00:00Alphabet, the Google parent raised the top end of its CapEx plan for this year to $205 billion.
00:06There's cloud growth, there's Gemini engagement all there, but the discipline on spending
00:11is a bit of a concern. Joining us is Eric Sheridan, Goldman Sachs, co-business unit leader of the
00:16technology, media and telecommunications group in global investment research. He says Alphabet's
00:22well-positioned to benefit from the growing demand for AI across both consumer and enterprise markets
00:28and reiterates a buy rating while lowering his 12-month price target to $435 from $440.
00:37Eric, welcome to the program. Not a surprise, really, that they would raise the CapEx expectation
00:44for this year, but the reaction to that seems a bit severe. They did swing to negative free cash flow
00:51for the first time as a public company. Was that it?
00:55They did swing to negative free cash flow, and I think there's a mixture of signals versus noise
01:01in this print. The long-term signals are search is a stable business, YouTube continues to gain
01:07momentum across the broader media landscape, and Google Cloud revenue continues to re-accelerate
01:13and will likely re-accelerate in an outsized way for most of the next one to two years.
01:18They made some decisions short-term to raise CapEx and strike deals for third-party Compute that are
01:25impacting OpEx that are all about closing some of the demand versus supply gap that exists around
01:31Compute today because they didn't want to slow growth and disappoint external clients. Now, we
01:37certainly are cognizant that in this market environment, over-indexing to investment and
01:42under-indexing to short-term return isn't being rewarded, but we think Alphabet is making the
01:47right long-term decisions when angling against the larger market opportunity for AI over the next
01:53couple of years.
01:54We got a lot of stats, stats about Gemini, stats about enterprise adoption, the cloud units growing 82%
02:02year-on-year. For me, the really simple question is, is Google doing well at AI?
02:10They are still an AI winner in our view. The market took a step back from that view overnight. The
02:17delays
02:18around 3.5 Pro and the fact that they no longer have a foundational model that sits right at the
02:24frontier of performance and benchmarking has definitely taken a little bit of the shine off
02:30the AI winner theme. What Sundar Pichai talked about last night is that they're likely going to have to
02:36wait for Gemini 4 to be back at the frontier of performance with AI models. Two points. I think
02:43generally when you look at access to chips, data, the ability to train these models, we think Alphabet
02:49is as well-positioned as anyone, but there can be short-term gaps that open up between performance
02:55and training runs around these models. More importantly, we think the world is broadly
03:00shifting from token maxing to token optimizing. And some of these other models that are around speed
03:06and efficiency, including some of the flash models that they've released, will allow them to remain
03:12very competitive for incremental workloads. But investors want to see companies spending this
03:17amount of money, then they want them at the frontier of model performance. They might have
03:21to wait a few months for that with Alphabet. That is a conversation I've been having with CEOs
03:26all across the stack recently, the difference between token maxing and token optimizing.
03:32If Google nails that, where does it show up, right? I think you write right at the top of your
03:36note that
03:37cloud revenue estimates now revised even higher. Is that still the metric to follow on how they are
03:43being used out in the real world? Yes. And we believe companies like Alphabet and next week
03:49we'll hear this from Amazon that are going into enterprise customers and saying, we're going to help
03:54you optimize your spend. It's not going to be about just buying tokens, no matter what the cost from a
04:00single model, but buying a wider array of tokens from a wider array of models is generally where
04:06this landscape is going. We wrote a note a couple of months ago about where the AI economy would go
04:11over the longer term. And I think what got lost in that note, Ed, would be the fact that to
04:16drive
04:16utility and to drive token growth, you need deflation. Every technology compute shift I've ever
04:23covered and analyzed has unit growth that comes with deflation because you have to incent
04:29adoption rates. And we don't think the AI economy is going to be any different than that.
04:33We don't have time for this, but China's focused on lowering dollar per token. America's focused on
04:37the quality of the token. I just note very quickly that the other hyperscalers as by association
04:43marketably lower today.
Comments