Karthik Vellanki

Full announcement: ontaurus.com/blog/raising-4…

We just raised $4.3M from @generalcatalyst with participation from @gokulr, @toddgoldberg, @rahulvohra, @aarthir to build an AI-native consumer brands company that owns, launches and operates brands end-to-end. Consumer brands are still built with fragmented teams, slow feedback loops, and manual workflows. We’re replacing that with a unified stack of AI agents that owns and operates our brands. We’re building the AI-native Procter & Gamble - powered by agents, not org charts. If you want to solve hard agentic problems and move atoms at the speed of bits, come join us. It’ll be fun!

@hahnbeelee grep works well for coding agents because you’re generally looking for function names which are precise. Outside of coding, semantic search still has a lot of utility.

semantic search is dead. file-based context retrieval is far better. let us know what you think and if there's more you want to see from our AI assistant

@evanjconrad @HotAisle “There’s always a buyer” ≠ Risk Management. Markets will clear even in a downturn but doesn’t mean every seller can stay solvent. I see how SFC is valuable by making it easy to find compute, but I don’t see it covering risk. Airbnb is a great marketplace but doesn’t cover risk.

@HotAisle there will always be someone who will buy it at some price, but if you buy without sfcompute, then there's no other option for liquidity at all

What SF Compute does. When you finance a GPU cluster, you need to get an "offtake" agreement. Basically, someone has to agree to rent the cluster from you, typically for a 3+ year period. If that agreement falls through (the person fails to pay), then the person who owns the cluster gets wiped out, and their lender ends up with a bunch of GPUs, rather than say, money. It really looks like the world is deploying more capital into the AI build out than any infrastructure project in the history of the world. You remember when people said there was going to be a Manhattan project for AI? The current build out is the size of 20 Manhattan projects. We’re so far past the Manhattan project it’s not even funny. This is the cost of a war. It would be really bad if that scale of capital was secured against offtake agreements (long term contracts) with application layer companies who turn around and sell to their customers on a month to month basis. If the AI SaaS has a bad few months, can the AI SaaS continue to front their compute bill? They could in CPUs, because in SaaS you might have a company with $20m in the bank, and has a $1m/year "CPU" bill. But in GPUs, you have a startup that raised $20m, but a $20m+/year compute bill. So a small shift in demand means lights out for your business, because the products are so levered. That works as long as you can plug the gaps with venture capital & high margins. But across the board, AI applications are lower margin than their SaaS counterparts, giving them less buffer to save them in a bad month. And even in a hot market, venture capital won't necessarily save you if you're running unprofitably with a massive liability. That’s the problem we solve. We let people buy long term contracts they can “exit”, by selling back. That lets them get liquidity in the most critical moments, ensuring they turn a profit rather than a loss on tight margins. In other words, we prevent a bubble. When we do that, it opens up blocks of compute for smaller use cases too, like academics or startups. When we started, we were "Junelark", a 2-person audio model company that bought too big of a cluster. We had bought 12 months, but could only afford 1 month. To avoid bankruptcy, we had to sublease the other 11 months by acting as GPU brokers. Our audio model company was forced to pivot or die because we didn't have liquidity. To make SF Compute, we split the company down the middle. One side of the house makes a billing company, a ledger, an order book, and a compliance program. The other side makes a systems engineering company. To make this work, you need to run the clusters. So we make the low level cloud stack that interacts with BMC (Redfish & IPMI), UFM, built a UEFI app that replicates PXE boot in weird environments, and a virtualization layer kind of like EC2. It’s a massively complex machine filled with nitty gritty challenges. Today, we’re growing faster than Cursor and we’re scaling to secure the risk of the largest infrastructure build out in the history of the world. We’re hiring across the board for rust programmers, systems engineers, and GTM, and we’d love for you to join us to prevent an AI bubble.

@tanayj @thinkymachines Second order effects of Nvidia having such insane margins.

Meta poached Andrew Tulloch from @thinkymachines after he initially turned down a $1.5B offer. One way to rationalize the comp: Meta will spend ~$75B+ in CapEx/yr. The 10–15 core researchers they're poaching are the hedge-fund managers of that compute. They get paid "management fees" like numbers in compensation to maximize ROI on that capex.

