DeepSeek vs OpenAI: Strategic AI Competition
Key Points
- DeepSeek’s playbook is to quickly re‑release cutting‑edge models (e.g., OpenAI’s latest) as open‑source equivalents, offering ultra‑low‑cost APIs to lure cost‑sensitive developers and capture market share.
- Their business model relies on cheap training tricks (e.g., the disputed $5 M claim for a Claude‑Sonic‑class model) and a “copy‑the‑next‑big‑release” pipeline that can pivot to any rival breakthrough (Anthropic, Google, etc.).
- OpenAI counters this by emphasizing data security—U.S. enterprises may avoid sending proprietary information to a China‑based provider—and by banking on its head‑start to deliver exponential performance jumps (e.g., moving from R1 to R3/R4 within months).
- OpenAI’s long‑term bet is that corporations will be willing to pay a premium for the superior intelligence that arises from those exponential gains, offsetting the pressure from cheaper, open‑source alternatives.
Full Transcript
# DeepSeek vs OpenAI: Strategic AI Competition **Source:** [https://www.youtube.com/watch?v=U6bZI-0oAjc](https://www.youtube.com/watch?v=U6bZI-0oAjc) **Duration:** 00:11:23 ## Summary - DeepSeek’s playbook is to quickly re‑release cutting‑edge models (e.g., OpenAI’s latest) as open‑source equivalents, offering ultra‑low‑cost APIs to lure cost‑sensitive developers and capture market share. - Their business model relies on cheap training tricks (e.g., the disputed $5 M claim for a Claude‑Sonic‑class model) and a “copy‑the‑next‑big‑release” pipeline that can pivot to any rival breakthrough (Anthropic, Google, etc.). - OpenAI counters this by emphasizing data security—U.S. enterprises may avoid sending proprietary information to a China‑based provider—and by banking on its head‑start to deliver exponential performance jumps (e.g., moving from R1 to R3/R4 within months). - OpenAI’s long‑term bet is that corporations will be willing to pay a premium for the superior intelligence that arises from those exponential gains, offsetting the pressure from cheaper, open‑source alternatives. ## Sections - [00:00:00](https://www.youtube.com/watch?v=U6bZI-0oAjc&t=0s) **DeepSeek’s R1 Model Strategy** - The speaker breaks down DeepSeek’s newly released open‑source R1 model and argues that the firm’s strategy of re‑creating cutting‑edge AI, providing low‑cost APIs, and targeting price‑sensitive developers aims to capture market share from larger players. ## Full Transcript
I want to talk about R1 which was the
new thinking model open source and
released by Deep seek today but I want
to talk about it strategically because
this is happening a lot and I am tired
of people being surprised so at the end
of the day I want to lay out both the R1
model and I also want to lay out the
strategy of each player in the room when
it comes to how they approach Ai and why
so we're going to go through them first
deep seek deep seek has done this before
they are in the habit of taking Cutting
Edge releases specifically from open Ai
and releasing them later open source
just as good and getting very cheap API
costs along the way why would they do
that why would it make sense these are
the guys that made a lot of headlines
for saying deep seek V3 which is sort of
a four class Claude Sonic class kind of
a model was trained for $5 million
people debated whether that was true but
the point is it's much cheaper than the
original cost of train chat gp4 which
started the whole four class Revolution
so at the end of the day their strategy
whatever their actual training cost was
is to gain market share using their apis
which are very cheap if they can shift
developers over to deep seek over time
by providing more and more effectively
equivalent
intelligence they're going to be doing
really well because developers are very
cost sensitive and will be happy to move
to a model in most cases that they can
get for cheaper and so to me I would
expect them to release 01 Pro copycat
next and as soon as there's another
model that's more Cutting Edge like 03
they're going to start working on that
if there's something that another lab
releases that is also Cutting Edge in a
different way maybe anthropic comes out
with something maybe Google comes out
with something they'll work on a version
of that too their whole strategy is to
essentially come from behind deliver
cheap API cards toath the market so much
for deep seek let's move to open AI why
are they not more worried this is a
situation where intelligence costs are
coming down in a matter of months 01 is
not a mo anymore what do you do well
they're betting on two things first
because deep seek is a Chinese model
they're betting that American
corporations at scale will find that
it's not secure to send their data to
China and would prefer to keep their
data in the US uh and so they want to
use a us-made model and so there's sort
of an inherent Advantage for open AI at
that point second they are betting on
the exponential curve at the end of the
day a little bit more time for them to
work on a Model A little bit of a Time
Advantage can translate to exponential
performance gains and so if you think
about it 03 can be substantially better
than 01 in just a matter of months and
04 beyond that and what they're betting
on is that that exponential gain an
intelligence is something that
corporations will pay a premium to
access over time we'll see if they're
right but essentially what they're
betting on is that there's
disproportionate gains to be had for
being an American company that lives at
The Cutting Edge and that is able to
continually deploy these extremely
Advanced models and that eventually they
will get into a point where they are
using a recursive feedback loop to very
very rapidly improve these models and
that could potentially help them expand
their lead what's interesting is we see
some sign of that already there was a
little leak that came out uh in the last
week that part of what made open AI so
excited over Christmas and getting into
New Year's is that they were able to use
four different instances of 01 to
rewrite their Transformers codebase I it
sounds like a movie but it's not it's
the Transformer based architecture
that's at the heart of large language
models and they asked these 01 instances
to look at the codebase and see if they
could make it more efficient and they
did substantially and if that's the case
then we are getting to a point where AI
can help build AI which means that for a
company like open AI if they are at The
Cutting Edge that feedback loop runs
faster and allows them to gain an Ever
bigger advantage over time that is
probably the corporate bet they are
making let's look at a couple of the
other players though let's look at
Google what's Google's bet here Google
is playing defensively at the end of the
day Google has a search position to to
