AI Week: Billion‑Dollar Deals and Policy Milestones
Key Points
- Nvidia and Intel announced a $5 billion partnership that gives Intel access to Nvidia’s AI chip stack, paving the way for powerful local large‑language models on consumer laptops.
- Microsoft committed an additional $4 billion to build two AI‑focused data centers in Wisconsin, underscoring its continued expansion of U.S. compute capacity despite earlier market rumors.
- OpenAI released a new model, adding to the rapid rollout of next‑generation generative AI capabilities.
- A new pair of smart glasses can interpret muscle‑activity signals for control but deliberately lack Wi‑Fi connectivity, highlighting selective sensor integration.
- For the first time, a national government publicly declared it is “ready for AI,” marking a significant policy shift toward embracing the technology.
Sections
- Nvidia‑Intel $5B AI Partnership - The excerpt highlights a $5 billion deal in which Nvidia invests in Intel to provide Intel access to Nvidia’s AI chip and firmware stack, enabling broader on‑device AI adoption while briefly touching on other headline AI developments.
- AI Race Fuels Browser and Notion Wars - The speaker highlights Chrome's push to integrate seamless AI features to fend off emerging smart browsers, while noting Notion AI's launch of agents that automate and centralize startup workflows, signaling intensifying competition in both browser and productivity tool markets.
- GPT‑5 CodeX: Efficiency, Competition Victory - The speaker contrasts cloud‑code platforms with GPT‑centric tools, then highlights GPT‑5 CodeX’s dynamic token allocation that boosts simple‑task efficiency by up to 94% while excelling on complex reasoning, and notes its unprecedented win over top human programmers in a collegiate coding competition.
- EU AI Market, Regulation, and Ireland Hub - The speaker highlights Ireland as a gateway for U.S. tech giants, emphasizes the EU’s massive AI market and its regulatory sandbox as a growth avenue—especially versus China—and notes how compliance‑first strategies can enable both EU and U.S. startups to penetrate highly regulated sectors like healthcare.
Full Transcript
# AI Week: Billion‑Dollar Deals and Policy Milestones **Source:** [https://www.youtube.com/watch?v=ImOo1BxAG7g](https://www.youtube.com/watch?v=ImOo1BxAG7g) **Duration:** 00:16:06 ## Summary - Nvidia and Intel announced a $5 billion partnership that gives Intel access to Nvidia’s AI chip stack, paving the way for powerful local large‑language models on consumer laptops. - Microsoft committed an additional $4 billion to build two AI‑focused data centers in Wisconsin, underscoring its continued expansion of U.S. compute capacity despite earlier market rumors. - OpenAI released a new model, adding to the rapid rollout of next‑generation generative AI capabilities. - A new pair of smart glasses can interpret muscle‑activity signals for control but deliberately lack Wi‑Fi connectivity, highlighting selective sensor integration. - For the first time, a national government publicly declared it is “ready for AI,” marking a significant policy shift toward embracing the technology. ## Sections - [00:00:00](https://www.youtube.com/watch?v=ImOo1BxAG7g&t=0s) **Nvidia‑Intel $5B AI Partnership** - The excerpt highlights a $5 billion deal in which Nvidia invests in Intel to provide Intel access to Nvidia’s AI chip and firmware stack, enabling broader on‑device AI adoption while briefly touching on other headline AI developments. - [00:04:55](https://www.youtube.com/watch?v=ImOo1BxAG7g&t=295s) **AI Race Fuels Browser and Notion Wars** - The speaker highlights Chrome's push to integrate seamless AI features to fend off emerging smart browsers, while noting Notion AI's launch of agents that automate and centralize startup workflows, signaling intensifying competition in both browser and productivity tool markets. - [00:10:44](https://www.youtube.com/watch?v=ImOo1BxAG7g&t=644s) **GPT‑5 CodeX: Efficiency, Competition Victory** - The speaker contrasts cloud‑code platforms with GPT‑centric tools, then highlights GPT‑5 CodeX’s dynamic token allocation that boosts simple‑task efficiency by up to 94% while excelling on complex reasoning, and notes its unprecedented win over top human programmers in a collegiate coding competition. - [00:14:32](https://www.youtube.com/watch?v=ImOo1BxAG7g&t=872s) **EU AI Market, Regulation, and Ireland Hub** - The speaker highlights Ireland as a gateway for U.S. tech giants, emphasizes the EU’s massive AI market and its regulatory sandbox as a growth avenue—especially versus China—and notes how compliance‑first strategies can enable both EU and U.S. startups to penetrate highly regulated sectors like healthcare. ## Full Transcript
This week in AI news had everything. We
had historic enemies becoming allies in
a $5 billion deal. The world's most used
browser got a brain transplant. We had a
new open AI model. We had glasses that
can read muscle signals but not Wi-Fi
signals. And perhaps most interestingly,
we had the first government that was
willing to say we're ready for AI in a
meaningful way. Let's get to the news
and why it mattered. Eight stories for
you. We're going to jump in with number
one, Nvidia and Intel inking a new deal.
