tensor processing unit
537
MENTIONS
325
EPISODES
214
PODCASTS
Search complete. 537 mentions across 325 episodes found for "tensor processing unit".
Oct 27, 2026
Seema ModyCORRESPONDENT
We're expecting OpenAI CEO Sam Altman to join Oracle co-CEO Clay McGurk on stage momentarily here in Vegas at AI World.
Seema ModyCORRESPONDENT
In addition to this news, you have Oracle Cloud Infrastructure expanding the number of accelerators or chips that can run on its cloud, including Google TPUs.
Seema ModyCORRESPONDENT
And we're told that this is the first time that Google AI chips will run on a hyperscalers cloud that isn't Google.
Seema ModyCORRESPONDENT
This announcement also including OpenAI Jalapeno, which is the Frontier Labs chip, and Cerebris.
Jack GambleHOST
How can you tell when the capital is defensively crouched? Higher interest rates are also making it harder to raise capital, giving lenders an upper hand as they can pick and choose which deals to finance.
Jack GambleHOST
In the North Star deal, a shell company is borrowing the $63 billion to buy the chips designed by Broadcom and Google known as tensor processing units or TPUs.
Jack GambleHOST
Why are they using shell companies? I'm telling you, whenever you hear shell company, run, do not walk for the exit.
Jack GambleHOST
Somebody is hiding something.
Jeremie Eliahou OntiverosGUEST
So effectively, all of that is just additional funding, which again, as I was saying earlier, just means these guys are going to have more money on their books, it's going to be easier to pay for stuff.
Jeremie Eliahou OntiverosGUEST
They got the Broadcom one as well, which is $42 billion, which is going to enable them to buy a whole lot more TPUs.
Jeremie Eliahou OntiverosGUEST
And so, yeah, overall, like, the 2027 compute roadmap for Anthropix fairly de-risked, which kind of goes back to what we said, like, in an article on a podcast a few weeks ago.
Jeremie Eliahou OntiverosGUEST
Like, the data started bailed out.
Ejaaz AhamadeenHOST
They train their own models, Gemini.
Ejaaz AhamadeenHOST
They have their TPUs, which is their own version of their GPU.
Ejaaz AhamadeenHOST
Arguably, they're going to be using their compute to train their models and to learn from all these other models that are being inferenced through their own cloud platform.
Ejaaz AhamadeenHOST
Now, what's interesting is Google is kind of playing both sides for a while.
8 MINS LATER
Ejaaz AhamadeenHOST
And remember, these contracts are non-cancelable.
Ejaaz AhamadeenHOST
So Elon Musk at any time can just shut Anthropic off from this and be like, well, I don't care anymore.
Ejaaz AhamadeenHOST
Now, another $115 billion Anthropic is spending on Google to acquire their TPUs, to diversify their GPU stack from NVIDIA's GPUs.
Ejaaz AhamadeenHOST
And this is a smart move, in my opinion, because too much reliance on a singular type of GPU or metal block means that you are just heavily reliant on that company, NVIDIA.
speaker_0HOST
Fortune, reading the filing, put the cloud commitments at at least $475 billion, and as much as 750 billion counting leases, stretching out to 2045.
speaker_0HOST
How do you buy that much computing power? Anthropic runs its models on four families of chips, Amazon's Tranium, Google's TPUs, and graphics processors from NVIDIA and AMD.
speaker_0HOST
It is now adding what it calls Anthropic Direct, leasing its own chips and housing them in data centers built to its specification.
speaker_0HOST
The idea is to lock in the building early, which is the slower and cheaper part, and buy the expensive chips later, once demand is clearer.
Type 3 AudioNARRATOR
Google, NVIDIA, and Sam Altman are all investing in companies pursuing nuclear fusion.
Type 3 AudioNARRATOR
The world's richest man is seemingly seriously discussing space-based data centers and Google have sent TPUs to space to test the concept.
Type 3 AudioNARRATOR
Terrestrially, next year US data centers will require 66 gigawatts of continuous power, up from 31 gigawatts in 2025.
Type 3 AudioNARRATOR
66 gigawatts is roughly equivalent to the average amount of electricity used by Germany in wintertime.
ColinHOST
The company has spent years designing accelerators around the demands of its own AI workloads, giving it an advantage that traditional chip companies cannot easily replace.
ColinHOST
Selling TPUs to third-party customers opens up a new revenue stream for Alphabet, with analysts forecasting a wide range of revenue from $100 billion to $250 billion as soon as 2028.
ColinHOST
The higher end of those forecasts represent what took Search decades to achieve.
ColinHOST
But the one caveat is HBM constraints and customer financing could limit how quickly the sales translate to improve cash flow.
Jason LangoneGUEST
You know, I'll say this is something that's going to continue to change, you know, in the coming weeks, months, years, as, you know, accelerators get more efficient, as, you know, additional players enter the space.
Jason LangoneGUEST
And we see things like TPUs and DPUs and, you know, just this wide variety of technology meant to accelerate things.
Jason LangoneGUEST
AI workloads.
Jason LangoneGUEST
I mean, I think today, if a decision needs to be made in real time or near real time, ideally, that decision is made on gear out at the edge.
Matt MurphyGUEST
yeah i mean i think you know when that's that's a really good new stat that they provided because i think people want to understand you know relative to what happened with core weave last week or a couple weeks ago like hey are people getting too far over their skis or are you getting leverage at scale and i think this you know this clearly shows that but you know like anthropic's always been a great story going back to when i invested you know they they've always had to do more with less when i first saw the company in early 2023 they had models that were comparable to and beating chat gpt on many dimensions and i think it hit at that time it was like 1 150 of the capital that that you know chat gpt is or you know open ai had raised to get there so this is a company that's always competed you know with special technology and to get a lot of leverage out of their infrastructure to optimize compute it's been amazing to see the way they've scrambled through the year.
Matt MurphyGUEST
It's all kind of public information about how they managed to get capacity from Elon and buy a bunch of TPUs.
Matt MurphyGUEST
It's just they're doing everything possible to secure compute.
Matt MurphyGUEST
And I think the game has moved from just having the best model to also having access to compute, being able to deliver that compute as efficiently as possible, and then really having the best harness where you're delivering a special set of services on top of your harness, like Cloud Code, Cloud for Life Sciences, and things like that.
Brett SchaferHOST
This is from Anthropic itself.
Brett SchaferHOST
There's also a TPU stipulation of $125 billion to $150 billion, which I'm a little confused about.
Brett SchaferHOST
I think they're required maybe to buy TPUs directly, which for anyone that doesn't know, TPUs are Alphabet's custom chip that they use for Google Cloud, and they've started to sell to other people.
Brett SchaferHOST
Amazon, and this is starting to add up here, $110 billion.
Brett SchaferHOST
in required spending.
44 MINS LATER
Brett SchaferHOST
Well, I became a little less.
Brett SchaferHOST
There's a lot of talk about chip diversification.
Brett SchaferHOST
Big commitment from Broadcom, big commitment from AMD, big commitment from Amazon and Alphabet on the Tranium and TPU stuff.
315 more episodes mention tensor processing unit.
Create an account to see the whole feed, search across every transcript, and follow the entities you care about.