maintain everything they've done for the
last 20 years is about defending their
search position and deploying a little
bit of spare cash to bets it's a very
conservative corporate strategy actually
it reminds me a lot of like a General
Motors strategy but if that's their play
why are they so hard in on open Ai and
why are they so hard in on artificial
intelligence I should say the reason
they're pushing so hard on AI is because
at the end of the day they see this as a
disruption to search
one of the things that uh I was reading
about is a CEO of a fairly large
corporation saying off the Record that
he has seen the search funnel collapse
with Google where organic search for his
Corporation has just started to just
erode and is like half of what it once
was look that's an anecdote right it's
not that I'm saying search has gone away
50% but I think if you're sitting there
in Google's chair you are worried about
the long-term erosion of search because
of AI you're not really worried about AI
tools like perplexity that just do
search what you're worried about is that
people will use search on chat GPT
instead over time you're worried that at
the end of the day the active searching
is really about gaining knowledge and if
these AI models have the knowledge why
would you go to
Google and so they're desperately
playing from behind to get the benefits
of AI so that you stay on the google.com
homepage and do your searches there
that's why they rushed so hard for those
summaries even though people laughed at
them and said the summaries were
terrible that's why they rushed so hard
to make sure that they deploying on
Google Cloud AI Solutions corporations
can trust at the end of the day if it's
the same position with Google Cloud as
it is with search if corporations can't
see AI solutions that they can trust and
leverage within their own cloud
footprint they're going multicloud
they're going some else for that AI
Amazon is actually similar if you look
at Amazon's position they deployed
15,000 engineers and they built a timer
that's what Alexa is Alexa is a smart
timer it was an open joke at Amazon
15,000 Engineers to build a smart timer
they missed the boat on large language
models they have been playing from
behind that is their incentive to work
with anthropic that is why they are
pushing so hard on their trinium silicon
they are desperate to regain a strategic
advantage that allows them to maintain
the margin leverage they have at
AWS that is what matters to them and so
they have to push hard on AI that is why
Jeff Bezos has gone back to work several
days a week just reviewing AI six pagers
they've got to get back in the
game and from their perspective if all
they do is defend their current market
share and continue to grow it the way
they have been they're doing okay but I
know Amazon because I used to work there
and I know that they're hungrier than
that and so they're actually not going
to be satisfied with just defending
their position they are going to insist
that they eventually be able to take the
number one slot in the AI world and
they're going to keep innovating and or
buying companies until they do so we
will see how that all plays out that's
their ambition right now they're just
trying to defend their market share the
way Google cloud is and this by the way
the the cloud products is why they
charge for these models
that is why Zuckerberg and meta do not
charge for llama because they're not a
cloud company they are a company that
makes money off of ads sold to
eyeballs and so for them if they can use
these AI models and they can juice an
ecosystem for free where developers know
their model architecture and their
ecosystem so they can pull in Talent
anytime they want which is exactly what
they did by the way you know the 5% riff
they did for performance it was
explicitly to dump out the bottom 5% of
the company and bring in fresh Talent
which is really hard on morale but it's
really easy to do if you have an open
source ecosystem that's that popular in
llama you can just bring in developers
that are already familiar with llama
super easy and that's what they're doing
because their goal is to use that
ecosystem to generate personalized feeds
for the billions of people that use
their products so that instead of
looking at friend generated content
you're going to be looking at AI
generated content and AI generated ads
and even if you feel weird about that
and I feel weird about that I guarantee
you it's going to perform well now they
have pushed too hard in some cases you
saw the botched roll out of the
avatars a couple of weeks ago where an
avatar famously said I was out helping
over Christmas at a Food Kitchen or
something like there was some like
charitable cause thing and everyone
dunked on them and said you're a madeup
avatar what are you talking about this
is really
offensive that is just a bump in the
road they are coming back they're going
to do it again and they're going to do
it again because it's really really
lucrative to increase the eyeball time
in their app that is what they're going
for so that's meta's strategy that's why
meta is open sourcing that's why Google
and Amazon are not what about Microsoft
Microsoft is interesting because
Microsoft has a cloud product but they
also have open AI in the fold so they
have two ways to win which is part of
why Sacha nadela is doing so well these
days uh they win because they've defined
a financial term for artificial general
intelligence that means that open AI
must cough up something like a hundred
billion in profits before AGI has been
achieved which by the way if you saw Sam
Alman today saying Tamp down the hype we
haven't built AGI he can't say they have
built AGI because it's a hundred billion
dollar statement he can't say it so like
take it for like with a block of salt
right like he's never going to say that
because it's a corporate statement okay
what is Microsoft's game though not only
are they getting paid if open AI does
well which it well they are also getting
paid by taking the open aai models
deploying them in Azure and selling them
with the openai label behind Azure
they're in a really good spot on that
and then they can pull the tech into the
consumer side and all of
that that's a really good spot to be
you're defending your Cloud business
you're pushing the open AI brand which
is the best known brand in AI at this
point and if open AI does well you also
make money that way it's a good spot to
be so that's a quick tour of the
different major players in the game what
they're thinking why they're thinking it
deep seek made me want to do that
because at the end of the day if we
don't understand the incentives we're
confused by the news so there you go
those are the Strategic incentives I
hope this was a nice tour for you um
yeah cheers