The deal is worth $5 billion.
Specifically, Nvidia is just investing
in common stock with Intel at 23 bucks a
share, which promptly popped. So, I'm
sure Jensen took a few billion off to
the bank on that one. But the the real
reason this matters is that Intel will
get access to Nvidia's technical stack,
which allows them to bring Nvidia's
chipsets and their firmware into the
laptop supply chain and the rest of
Intel's products. What that means is a
lot more AI availability over the medium
term for consumers and for businesses.
Right now everything is running through
the cloud. We have a billion people
accessing AI through the cloud. I don't
think that's the way it will be forever.
Andre Carpathy actually called this out
really effectively. I thought when he
commented that we are going to see a
future where everyone just has
intelligence too cheap to meter that
lives locally on their device and they
can get other intelligence if they want
it. To realize that you need a chipset
change. You need a C change in what
chipsets are available and this deal is
the deal that unlocks that future. Intel
is so ubiquitous across chipsets and
consumer laptops. This deal is what
enables much more powerful local LLMs to
run on consumer laptops, which enables a
future where you will just have a local
LL running on your machine and it's
local intelligence and it knows your
machine really well. And yes, it can
probably browse the web and do other
stuff. And yes, you may call in cloud
AIs as well, but it's just there to help
you run your machine and run your day.
The idea of a personal ondevice
assistant was made real this week
through this deal. Story number two,
Microsoft is investing in its Wisconsin
AI data center to the tune of an
additional 4 billion and they are
planning on bringing two data centers
online there. So, there's a data center
uh coming online in 2027 and another one
in early 2026. And as you would expect,
they're housing a ton of NVIDIA GPUs and
they're basically looking to establish
more compute capability for what they
anticipate to be surging business demand
specifically in the United States. This
contradicts the narrative that was
popular early in 2025 that Microsoft was
scaling back its AI investments because
it was feeling cautious. I don't know if
you remember the brief market turmoil
when Deepseek was launched, but it led
to rumors circulating that Microsoft was
pulling back on its AI data center
investments. This investment reminds us
that the long-term trajectory of
investing ahead of business demand in AI
remains intact. Story number three,
Google Chrome got a Gemini integration
and it has a bunch of features. The one
that popped out to me is the idea that
you can look at your email, type it into
Gemini that you want to get dinner with
someone and you want some ingredients
and Gemini can go order from Instacart
for you and get your ingredients. That
stood out to me for two reasons. One, it
is a classic demonstration of Aenta
capability, particularly because it's
tied to Instacart, which is a highly
predictable interface in which you can
partner with the Instacart team on to
make sure you don't have mistakes. So,
it's a little bit like driving on the
closed course. They always tell you the
stunt drivers drive on for ads. Yes, you
can do this, but drive it on Instacart
because that's what that's where this
works. I think the reality is aentic
browsers are going to feel like a
gimmick until we find stable readrfaces
that actually work. Now, I am the first
person to say Gemini has footprint,
Chrome has footprint. The fact that they
felt comfortable rolling this out to the
whole Chrome user base says something
about how committed the team is to
shipping at Google. I am sure it will
get better. But init and I should add
one caveat. It is rolling out to the
whole world, but it's starting with US
English users and global expansion will
follow. So if you're in Europe, sit
tight. It's coming. It integrates with
calendar, YouTube, maps. That's really
where the power is going to be. If
you're looking down the road at what
could this be, not what is this now,
which I think is a great way to look at
a lot of these AI stories, the
trajectory for agentic browsers is how
can they establish readwrite interfaces
to the places that matter. Not the
things that feel like made up. Because I
got to tell you, when was the last time
I wrote a friend and was like, let's
have dinner and went to Instacart? Not
very frequently, but when was the last
time I went to Google Maps? All the
time. When was the last time I went to
YouTube? All the time. When was the last
time I went to Calendar? all the time.
So those kinds of concrete use cases, if
they can make them feel seamless, have
the potential to really build
stickiness. And it's also a move that
indicates that just like the rest of
Google, the Chrome team is aware that
they need to play catch-up here. There
are smart browsers on the market that
are already out there already delivering
aic readr experiences. Chrome doesn't
want to lose market share to those other
browsers. So watch this space. I think
it's going to be a really, really
interesting browser war heating up here.
