Scale or Bail: Meta’s $15B Panic Play

Zuckerberg's shaking off the metaverse, putting Yann LeCun in the corner, building a new line of business in cloud services, and looking for an end run around model collapse while throwing sand in the eyes of his competitors. That's what $15B and Scale AI gets you. Let's talk about it today on Minimum Viable Banter.

Episode Transcript

Rod Edwards (00:00):
Hello, and welcome back to Minimum Viable Banter.

(00:03):
My name is Rod Edwards, and youcan find me on LinkedIn. Search
for Rod Edwards in Winnipeg orat rodedwards.ca. Today, I'm
here to talk about somethingthat I feel like is a little bit
under the radar, which is Metapumping $15,000,000,000 into
sort of buying Scale AI or atleast 49% of it in a deal that

(00:24):
would also make the CEO ofScale, Alexander Wang, the head
of Meta's new superintelligenceAI unit. What does all this
mean?
It means that Zuckerberg'sshaking off the metaverse,
putting Jan Lakhan in thecorner, building a new line of
business and cloud services, andlooking for an end run around
model collapse while throwingsand in the eyes of his
competitors. Let's talk aboutit, today on Minimum Viable

(00:47):
Banter. Thanks so much forjoining me. I'd like to set the

(01:15):
stage by starting with aquestion asking what exactly is
Scale AI? I think there'sprobably a little bit of
confusion or misperception inthe community about what the
company actually does.
The name of course makes itsound like a large language
model company like a peer toAnthropic or OpenAI or something
like that. But the truth of thematter is Scale is a data

(01:37):
annotation company, which meansthat it's a labor company. It's
a staffing organization. Scalehas thousands of employees
working for shell companies inlow cost labor markets like The
Philippines, Africa, and soforth. And their job is to tag
millions of photos, classifymillions of snippets of text and
so forth.
This is the work of dataannotation, creating the

(01:59):
datasets that large languagemodels are trained on. The short
of it is that scale is notradically different from
Amazon's Mechanical Turk productfrom back in the day. It's a way
to provide, and I'm going to saythis in sort of a buzzy way, but
it's a way to provideprogrammatic access to human
cognition. So when Scale AIdescribes itself as data as a

(02:20):
service, it's because they'vewrapped a layer of technology
and APIs around the messy gruntwork labor business of
annotation. And that made Scaleuniquely attractive to the AI
industry just as it hitsinflection point.
The stars really aligned forAlexander Wang. He was in
exactly the right place andright time to pivot what started
as something closer toTaskRabbit into an annotation

(02:41):
service tailored to servefrontier language model
developers and in doing so, getaccess to their incredible
spigots of money. So twoquestions to address given what
Scale AI is. First, what isZuckerberg hoping to achieve?
And second, why with Scale AI?
The Scale AI acquisition is onlyone part of what appears to be

(03:04):
an existential push at Meta toreally restore their place in
the AI industry by hiring somebig players. I'm sure everyone's
heard about Sam Altman'sresponse to the $100,000,000
signing offers for top AIengineers, which is absolutely
bonkers. Today, there was morenews about Meta making
significant acqui hires fromGitHub and elsewhere. Meta is
spending the order of billionsof dollars to capture top AI

(03:27):
talent. That's intense.
The takeaway from that, theinevitable conclusion is that
Zuckerberg is existentiallyscared for his core businesses.
And I would also say that he'sfrustrated that he missed a big
thing. Zuckerberg spent$65,000,000,000 and several
years fiddling around in themetaverse and missed most of the
generative AI gold rush. Not tosay that Meta wasn't in the

(03:50):
space, but it was in a way thatwas clearly not appear to the
OpenAI's Anthropics and Googlesof the world. In recent months,
Meta's model credibility hasreally tapered off since LAMA
four, which was widely regardedas pretty mid.
So remember for a moment whoMeta's customers are. It's
advertisers. Their product,Meta's product, is attention,

(04:12):
yours and mine. And you and Iare spending more time
interacting with AI and lesswith traditional web properties
and social media. At the sametime, social media is getting
devalued by AI slop, whichcreates a downward spiral that's
going to impact Meta's margins,prices, and ultimately revenue
and enterprise value of alltraditional social media

(04:34):
properties.
So Zuckerberg has to be feelingthe heat and has correctly
identified that Meta needs to bea player in the AI space to
shore up its core properties.Beyond those core properties,
Meta is doing another smartthing, getting into the cloud
services space, hosting theirown models, and making them
commercially available. Thisfeels like a good hedge against

