The latest 2023 report is worth a flick (all 160 slides).
This graph, for example, shows the largest Nvidia H100 chip clusters – interesting to see TSLA there, who also run the 4th largest A100 cluster in the world.
Or see Slide 76 which suggests that Nvidia’s advantage (the use of its chips in academic papers) continues to increase.
Fascinating history of how we all ended up doing PowerPoint slides all the time.
The creator of PowerPoint also has a colourful story – “It’s hard now to imagine deafening applause for a PowerPoint—almost as hard as it is to imagine anyone but Bob Gaskins standing at this particular lectern, ushering in the PowerPoint age. Presentations are in his blood. His father ran an A/V company, and family vacations usually included a trip to the Eastman Kodak factory. During his graduate studies at Berkeley, he tinkered with machine translation and coded computer-generated haiku. He ran away to Silicon Valley to find his fortune before he could finalize his triple PhDs in English, linguistics, and computer science, but he brought with him a deep appreciation for the humanities, staffing his team with like-minded polyglots, including a disproportionately large number of women in technical roles. Because Gaskins ensured that his offices—the only Microsoft division, at the time, in Silicon Valley—housed a museum-worthy art collection, PowerPoint’s architects spent their days among works by Frank Stella, Richard Diebenkorn, and Robert Motherwell.”
Everyone is trying to figure out what AI means for GPU demand.
It’s hard as the true picture is muddied by providers investing in their own customers, demand-pull forward, and strategic buying ahead of having a real use case (see Saudi, UAE, U.K.)
Confounding all this is Meta releasing Llama 2 for almost free, followed most recently by its coding version (by far the most useful application of AI so far).
This matters because training is a lot more GPU-intensive than inference. Free models mean less training needed. This specifically matters for Nvidia’s H100 chip (which by the way weigh over 30 kgs!).
Qualcomm actually thinks processing might happen right in our phones (they of course would benefit most from this).
“Eventually, a lot of the AI processing will move over to the device for several use cases. The advantages of doing it on the device are very straightforward. Cost, of course, is a massive advantage. It’s — in some ways, it’s sunk cost. You bought the device. It’s sitting there in your pocket. It could be processing at the same time when it’s sitting there. So that’s the first one. Second is latency. You don’t have to go back to the cloud, privacy and security, there’s data that’s user-specific that doesn’t need to go to the cloud when you’re running it on the device. But beyond all of these, we see a different set of use cases playing out on the device.” Qualcomm CFO Akash Palkhiwala (via The Transcript).
Data for the UK suggests we have hit saturation point.
“The growth of SVoD household penetration slowed in 2022, and this continued into early 2023 as the rising cost of living, combined with SVoD service price rises, put greater strain on household budgets.“
Podcast advertising leads to pretty good returns for brands.
“After conducting a study with 250 advertisers and marketers, it says two-thirds (67%) of podcast ad buyers say that every $1 spent on podcasts returns between $4 and $6 for their brands.“
Yet SPOT is struggling to capture this – why?
This blog post covers a lot of reasons. For example:
“[What’s] most misunderstood about Spotify is Spotify doesn’t get to monetize all the podcast content that they have. So in the most recent quarterly earnings report, they say that they had 5 million podcasts on their platform, but 99.9% of those podcasts, Spotify does not get to monetize.“