For those of you who remember the really
old browser wars in the early 2000s,
this is like the 2.0 version. It's going
to be really fun to watch. Next story,
Notion AI launches agents. This is part
of their so-called 3.0 launch where they
launched the idea at the Make It With
Notion event. And Notion's value
proposition is very clear. They
basically want to centralize all of the
tool stack that goes with running a
modern startup and put it into one tool.
So, you don't have to worry about CRM as
much. You don't have to worry about
email as much. You don't have to worry
about publishing a website. Everything
comes in under notion. Agents make all
of that easier in that product strategy
because agents enable you to take the
friction out of all the work you need to
do. You don't need to learn Notion to do
stuff in Notion anymore. If agents can
create and update pages, agents can
update databases. Agents can search
connected tools. You get the idea. So,
this features editable profile pages for
customizing your agents behavior. And it
enables you to store a dynamic memory or
user preference. I will be really
curious to see how people start to
actually use this in the wild. To me,
this feels like a power user feature for
notion where if you're in the notion
ecosystem, this is going to be a huge
help. In fact, I had someone email me
today and say, I am going to have to
double down on my notion investment
because this is going to really really
help me and I can cut down on other
tools, which is exactly what the notion
team wants. It integrates with Slack, it
integrates with email, with Google
Drive, with Zenrest, with GitHub. You
you get the idea. You can also plug in
MCPs to it. Now, if you think about it,
this is a another case where there's an
arms race going on because notion is
playing in a multifront battle between
the sort of tool sets of the world for
B2B SAS like the CRM teams, the email
teams, etc. And they have like
competition across all of those fronts.
But they're also now at a point where
they're playing with people who are
using Chad GPT projects, people who are
using claude projects. So the big model
makers and Notion has picked the goal of
being a hub which has made them a target
for all of those companies. It's a
really tricky place to be and it's not
surprising to me that notion has seen
their gross revenue uh profitability
decline. I should have said that better.
their gross profitability declined from
a 90% profit margin to an 80% profit
margin over the last year because
they've had to invest so much in AI
tooling to stay competitive. So watch
this space. I'm really curious to see
how power users actually level up with
notion using AI agents. Next, Ray-B band
display glasses. It was not a good
launch, people. So they launched it with
an aggressive price point at $799. The
idea is that this is the first fully
functional pair of glasses. I got to
say, I've heard it before, but this one
looks like the best yet, right? It has a
600 by 600 pixel fullcolor display. It
refreshes at 90 Hz. It is bright enough
that it looks like you can read it in
daylight, but it does not have light
leakage, so other people can't see what
you're seeing. And it looks like it's
small enough that it doesn't look weird
to wear. The drawback, as we all saw on
the demo, is that it is a little bit
sensitive. And so it seems like the team
had some Wi-Fi issues in the room at the
time. They're trying to say that this
was all a function of dev servers and
the Wi-Fi. My view is we will see how
reliable the AI is. Once we get people
out in the wild who are able to wear
this and actually interact while they're
at the park or, you know, while they're
taking the kids for a walk. My suspicion
is that some of the issues that came up
in the demo are actually not just Wi-Fi
issues. It sounded to me when I watched
the demo like the AI actually heard
queries and got mixed up. Put a pin in
this one. There's an ongoing debate,
ongoing battle between Apple and Meta
over who is going to be able to convince
people to get into wearables for
glasses. Nobody has really gotten
something that everybody's wearing on
the street yet. And that is the bar,
right? That is the iPhone bar. And I
think that to get there, you have to
make something that feels not just
unobtrusive, but actively cool. And that
was the bar that the iPhone was able to
cross to drive adoption. And that's
something that I know Meta wants to get
to with their partnership with Rayban. I
don't know that they got there with this
one, but they probably got closer than
anyone has before. Next, we have
OpenAI's GPT5 codeex release. This is
the new model from GPT5 from the the
OpenAI family. This model is designed
specifically to help with the
nonlinearity of coding tasks and it
follows the ongoing trajectory of the
codeex team. It's shipping really,
really quickly. One of the things I've
been most impressed by is how quickly
the codeex team has been shipping. So
much of what you buy in AI is not the
capability today. It is the trajectory
of the team building the product. And
Codeex has been outshipping cloud code
lately. I noticed that a lot of claude's
recent releases in the last few weeks
have been productivity tool oriented but
not cloud code oriented. And Codex has
been catching up and according to some
developers surpassing cloud code. Now my
view is that it is a little bit like
apples and oranges and it's going to
depend on your developer use case what
matters. Codeex seems to be positioned
more for larger teams that have more PRs
hitting GitHub at the same time and you
need something to sort of pre-review.
