I spent more than 20 hours looking at AI

news, so you can get this in 10 minutes.

This was the week a robo taxi hit a

child, a weekend coding project hit a

100,000 GitHub stars, and Wall Street

decided that spending $135 billion on AI

infrastructure was either genius or

insanity, depending on which earnings

call you listen to. Let's strip away the

noise and figure out what is a single

thread running through all of it.

Fundamentally, AI is no longer a thing

companies are talking about doing. It's

a thing that is happening all around us

in school zones, on your laptop, and

across factory floors. Let me walk you

through what this looked like this week.

Number one, the market cannot decide

what AI spending means. This is a really

big deal, and it gets to the lack of AI

fluency on Wall Street. Meta and

Microsoft reported earnings within hours

of each other this week. Both beat

expectations. Both announced they're

spending ungodly sums on AI

infrastructure. One stock jumped 10%,

the other dropped 11%. I think that

divergence is really instructive. Meta

posted $8.88 EPS against expectations of

$816 with revenue clearing $59.9

billion. More striking was their

forward-looking capex guidance 115 to

$135 billion up from 72 billion in 2025.

That is not a typo. Meta is nearly

doubling infra spend in a single year

and investors largely cheered. Why?

Because Meta can point to where the

money goes. Its recommendation

algorithms already run on AI, its

advertising engine, the actual business,

improves measurably with every model

generation. And when Zuckerberg spends

$130 billion, shareholders can trace a

line from that capital to revenue. At

least that is the story Zuck is telling.

Now I think what's interesting to me is

that he has also spent historically a

tremendous amount of money, billions and

billions on the metaverse which did not

have a clear revenue line item and

investors largely rewarded him because

the ad revenue model kept ticking over.

My sense is investors are happier with

the investment in AI versus the

metaverse because at least with AI, Zuck

can point to his core business model and

say, you know, if you have a wallet, AI

is going to make it possible for you to

spend on Meta. It's that easy. We'll

create the ads with AI. We'll optimize

with AI. We'll create the campaigns with

AI. We'll literally do everything. Just

give us your card. Investors are going

to love that. Microsoft told a very

different story. Azure grew 39%, which

sounds great until you remember it grew

40% last quarter. A single percentage

point seems like it shouldn't matter,

but in context with investors, it does.

Microsoft also disclosed that 45% of its

$625 billion commercial backlog is tied

up in OpenAI commitments. Nearly half of

Microsoft's contracted future revenue

thus depends on a company it does not

control. Something that Deutschbank

described as a makeorb breakak moment

with deferred commitments coming due for

Microsoft. The market's message to

Microsoft was quite clear. Spending on

AI is fine. Spending on AI you don't own

is a different story. And that

distinction is going to define

enterprise AI strategy for the next

couple of years. Companies face a choice

between building proprietary AI

capabilities expensive, slow but

defensible, or renting frontier models

from labs where the economics remain

unproven. Microsoft chose the latter and

is now structurally exposed to OpenAI's

execution. Meta chose the former and is

being rewarded by the street for it.

Even as Meta's Reality Labs is burning

$6 billion a quarter on a metaverse bet

that is yet to pay off. Yes, Zuck isn't

quite done with the metaverse. The

lesson isn't that one approach is right

here. It's that markets have started

pricing AI strategies differently

depending on who controls the underlying

asset. I think this is instructive as we

look to Apple's deal with Google because

Apple is choosing not to invest in

Anthropic because Anthropic was

reportedly nickel and dimming them over

the pricing for powering Siri with

Claude and Apple went with a longtime

partner in Google because Google was

willing to more expansively partner with

them in supplying Gemini to run Apple.

This is a case where Apple needs to

leaprog and investors may make an

alternate and opposite decision versus

their punishment of Microsoft because if

you think about it, I'm not sure that

the street believes that Apple can jump

and leapfrog with AI on their own. And I

think they're looking for an AI deal to

enable Apple to compete and to support

the continued dominance of the iPhone.

And so if Apple comes out and starts to

push and say, "We're launching with a

Gemini powered Siri. It's going to be

incredible. I think that investors in

both Google and Apple are going to see

that as a win. And so the other piece of

this is that even though you have to be

careful about owning your assets, if

you're going to rent them like Apple is

doing for Google, make sure you rent

them from a provider where the street is

like, "Yes, this is going to work out."

