Razor's Edge: ClawdBot is the Thing

"the computer is not the thing. it's the thing that gets us to the thing."-Joe Macmillan

One of the best moments in Halt and Catch fire is when Joe is celebrating the success of the cheaper/faster Giant IBM pc clone sale at Comdex, which required removing Cameron's interactive/memory hungry intuitive human UX.

Joe goes to get some ice down the hall from their room, and stumbles into a Macintosh unveil.

The whole scene is pure visceral technology as computer is turned on and it says:

"Hello, I'm Macintosh."

Joe reaction though is what makes the moment as he clearly is overwhelmed and realizes how much the computer can be, and its captured so brilliantly by his only spoken words in the moment : "It speaks".

Clawdbot has that vibe. It takes llm's and marries them to chat rails and personalization to complete the dream Siri was supposed to be. Yea, sure its comes with all kinds of security concerns, but for a brief moment we can all dream about what open source and unlimited compute can do for us as humans.

The Opensource community is adding skills for Clawdbot faster then any company will ever be capable of doing because, well, customization with intelligence is essentially limitless. Remember when I criticized the Rabbit R1 after it won at CES for essentially making a companion voice integration skills activator for some of your apps? My pushback was this will no brainer in the os or in the cloud very fast, and here we are. But for now adoption will remain limited to the diy crowd. Whether your buying a macmini or setting up in vms; the whole thing has some kodi/napster vibes to it. But unlike those moments which were IP skirting driven, one can now very clearly start to think about the compute limitations here.

The already highly subsidized compute of the frontier labs nowis going to become an obstacle. Power user throttling is gonna be a big deal and figuring out how to get these assistants to not consume $300 in tokens for something a human can do in 3 seconds is gonna be a thing too. Will the broad model of subsidizing consumption to stoke adoption continue to work or is Ai about to get a lot more expensive for those who can afford 24/7 assistants perfectly customized to one self.

Why can't I pay 2k a month for a managed service provider to deliver me the perfect digital assistant for all facets of my life from controlling my home to handling work tasks? I want security and reliability, and a 3rd party to handle all this for me. That is obviously coming.

So, time to think about who wins in this world..

COMPUTE SCARCITY: TIE GOES TO THE PLATFORM

If you been trading memory stocks on AI DC you probably have given some thought to why hyperscale/nvda didn't see this shortage coming and invest accordingly. Seeing as Jensen told us 18 months ago we would be short compute forever; why not address the memory bottleneck with acquisitions/investments? And the answer to that is probably because it is an entp biz that is limited by their own qualification cycles etc. I.e. the surge in commodity prices don't matter as much to them as their own contracting of supply for the long-term is what is causing that in cyclical commodity end markets.

Which means when someone tells you everything is sold out till 2028, it's kind of hard to get excited about compute demand upside as an earnings driver. When the constraint is physical supply (compute, memory/storage, cooling, power), incremental demand mainly shows up as higher prices / rationing, not unlimited earnings growth. The alpha at that point is in owning the layers that turn a unit of compute into more units of user value via routing, caching, model choice, tool orchestration, safety/privacy, and distribution.

Google has had this going for them and now I think Meta gets a new narrative look along these lines.

Meta: The Purpose Built Jarvis Platform?

I have discussed the Meta Ai investment cycle as essentially a defensive positioning against potential social lost engagement to Ai assistants/companions. One can already see Duobao's success in China and we all generally have known where Meta will need to go with this. But them being behind on intelligence has meant that their own MetaAi chatbot isn't as compelling engagement wise. Clawdbot potentially flips the Meta strategy script going forward. If Chat is the best UI interface for humans, then WhatsApp/Instagram can be the default rails. All you need is for Meta to embrace third-party assistants, and it can subsidize the compute cost of the experience w its core (ads, engagement, commerce) and capture the identity graph and data exhaust. So, far they don't appear to be going this way, at least at the entp level w whatsapp for biz they are closely guarding the rails. But if they were to do that here, they would just make another chat platform the default ux.

And this shit isn't simple. Someone is going to need to manage the identity/security element of all this for it to truly be scale capable and meta could deliver on that. Apple for sure is going to try and do the same likely with google in tow, but Meta has a huge advantage here with Whatsapp. And from a stock perspective the narrative pivoting in that direction looks great at that valuation as victory would probably 3x the company value, so there is a lot of room for discounting.

Cloudflare: Monetizing Compute Scarcity

First company you gonna think about infra wise in a compute constrained world with custom AI assistants for everyone is Cloudflare. Clawdbot workloads are bursty, latency-sensitive, and not very compute efficient at the model layer. Cloudflare sits in front of any model provider and monetizes immediately via: AI Gateway (routing/observability/cost controls), rate limiting, caching, and edge execution. That’s a AI infra tollbooth on every request, and it directly turns scarcity into margin by optimizing utilization, not fighting for HBM and GPU allocations.

I imagine Prince is gonna be giddy over this on the next earnings call. The only issue with Net is its so pricey. For me this is a bit interesting as I been bearishly biased on the high valuation security/infra saas names for the past several months, but what's kept me in check on a aggressive cloudflare short has been the compute scarcity environment tailwind, which I know all too well from the long side of it for most of 2025. I just been worried at some point this could get stupid momo popular again because of where they sit in the stack. My guess is it catches another wave of interest here on that.

As for the infra guys, yea this is secular great as everything really been everyday for them for past few years. But like many experienced in covid, when you can't make more to sell supply chain limitations become the bottleneck for multiples. Cause they slow things down long enough for actual meaningful changes to happen which basically means they become the dominant risk factor intermediate wise as the problem is supply not demand visibility.