Meta is about to spend $130 billion this

year, mostly towards data centers. And

we know that Meta has been far from

relevant in comparison to other frontier

labs. And yet, Meta is doubling down to

spend up to $600 billion by year 2028,

which is nearly 2% of the US GDP in a

single year. And currently, Meta is

finishing up their $30 billion data

center in Louisiana. That brings up

their number closer to 30 data centers

in the US alone. So, why would Meta

continue to dump all this money towards

AI? Welcome to Kilobytes Code, where

every second counts. Quick shout out to

Nvidia. More on them later. We don't

necessarily think about Meta when we

talk about AI, but Meta's contribution

to AI actually goes further back than

OpenAI and Anthropic where in 2013 Meta

formed a group called fair that released

important libraries like PyTorch,

different versions of Llama from Llama 1

to Llama 4, as well as different

expressions of GAN and many more. As you

can see, unlike what we currently think

about Meta right now, Meta used to be at

the forefront of AI research and

innovation until closed proprietary labs

like OpenAI and Anthropic started to

dominate in performance. And what's

worse is that Meta also fell behind in

open models against Chinese frontier

labs like Deepseek, Alibaba, Moonshot,

and Miniax. as well as more recently

Nvidia actually started to overtake in

open model contribution according to

hugging face given their contribution

towards Neotron Cosmos and Earth 2. So

after a decade of contributing in the AI

research and innovation meta fell back

in both performance against proprietary

labs in the US and research against open

labs in China and now Meta is sort of

stuck in the middle worst of both

worlds. Now when we look at MATA's

capital expenditures over time, we can

see an inflection point right around the

DeepSc moment where they started to

increase their capex heavily. Deepseek

released their R1 model back in January

2025 which proved to the world strong

capabilities while keeping the cost down

to less than $6 million in GPU run cost.

And only 3 months after the Deep Seek

moment, Meta finally released their next

Llama 4 model, which basically flopped

even though benchmarks showed otherwise.

And that April sparked a code red moment

for Meta to do something drastic. Around

June 2025 was when Meta started to

change their business model entirely.

Mark Zuckerberg went on a spree of

acquisition where he took majority

ownership stake in Scale AI and poached

Alexander Wang to form a new initiative

called Super Intelligence Lab. And he

followed up making insane offers to top

AI talents to join his team. And since

then, Meta's capex has not only

increased towards AI, but the projected

capex over the next four years will add

up to an insane number in spending. So

much so that recently Meta announced a

multi-billion dollar deal with AMD as

well as multi-year deal with Nvidia to

purchase their CPUs, GPUs, and Ethernet

switches to fill their growing data

centers across the country. And speaking

of GPUs from Nvidia, if you're

interested to learn more about AI,

Nvidia's GTC conference is coming up

from March 16th to 19th. You can join

the session deploying AI agents at

Enterprise Scale online for free. I'll

be attending the GTC this year thanks to

Nvidia's invitation to host me with a

complimentary badge and stay. And this

year's GTC has a lot of interesting

topics to go through from data centers

to robotics as well as all the sessions

surrounding Neotron models as you can

see here. And if you sign up for any

session using the link to participate

online, I'm doing a giveaway courtesy of

Nvidia for an RTX 4080 Super GPU. And

joining sessions online is completely

free. So make sure to sign up for the

giveaway and take a selfie to qualify.

link in the description. Now, despite

Meta's strong cash position, people are

still very doubtful towards Meta's

ability to actually execute on their AI

initiatives. The closest hint that we

have in innovation is in Meta AI glasses

that uses Llama 3.1. And their first

proprietary model that's supposed to

drop in early 2026 called Avocado, which

supposedly already beat other top open

models, comparing base model against

base model. But around mid 2025, as more

and more compute became available, what

we started to see is the increase in

innovation speed among Frontier Labs

where new checkpoints were being

released every month or every 2 months.

So if you factor in other Frontier Labs

that all release on their own schedule,

we practically have a new model release

every other week. So in a way, Meta is

met by this dilemma between going the

route similar to Apple and Microsoft

where they're licensing LLMs from other

labs or just compete directly by

spending hundreds of billions of dollars

in infrastructure. Meta does have a huge

upside when we think about their

distribution network on platforms like

Facebook, WhatsApp, and Instagram where

each has more than 2 billion active

users at a given time. This kind of

scale is something that OpenAI and

Anthropic just does not have. Which

means even if Meta is slightly behind

innovation, Meta's ability to convert

their innovation to revenue dollar is

significantly faster and easier than

OpenAI and Anthropic. This kind of

family apps that Meta has also give them

huge access to training data that exists

in the form of texts, images, and videos

that other Frontier labs might not as

easily get their hands on. And even if

the underlying model isn't as strong,

the growing compute demand from agentic

use cases will require Meta to use their

data centers across the country. But

meanwhile, people are still highly

skeptical towards Meta given their track

record of Metaverse and Llama 4 that

both flopped. There are other reasons

why people are still nervous about Meta,

especially around how they are financing

their data centers. Meta's recent data

center deal involved offbalance sheet

maneuvering to secure financing to cover

construction cost while maintaining

their credit as if they don't have a

direct exposure to a $30 billion debt.

And this isn't necessarily because

people think Meta can't make the payment

per se, but rather it's the involvement

of private credit like Blue Capital,

unlike a cash goal company Meta, that

could crumble, which might trigger a

huge domino effect to the rest of the

market and the rest of the economy. What

do you think? Do you think meta has a

good chance in actually working through

all of its difficulties?