparable price, and that up to 256 SN50s can be connected.

Be afraid of circular financing in AI, but AI frenemies or co-option seems like a common

thing too.

Intel invests in Sam Bonova, while Nvidia took a stake in Intel, value that $5 billion,

which was a contributing factor in Intel's share prices going up.

Intel shares were up 75% over the past year.

And of course, the US government took a 9.9% stake in Intel worth nearly $9 billion,

which may have shown the way to Nvidia.

GTC is the week after next.

And we're expecting the same Sardine-Can crowds at the San Jose Convention Center that we've

seen in recent years, along with a huge crowd at a professional sports arena for Jensen

Wong's keynote speech.

For electricity, general excitement, and avid interest, I don't think I've ever seen

anything quite like GTC unless it was at Comdex if people remember that show back in the

80s.

I'm not trying to flatter Nvidia, just stating a fact as I see it.

Moving around on the conference floor is a battle as is getting a hotel room, a noob

room, a bottle of water, everything.

The tech publication Tom's Guide ran an article late last week on what may be coming up

at GTC.

This includes the possibility of a new AI chip, something hinted at by Jensen.

And Tom's Guide says, quote, we can expect a major reveal of a new inference-focused chip

possibly related to the Feynman architecture designed for faster AI tasks.

This wouldn't be a typical GPU since it's built for a genetic AI.

Jean, I have to imagine Nvidia's major competitors are already wincing in advance

over what Nvidia may announce.

Among the company's hallmarks is acceleration of the cadence of new chip releases putting

immense pressure on the rest of the industry.

Yeah, for sure.

I'm really looking forward to it.

And as we look forward to GTC, this week we have the big mobile world congress in Barcelona

Spain.

And next week there's Embedded World in Nuremberg, Germany.

Mobile World Congress expects over 100,000 attendees and about 3,000 exhibitors, focusing

on advanced communications and mobile innovation in general.

And there will be a lot of AI there too.

The conference has themed the IQ era and what we should expect is AI's expansion outside

of the cloud towards edge AI, agent AI, physical AI and ambient AI.

There will be robots and smart things and examples of autonomous systems built directly into

hardware and networks.

Really, the return of the Internet of Things this time propelled with AI.

We talked a little bit about network intelligence after Nvidia's GTC in Washington DC last

year and the work they did with partners towards AI ran radio access networks.

That will be on display as we will 6G build on AI native platforms, complete with its

own network autonomy, AI enabled optimizations and telco reasoning models.

In many countries telcos are a suitable vehicle and partner for AI enabled communications,

for military applications and sovereign AI deployments.

So we will see that as well.

Alright, that's it for this episode, thank you all for being with us.