(04:54):
social media spending beingeroded. If AI is damaging core
properties, Meta can be at leastthe one benefiting off that
cannibalization, making moneyoff of it, and kicking off a
completely new revenue stream inthe fastest growing market. So
to recap, why does Zuckerbergsuddenly care about AI?
He's scared for his corebusinesses. He's aware that he's

(05:17):
missed a window to be a peer toanthropic and OpenAI, leaving a
lot of shareholder value on thetable. So great. Zuckerberg is
course correcting. That's greatnews and astute management.
And Zuckerberg is lucky becausehe's in a position where he's
got the resources to actuallythrow money at course correction
in a way that not manyorganizations would be able to
muster. So the next thing is whyScale AI? First thought, maybe

(05:43):
it's Alexander. Maybe he thinksAlexander Wang as the leader of
the super intelligence unit atMeta is actually the person best
equipped to win the race tosuper intelligence or at least
artificial general intelligence.I think part of the x factor
here is the mystique ofAlexander Wang himself.
He's super young and supersuccessful, and I feel like
Zuckerberg must see a lot ofhimself in Alexander Wang. Does

(06:06):
that make him a good bet to getto AGI or ASI? I think that's to
be determined. I would arguethat Alexander Wang today is a
great operator. He's got a trackrecord of smart pivots and
creating an effective, fastgrowing product in a low margin,
high headcount industry.
That's a superpower in and ofitself, but he hasn't been hired
to build a highly performingoperations unit. He's not a

(06:29):
frontier model developer or amodel of the super intelligence
Illuminati or whatever. So to bedetermined. This is a bet that's
being placed on someone who'sdone some amazing things very
quickly and someone that'sprobably very aligned with
Zuckerberg's line of thought.Time will tell if that works
out.
Next thought about why Scale AIis, well, it's in the name of

(06:51):
the company. Scaling laws, Iwould say today, are up in the
air. That is to say, what'sgoing to power the next round of
AI performance gain? Some thinkit's inference time. Some think
it's architecture, biggermixture of experts collections.
Others think it's the quantityor quality of data used to train
a model, especially now asInternet data is increasingly

(07:14):
locked up behind intellectualproperty claims and increasingly
vulnerable to model collapse asAI content infiltrates
everything. If the next modeladvances are going to be driven
by data quantity or quality,then Zuckerberg is making a very
smart bet, getting into bed withthe foremost supplier of AI
data. The other possibility, ofcourse, is that this is throwing

(07:35):
sand in his opponent's eyes. Thecynical angle to all this is
that the deal was engineered toslow down other players in their
pursuit of AGI. XAI, Google, andOpenAI are all pausing work with
scale.
Whatever they were doing withscale, they were doing because
presumably they needed it. So,this buys time for Meta to catch
up as those other tier oneplayers are searching for new

(07:57):
annotation partners. Alsointeresting to note is the KG
49% number. Meta deliberatelystayed just under the margin of
control, and that 49% feelsfinely calibrated to avoid
regular regulatory oversightwhile scaring off other scale
customers. My initial reactionto this is that I find it hard

(08:19):
to imagine that it's not part ofthe equation, but the fact that
it is probably feels a bitgross.
The truth of the matter is, Iwould say AI is firmly in sort
of like Games of Thronesterritory right now. Battle
lines are being drawn, backs arebeing stabbed, lots of gross
things are happening and willcontinue to happen. This is just
sort of a more public examplethan what we're used to seeing.

(08:42):
So at the end of the day, whyscale? One, Zuckerberg likes the
founder and probably finds themvery relatable.
Two, data quality is probablyreally important in the next
generation of models. And three,the ability to throw a wrench
into the data pipeline of othertier one AI players probably has
some value in and of itself. Thebottom line on Scale AI is that

(09:02):
Zuckerberg's come to terms withthe fact that he's had the wrong
people with the wrong prioritiesand has left money on the table
as a result. He thinks Scale isgoing to plug at least some of
those gaps for him and help himjump start Meta's AI factory, if
not leapfrog his peers. Theother thing to consider here is
that Meta's market cap is closeto $2,000,000,000,000.
And so a $15,000,000,000 bet onScale AI feels like a small bet

(09:26):
to place given the potentialupside here. Those are my
thoughts on Scale AI. If you'dlike to connect, find me on
LinkedIn or at rod edwards. Ca.And thanks so much to listening
to minimum viable banter, andI'll see you next time.