NVDA通过垂直整合（拥有土地/电力/租赁）来缓解GPU需求风险，并通过NVLink Fusion生态系统锁定维持竞争护城河。

Now, on today's episode, I want to take a closer look at Nvidia, especially because two big reports came out came out about Nvidia. First, they're investing in the new semiconductor company. Why are they doing that? The second thing has to do with the massive data center for Anthropic and a lot of questioning of why Nvidia is involved so much.

The first thing with Nvidia is there was a media tech investment that came out yesterday. Nvidia has also invested 3.5 billion in convertible bonds issued by MediaTek.

Nvidia, for those that are not familiar, the DGX Spark, which is kind of this AI supercomput meant as a desktop. I actually just ended up purchasing one about a week ago and working on it right now. that is partnered with MediaTek.

The upcoming laptop, the I forget what it's called. Um the RTX Spark, which is a laptop that has Nvidia's AI supercomputer in it, where they're trying to really combat Apple in kind of that consumer AI space.

But the other side of the story is kind of the custom solution, right? that custom solution of chip big players designing their own AI chips. The fear for Nvidia or one of the biggest bare cases for Nvidia has always been hyperscalers designing their own AI chips.

But Nvidia about a year ago decide to come up with NVLink Fusion where it actually opens up its ecosystem. If you are a chip designer and designing your own chip, you can actually use a lot of the stuff Nvidia has done.

And the great thing about this is one it keeps you in the Nvidia ecosystem.

So if you're first let's say a hyperscaler and you're buying a lot of barrel rubins but then you in the future you want to buy a lot of your own chips by being part of the Nvidia ecosystem with Nvidia link eusion the collaborations and communication within these two types of racks become a lot easier right if you're designing to similar specs that Nvidia is designing if you need to keep swapping if you want to build one data center that data center is not just going to be focused on a custom chip it could be focused on all of Nvidia's products and your custom chips.

So, it gives the flexibility and Nvidia says that it's no longer about designing a chip, right? It's no longer about that. It's the supply chain. It's the rack scale solution. It's the communication. It's the optics. It's the memory, right? It's the way you communicate the GPU to the CPUs for Atantic tooling. It's the way you build the whole rack. It's the way you build the data center. It's the type of cooling that you need. It's the type of software that you need. It's no longer just an AI chip, right?

And and and that's where Nvidia has really gone with this NVLink fusion now, right? Nvidia is not going to share that data purely to its competition, right? What it is, there has to be a middleman, right? A middleman that helps kind of okay, this is the part that we're working with. This is how we're going to connect it with.

Cuz they're not going to share their information. They're going to give it to someone that's the middleman that helps with the connection process. So Nvidia gets some form of collection, and even if a company designs their own ASIC or own custom GPU or AI accelerator, Nvidia still wins.

With the NDLink fusion, you have a lot of things like memory solutions like communication solutions. Here they talk about it fusion chiplet which is connecting XPUs to the Nvidia NVLink scaleup fabric using Nvidia Photonix or electrical interconnects. You have the Nvidia NVLink C2C which provides high bandwidth energy efficiency connected between XPUs and Nvidia CPUs. And then you have Nvidia NV HBM which is integrating customized memory capabilities to increase bandwidth and energy efficiencies while dedicating more silicon area to compute.

Now the thing that I found interesting so it's great news right I feel like this is Nvidia saying that look we're investing in more players like this because it does seem like the demand from hyperscalers are there.

which we know is going to be part of the NVLink ecosystem. I wonder this is just my thoughts right now. Broadcom is not part of the NVLink fusion and I don't think they want to be part of it but I think Google wants to be a part of it right Google wants to be a part of this NVLink fusion and if brocom doesn't want to do it Nvidia says that's okay Brockcom you don't have to do it but Google now has the option to kind of work with us with either Marll which they are with AI accelerators and attach things to the TPUs not necessarily the TPU itself and now you have MediaTek also part of that ecosystem system, which again is pretty close to Google.

Anthropic will pay Nvidia backed cloud provider Lambda for compute which is funny. First, Nvidia has invested in Anthropic obviously. Then Anthropic will pay cloud provider Lambda for compute which is already which is also backed by Nvidium and Lambda will use data centers developed by HU 8 to provide compute who also recently signed an agreement with Nvidia. Nvidia holds the lease on the data center.