@solarizid @typedfemale Hiring TPU kernel engineers definitely doesn’t mean they’re losing money on inference. Every AI lab has kernel engineers, it’s just table stakes.

just another HN thread with nearly 400 comments in which no one points out any of the article's many blatant mistakes

@mayfer This seems to be conflating representation with the actual number? The representation of irrational numbers maybe a function but the number itself is a fixed and single value. The hypotenuse of a right triangle with two sides of length 1 is very much a fixed value.

irrational numbers aren't "numbers" they are functions they aren't defined by anything other than infinite series. they are result of conceptually derived tools of infinitely expanding levels of precision of PROCESSES, not countable things that exist in stationary fashion

@tszzl This isn’t a great analogy. Ads manager doesn’t have any value by itself, even if didn’t exist at all, media buyers would physically call to buy ads on Meta. If ads manager stopped growing and became culturally irrelevant, people would still use it.

the greatest apps in the world have a monopoly chokehold over a single highly valuable market function and everything ancillary just doesn't matter. the facebook ads manager has been an impossible to use POS for years, doesn't change, drives 200 billion in revenue

@Austen Not to sound pedantic, but low level coding still exists in that scenario. Even if you abstract away everything, the instructions sets to do matrix multiplication is still code.

wtf

@skdh @bernieblume Shouldn’t theoretical physics be before math? Math is more fundamental than physics.

@bernieblume Theoretical physics is next basically

I find it mildly amusing how mathematicians are currently clinging onto the hope that there is something special about their creativity that AI will not also be able to pull off in a year or two.

@lulumeservey He gets criticized for inaction but I’ve never heard a compelling argument on what “exactly” he could have done differently. He could have managed Wall Street better but there’s no way Larry/Sergey/Zuck/Satya would have even attempted to disrupt the greatest cash cow ever seen.

@deedydas @JeffBrines These are all very good when you start from scratch and want to generate a video, but they fall short when you want to maintain continuity and consistency with characters/scene. And there’s really no concept of “editing” yet - they regenerate the entire video again.

@JeffBrines Google Flow (w/ Veo) Runway Invideo Flora All come to mind. I imagine the video gen players also want to get into the space

Chinese researchers just dropped an image gen model that does Photoshop-grade edits without affecting the rest of the pic. "Make her dress blue" "Make him smile" (multi image) "Put this jacket on him" GPT 4o was always bad at this. It's OmniGen2. Open source. Photoshop killer.

@dnlkwk M&A, and more specifically the serial M&A you need for roll ups is not trivial. There’s a huge possibility of adverse selection where you’re only able to acquire bad companies or overpay. The well run companies with a high NPS are equipped to infuse AI and scale on their own

@andrewmccalip You need both a high tolerance for risk and a strong, centralized regime to build mega-projects. Think pyramids, Trans-Siberian Railway, Panama Canal, Three Gorges Dam. Even the Hoover Dam cost 100+ lives - something no modern democracy would politically tolerate today.

Contrarian take? Most engineering now moves slower than half a century ago—every boost in efficiency got bartered for risk reduction.

@gauravvohra > Away from laptop for 10 mins > Repeat security theater We don’t give Google enough credit for making auth nearly seamless and incredibly secure. Every other auth system punishes you with bad UX to pretend to be more secure.

Signing into stuff in 2025: > Sign in > Forgot password > Check email > No email > Try Google auth > Hooray, works > Is this really you? > Grab iPhone > Open Youtube app > Yes it's me > Okta sign in required > Enter passkey > Passkey expired > Need backup code > Open password manager > Signed out > Enter password manager password > No backup codes > "Try another way" > QR code > iOS Okta app > Sign in > Face ID > Is this really you? > 2FA SMS code > Your account needs another layer of security > "Do this later" > Your org admin requires Passkey > Set up Passkey > Sign in > Congrats you finally got into Salesforce or some shit

@ludwigABAP Since the moon landing, infant mortality rate dropped 73% and life expectancy is up by 15 years - which surely counts as progress outside the world of bits. Stagnation narratives are driven by unmet sci-fi dreams (flying cars, jet packs). We just progressed on different vectors.