Because to be honest, Google is the only

hyperscaler in the mix where it's an

affordable bet. They make so much money

off of search that all of AI is an

affordable bet for them. and there's no

real threat to the business. And isn't

that ironic because a year ago we were

looking at search and saying Google's

going to be disrupted and here we are

recognizing that Google came back. All

right, let's talk about the car crash.

On the morning of January 23rd, a Whimo

robo taxi struck a child in Santa Monica

during drop off hours. The NHTSA has

opened up a preliminary evaluation into

the incident and this is not Whimo's

only federal investigation this week.

The NTSB had already opened a separate

inquiry into the company's Austin

operations where Whimo vehicles have

illegally passed stopped school buses at

least 19 times since the school year

began this year. Whimo's response was

defensive and kind of missed the point

because autonomous vehicles aren't

really competing against the median

human driver. Whimo tried to come out

and say, "We have a peer-reviewed model

suggesting a human driver would have

struck this child harder and the Whimo

vehicle struck this child softer. You

struck the child. This is tonedeaf. This

is inappropriate and this is wrong. Now,

Whimo has logged tens of millions of

autonomous miles. By most statistical

measures, its vehicles are safer than

human operated cars. But data doesn't

matter when there's a narrative. And

there is no narrative as compelling as

the risk to safety from an autonomous

and unaccountable vehicle striking a

child. If if we are going to have any

kind of acceptance for autonomous

vehicles, we need to do better at taking

accountability when things go wrong. We

can't just say, "Oh, it's more rare. The

humans are worse." Like we have to

actually have accountability. And that

is one of the underlying fears about

autonomous vehicles and autonomous

robots in general is that who's going to

be accountable. We need to answer that

question directly or we are not going to

get societal acceptance. Meanwhile,

Nvidia is not done investing in

companies. Big surprise. Nvidia invested

$2 billion in Coreweee this week at

$87.20 a share. The money will

accelerate Cororee's buildout of AI

optimized data centers with a target of

5 gawatt of capacity by 2030. For

context, that's roughly the power

consumption of a middlesized city.

Jensen Huang called it the largest

infrastructure buildout in human history

and he is not wrong. The capital

continues to flow into AI compute and I

feel like I have a new story to tell you

here every week and it all adds up to

more gigawatts of data centers. And the

deals that are being done now reveal a

maturing infrastructure market.

Microsoft signed another big deal this

year as well. They signed a $750 million

three-year deal with Perplexity, giving

the AI search company access to open AI,

anthropic and XAI models through

Microsoft's founding foundry. Meanwhile,

Perplexity was careful to note that AWS

remains its primary cloud provider. a

hedge against over reliance on a single

hyperscaler by an AI startup. Both the

Nvidia deal and the Microsoft Perplexity

deal reveal that we are moving past the

first wave of AI spending which went to

model development and we are moving into

a second wave where we are figuring out

physical and logical layers that make

model deployment possible at scale. So

companies like Coreweave and platforms

like Microsoft Foundry are becoming

essentially plumbing for the AI economy

and are getting involved in major deals

as a result. The ultimate question,

which is one nobody has an answer to, is

whether the infrastructure buildout is

sized correctly. Is 5 gawatt correct

because demand curves will continue to

explode? Is 5 gawatts incorrect because

it's too small a bet? Is it incorrect

because it's too big a bet? We're not

going to know until we get close. And

none of the hyperscalers want to risk

leaving cash on the sidelines. The next

one I don't want you to miss is that

Tesla is now an AI company that sells

cars. Tesla's earnings call this week

wasn't really about cars. It was about

robots, AI investments, and dismantling

product lines that no longer fit the

company's direction. Tesla is

discontinuing the Model S and the Model

X. The factory lines that produce them

are going to convert to Optimus robot

production with a target of a million

robots a year at their Fremont facility.

Optimus production is slated to get

started at the end of this year in 2026.

Tesla also announced a $2 billion

investment in Elon Musk's AI lab, XAI.

Grock is now deployed across the Tesla

vehicle fleet. The company disclosed

more than a million active full

self-driving subscriptions with robo

taxi service expanding to seven new

metro areas. Dallas, Houston, Phoenix,

Miami, Orlando, Tampa, and Las Vegas,

all in the first half of this year. They

do move fast. When you add all of this

together, you get a company that has

decided its future lies in AI and

robotics, not premium sedans. The Model

S and Model X were always low volume

products that consumed a lot of factory

capacity. And I think that the company

has figured out they have higher margin

opportunities in robotics. So killing

those frees resources for bets that

Tesla considers to be more important.

Tesla's $20 billion capital expenditure

plan signals the scale of these bets.