Nvidia I think is really smart cuz everything looks like circular financing around here right and it's super scary but let me share my thoughts. Nvidium when they first saw this AI wave, the first thing they wanted to do is make sure the AI companies were okay because they're the ones that are going to force competition and when you force competition, you force innovation. So, and Nvidia has invested in a lot of AI companies. A lot of those AI companies are doing well, right? Nvidia has said that he's most likely not going to invest in Enthropic or OpenAI or any of these other companies much more because they're doing well. They're they that ecosystem has been created.

All right. outside of that ecosystem that they didn't want to fail. Nvidia didn't want the neocloud space to fail, right? Because you don't want just your hyperscalers to succeed. You don't want just your Googles and Amazons to be the rulers of all the data centers because that's bad for business.

So, Nvidia did two things. First, to reduce this dependency of some of these players, helped improve the neocloud market, right? The other thing is it helped accelerate the AI the AI infrastructure market. Meta and all these companies weren't going to be investing heavily like they are now in the AI space. A few years ago, Nvidia was helping Coree built up. Nvidia was helping Nebas built up. Nvidia is now helping a lot of other NeoCloud players, but it wanted to build the road map for that AI cars that were designed by the AI companies to move. And that's what it did.

Now Nvidia seems to be believing one of the next issues is actually the land power and shell. We actually saw Nvidia do it. So now Nvidia what we see is really providing a nice amount of agreements with these land power and shells.

So Enthropic is going to rent Nvidia's compute. Lambda from Lambda. Lambda is going to buy the compute from Nvidia, right? And they're not going to own the land power or shell. The lambda is going to rent the land power and shell from HUD 8 and at the end of the day Nvidia holds the lease on that data center.

Now I believe that bottom line is also very important. Nvidia holds the lease on the data center. I also believe Nvidia wants not the hypers the AI companies themselves to own the compute itself. The reason being is first Nvidia holds the lease on the data center. Why? Because if anything goes wrong with Lambda or Anthropic, Nvidia still holds this massive power which is a constraint and a bottleneck in the industry that if they hold it and something happens bad to Anthropic or Atlanta, Nvidia can give it to another customer that's succeeding in the AI store.

Right? If power is the limitation, let me grab as much power as I can and if anything goes wrong, that power is worth a lot and I can make sure that Nvidia's solution end up in this land.

Now the other thing I mentioned lambda right I think Nvidia does like Neoclouds to own a lot of it because if Anthropic goes bad right if Anthropic goes bad doesn't mean Lambda goes bad Nvidia already has this partnership with Lambda okay anthropic you don't need the compute stop renting it I actually have another customer that will take this compute gladly from you open AI or or maybe Meta or or some new startup that just came up so Nvidia has really in my opinion eliminated a lot of the risk of this GPU business, a lot of dark GPUs by owning the compute, by owning the LAN PowerShell, by having partners in the Neocloud space.

And then because they're such a top level in the chain, they are able to see where the demand is and if anything ever sees a hiccup, they can say, "Okay, if you need a better customer or if this customer error fails, this is who you're going to go with." Uh and then obviously Nvidia helped improve in my opinion this innovation in the AI story which is driving the crazy demand that we're in right now.

So many people see circular financing but all this is happening during different time frames. I think right now the main focus is land power and shell.

Now, what I again think that would be bullish would be the NeoCloud players because their AI compute. If you can't put more compute on your compute that's online is going to be worth a lot more.

At the same time, if there are going to be dark GPUs, I would completely believe Jensen would not be making these types of investments that he is doing right now. Right. Jensen is investing so much in memory, so much in in the supply chain. If he truly believes there is dark GPUs in 2027 that he wouldn't want to do that, right? Because what's going to happen, right? If 2027 you have dark GPUs, then 2028, even if you have power turning on like crazy, you still have all those dark GPUs to turn online be and and you have all this chip capacity that is being built out.

And this is why I actually think Meta is one of my favorite Mac 7s right now outside of Nvidia and Amazon, but I do believe him selling intelligence is a great great thing and it's going heading in the right direction.