he’s also deeply obsessed with the concept of acceleration or fast movement, arguing we’ve been stagnating for 70~yrs (in which we’ve made more progress than the entirety of the rest of history in most fields) there seems to be a confusion about breadth of research vs depth, and some looking down at depth bc of lack of understanding of how it contributes to depth pre-19th century scientific discoveries were more disruptive bc we knew less, and we added breadth over time, we shifted to more and more and more depth but only the blind doesn’t see how the depth in one field tends to bridge to other domains and actually clarify the whole of reality? or am I tripping and I just don’t like Thiel

I’ve listened to Peter Thiel for 5mn and he’s already said: - CS was basically people who flunked out of mathematics / physics / EE - how it’s weird that the field ended up working out - he immediately clarifies that it worked on the scale of people making some “fantastic companies” and some “important social transformations” this is not the history of CS at all? the field was literally created by brilliant mathematicians, EE folks etc what’s his beef with modern times, computer science etc? what’s with his obsession with stagnation?

@ptsi @haridigresses @altcap a) probably doesn’t apply in this deal right? I don’t think Thrive is struggling to get into any deals. If anything, their “loss leader” deal was OpenAI which allows them to get into any deal they really want to.

Mostly agree right back at you, Hari! 😀 But two differences: For investors, I can totally see 2 things working for this deal: a) Perfectly legit to construct a portfolio where Cursor is your loss leader, but it lets you on the margin get into a bunch of deals you otherwise would not get allocation to b) Especially in a world where you could argue that ALL hot AI deals are overpriced and the play is more about AUM maxxing than DPI maxxing 🙃 As a founder, yes high valuations are often risky. And founders forget that too often. But they can also let you do unique things like… e.g. buy Vercel or GitLab or Figma and break out of the same competition as everyone else. Yeah, take some secondary, but otherwise — my bet is it’s worth playing it big. :-)

I know lots of eng teams already switching away. An ARR multiple of 66x would be reasonable for the growth rate; but this isn't ARR, it's more like pilot revenue (what @altcap would call ERR). I hope the Cursor founders, employees and early investors are taking some secondary.

@celinehalioua @lisabonos @washingtonpost Noob question: what’s the expected increase in lifespan (either in number of years or avg lifespan)?

Loyal’s drug for senior dog lifespan extension drug LOY-002 has completed its FDA efficacy package (RXE) We are on track to hopefully bring the first longevity drug to market *this year* Thanks @lisabonos + @washingtonpost for the exclusive

@signulll Counterpoint: they “lost” the shopping walled garden 2 decades ago to Amazon, but product search is still a massive business. They own YouTube and Meta allowed Instagram to be indexed. Google can’t/won’t build creator ecosystems on all paradigms but it’s not a lost cause.

google’s core problem is that it was built to organize a web that no longer exists. the open web has been replaced by walled gardens, discord servers, newsletters, private forums, & algorithmic feeds that are never exposed to search. worse, the visible parts of the web that google still indexes have been overrun by seo-optimized sludge, ai-generated spam, & paywalls. their dna is fundamentally extractive. they never built a creator ecosystem because their whole game was to scrape, index, & serve ads against other people’s content. the entire ecosystem slowly but surely shifted drastically—with llm’s anyone can organize anything so the mission breaks down.

@chamath @levie Forking android would have made no difference. They wanted to reach the Apple consumer base in ways Apple wasn’t permitting them to do. Apple consumers are not switching to a fork of android. They can do anything they want on Android now (without worrying about anti trust)

The biggest strategic error Meta made was not forking Android and building their own mobile OS and launching their own phone. Had we (I say we because I tried, unsuccessfully, when I was there) done that it would have slowly but likely created a three horse vs two horse race for the most important compute platform today - mobile phones. Considering how much money has been poured into VR/AR subsequently - essentially trying to create a different compute environment - it’s clear they have realized this. Assuming AI is as consequential as we all think for them to not do it and end-run Apple, Google and Microsoft would be yet another huge mistake when looking back a decade from now. If you own a plurality of the underlying substrate model and the compute environment, it will be much easier for them to launch disintermediation hardware a few years from now.

Meta should launch an AWS-style cloud for AI. It’s the smartest next step in consolidating mindshare around Llama + PyTorch by subsidizing compute and developer mindshare. They will need a lot of heterogeneous silicon to do it properly. A mix of build and buy can get them there fast.