It's building out AI training

infrastructure, robotic manufacturing

capacity, and autonomous vehicle

deployment networks simultaneously. This

is a portfolio approach to a tech

transition that most companies are just

barely beginning to navigate. We'll see

whether Tesla can execute. But one thing

I will say is that they have never been

afraid of jumping five years into the

future and making a bet on where they

think the market is going. And it is

significant to me that Tesla thinks they

would rather be in the robotics industry

at scale than the car industry at scale.

That's something worth thinking about.

Meanwhile, Hyperscaler fundraising isn't

done. Anthropic closed a funding round

this week valuing the company at $350

billion, nearly double the September

valuation. That's how fast we're going.

The round was led by Singapore's GIC in

Koju and the potential to reach 10 to 15

billion more in funding if Microsoft and

Nvidia choose to hop in on the tail end

of the round. This is a third of a

trillion dollars. This is half the

market cap of Walmart and the valuation

reflects several converging factors.

First, anthropics cloud models have

gained enormous traction at the

enterprise level and they are

particularly gaining traction in spaces

where safety and reliability matter

which tends to lead to sticky revenue.

The constitutional AI approach is

clearly differentiated in the market

from competitors and leads to trust. And

finally, the funding market for Frontier

AI Labs has become intensely competitive

and investors continue to be willing to

pay really steep premiums for exposure

to what they believe will be a winner

take most. Ultimately, Anthropic is

betting that safety focused development

will produce more capable and deployable

systems than approaches that optimize

purely for performance. And that is not

to say they're not leaning in on

performance. Anthropic is openly and

publicly saying they are building AI to

build AI. They are working on AI models

internally that are going to enable them

to scale their AI model building efforts

even more. If that bet is correct, the

company's valuation may well look cheap

in retrospect. And and if they end up

not being able to deploy and scale at

the pace they're looking to accomplish,

the third of a trillion dollars is going

to look like a big monument to hype. I

got to say all the signals are toward

the former. All the signals are toward

Anthropic continuing to scale revenue

extremely quickly and that valuation may

indeed look cheap in retrospect. Last

but not least, Peter Steinberger hacked

together a weekend project a couple of

months ago. It has crossed 100,000

GitHub stars and it's named Open Claw.

Formerly Clawbot, then Moltbot after

Anthropic raised trademark concerns.

This is basically an AI agent that

connects to your messaging platform of

choice, WhatsApp, Slack, Discord,

Telegram, iMessage, and it lives inside

your hardware and enables autonomous

task execution across your digital life,

email management, calendar updates, file

operations, web browsing, any kind of

cross-platform automation. The the

simplicity is is an incredible appeal

for this, right? Instead of using AI

through a chat interface, which feels

cumbersome, users can deploy an agent

where you just text it and it does

stuff. And the risks are equally

obvious. Open claw requires broad system

access to function at all. And so

effectively it becomes a lethal trifecta

of private data access, exposure to

untrusted content, and the ability to

take action. Like when you put all of

those together, that's what makes open

claw so powerful. But it's also what

makes it so risky. That is not stopping

any of the members of the community from

deploying it. And it reminds me a little

bit of where we were decades ago in the

age of music piracy when we had startups

that said music wants to be free and

despite all of the regulatory issues,

they just kept doing it. Despite all of

the concerns around safety this time,

humans are just so excited about AI

agents being autonomous, they're just

going to keep doing. The democratization

of agent AI means more people can build

and deploy autonomous systems. And it

means we're going to need to think more

and more and more about security on the

internet as we have autonomous agents

operating on the net like crazy. If it's

at 100,000 GitHub stars now, it may well

hit a million gets GitHub stars within a

couple of months. The growth is

absolutely hyper exponential. Look, the

throughine connecting all these stories

isn't really tech. It's commitment.

Microsoft is committed to OpenAI at a

scale that shapes its entire financial

structure. Meta is committed to

infrastructure spending in a way that

assumes AI will transform the business.

Whimo has decided they're committed to

autonomous deployment and are willing to

brave the regulatory scrutiny and are

willing to be frankly quite defensive

about a human tragedy. Tesla is

committed to an incredible pivot into

robotics. Anthropic is committed to

scaling their enterprise revenue and is

earning a valuation that reflects that.

None of these commitments are cheap.

They're very expensive. They are not

reversible at a low cost. the the phase

of AI where companies could talk about

potential is ending. What remains is

execution against bets that companies

are too far in on to walk back. We are

in a phase where you have put your chips

on the table and we are all going to see

how AI turns out. There's no going back