But on the other hand, Anthropic claims
that's what their cloud code does
internally. And so I don't think it's
that cloud code can't do that. It's just
that there are people who have got
stacks that are oriented around cloud
code, which certainly can scale from the
individual terminal on up. And then
there are folks who are used to the GPT
ecosystem and they just want Chad GPT
all the way down. To me, it is almost
feeling like the Windows versus Mac wars
of the '9s where people have a brand
affiliation and that drives a lot of the
downstream product choices. Anyway, from
a capability perspective, this new GPT5
codeex model adjusts thinking time based
on task complexity, which makes it much
much more efficient, up to 94% more
efficient on simple tasks while
allocating a ton more tokens to very
complex reasoning tasks. What that means
to me is that they got better at parsing
prompts, which is a huge deal. The other
thing to notice about GPT5 codecs is
that this is the model that later won a
collegiate coding competition with a top
score better than the best human in the
competition, which is a new watermark
for AI because previous high water marks
for AI, AI has been slightly below or
equal to the best humans in a particular
technical competition. As far as I know,
this is the first time an LLM has been
better than a human unambiguously at a
top technical competition. So, I am
really curious to see where we go from
here. Couple more stories for you to
round out our day. OpenAI and Microsoft
finally cut a deal. If you recall,
they've been debating back and forth
about their previous deal that they made
and there have been questions around the
equity stake that would be taken. And
there have been questions around that
definition of AGI clause that was
famously there where OpenAI could void a
lot of the terms with Microsoft if they
declared AGI. Fundamentally, what was
going on is that Open AI had grown up
and Microsoft needed to readjust their
relationship with Open AI. Well, they
have now. uh the nonprofit arm of
OpenAI, I you I can't believe they're
still a nonprofit, but that's a separate
story, received a hundred billion dollar
equity stake in the for-profit OpenAI
company as part of a comprehensive
reorganization that Microsoft approved.
Open AAI will then share only 8% of its
revenue with Microsoft and other
commercial partners by 2030, down from
the current 20. So essentially, Open AI
successfully made the case to Microsoft
that they'd grown up enough, that they'd
scaled the business enough, that 8% of
their projected revenue through 2030 was
bigger than 20% would have been when
they were at a smaller scale and a
smaller run rate. What that says is that
everybody in the room making that deal
thinks the demand picture for AI over
the next 5 years is even bigger than
they thought a couple of years ago to
the point where they are willing to take
significant dilution on revenue share.
Last story, Ireland EU AI act. It might
sound boring, but it's not. Ireland
designated 15 different authorities to
enforce the EU AI Act, including their
central bank, a data protection
commission, a health products regulatory
agency. And they established a single
point of contact within the Department
of Enterprise, Trade and Employment to
coordinate with EU bodies, member
states, stakeholders, etc. You care
about this because they're actually
putting in place the regulatory
environment and framework to enforce the
EU AI act. This is going to create a
national AI office in Ireland by August
2nd of 2026, which will centralize
coordination, host regulatory sandboxes,
drive AI innovation, etc. This will
enable controlled AI testing under
compliance safeguards inside the EU,
which I know EU entrepreneurs are
excited about because one of the things
they've sort of complained about is that
the new EU AI act is going to hamper
European AI innovation. Other companies
like American companies care about this
because they are doiciled in Ireland.
And so where Ireland goes, there goes
the headquarters of many of the most
prominent US mega corps. There are a
number that have offices there. Amazon
has offices there. Apple has offices
there. And there are others as well. You
want to be paying attention to this
because the EU 500 million member market
is probably the second biggest market
for US AI companies after the US itself.
With the divergence with China in terms
of both chipsets where China is saying
no to Nvidia chipsets now and now also
sort of a divergence around purchasing
patterns and consumer patterns and the
divergence in internet behavior. US
technology companies selling AI are not
going to get very far in China right
now. There are too many open source
models. It's a different ecosystem, but
they will have a chance to market in the
EU. And conversely, EU companies that
come up and are able to use this
regulatory authority sandbox get better.
Maybe Mistl comes to mind are going to
have an option to bring a compliance
first perspective with AI to the US if
they want. And it's it's still valuable
to have a compliance first perspective
in the US. There are a lot of highly
regulated industries in the US where AI
is being slowwalked because no one can
explain how to do it safely. Healthcare
comes to mind. How do you make sure that
you roll out across the gigantic
healthcare industry HIPPA compliant AI
workflows? There's a whole forest of US
startups on that, but there's no reason
why EU startups can't compete there,
too. So, that's the stories for the
week. It's been a busy week. We will see
what next week's holds. Cheers.