{"slug": "is-software-dead-no-it-just-got-a-lot-harder-to-win-the-saastr-ai-deep-dive-with", "title": "Is Software Dead? No. It Just Got a Lot Harder to Win. The SaaStr AI Deep Dive with Rory O’Driscoll", "summary": "Veteran software investor Rory O'Driscoll told the SaaStr AI conference that software is not dead, but winning has become much harder. He noted that hyperscalers will spend roughly $688 billion on AI capex in 2026 against only $110 billion in revenue, a gap that won't close until around 2032. The bet relies on AI capturing a significant share of the knowledge worker wage bill, with potential market disruptions along the way.", "body_md": "You already know Rory O’Driscoll from our weekly 20VC x SaaStr podcast, where he, Harry, and I go back and forth on exactly this stuff. He’s also been investing in software for 30+ years. Long enough, as he put it on stage at SaaStr AI, to have stayed employed but not so successful he could retire and buy a football team. So when the whole industry starts asking whether software is dead, he takes it a little personally. Then he does what a good investor does: narrows the question until it’s answerable.\n\nHis answer, after weeks of his firm arguing about the numbers every Friday: software isn’t dead. It got a lot harder to win, but it isn’t dead. The math is below.\n\n## AI Is in “Invest Mode,” and the Gap Is Enormous\n\nOne fact colors every other decision right now. In 2026, the hyperscalers alone are spending roughly $688B on AI capex. Think of that as all the money going in to make AI happen.\n\nWhat’s coming out the other side is about $110B in revenue. Around $89B of that comes from the two foundation model leaders on a GAAP basis. Round up everything else and you get another $20-30B.\n\nSo the industry is spending on the order of half a trillion dollars a year more than it’s taking in.\n\nRory’s framing: roughly right beats exactly wrong. Argue about $10B here or there and nothing changes. The shape is a massive bet funded years ahead of the revenue, and every strategy has to account for it.\n\n## It Takes Until ~2032 for Revenue to Catch the Spend\n\nRun the revenue estimates forward and the crossover, where the two model leaders finally out-earn cumulative capex, doesn’t hit until roughly 2031-2032, at around $1T of revenue.\n\nTwo takeaways for founders from that timeline:\n\n**One, this goes on for years.** We’re five or six years from being out of invest mode. That’s a long runway of capital pouring in ahead of returns, and a long runway of opportunity on top of cheap, improving models.\n\n**Two, expect a hiccup.** When you’re spending far more than you’re taking in, someone eventually wakes up and asks why. Even if the long-term story plays out, don’t be shocked if the market takes a hard breath and pulls back before 2032. Plan your burn like that moment is coming, because it might.\n\n## Why This Is Even Possible: The Knowledge Worker Wage Bill\n\nFor the foundation models to earn a trillion dollars, the money has to come from the largest pot there is: the knowledge worker wage bill.\n\nTo hit these numbers, AI has to take an appreciable slug of it. If all the spend landed in the US, you’d be talking about something like 15-17% of all knowledge worker dollars. For software developers specifically, well north of 25%. Put concretely: for every $200,000 developer, roughly $50,000 going to tokens.\n\nThat’s the bet American capitalism is making. The scaling laws showed that money in produces better models out. The ChatGPT moment showed people actually want to use it. Combine those two and the capital keeps coming. But it only pencils if AI eats a real share of what is currently paid to humans. Possible, maybe even likely, but not without bumps.\n\n## The Stack: “Making AI” vs. “Using AI”\n\nBorrow the model Jensen uses: energy, chips, infra, models, apps.\n\nThe bottom three layers are all about **making AI**. This is the stuff it takes to build $688B of capex, and roughly 80% of it isn’t venture-backable. Half of it is chips, and the vast bulk of those dollars went to one company. Capex went from about $200B four years ago to about $600B today, which means someone is pocketing roughly $400B a year, mostly Nvidia. Energy is a big, hard block but a relatively small dollar amount. Infra ranges from plain buildings up to the labeling companies and RL training gyms the model companies buy.\n\nThe top layers are **using AI**: how software takes a model and turns it into value for a business customer. That’s where the rest of us live.\n\nThe money question: how does enterprise want to consume ~$1T of AI over the next five or six years? Do they buy it all direct from the foundation models, in which case we all go home and rent a very small conference room? Or is there room for a whole set of companies to buy AI from the model providers, add real value on top, get close to a specific customer, and ship an economically differentiated product?\n\nRory’s read, and mine: there’s room. What defends that room is the next question.\n\n## The Harness Is the New LAMP Stack\n\nThere’s a term going around, “the harness,” for the software layer on top of a raw model that turns it into a dependable system. Deciding what context the model sees, allowing actions, selecting outputs, logging what happened, escalating and managing models by task.\n\nRory doesn’t love the word, and neither do I. The better mental model: for 15 years we built B2B software on the LAMP stack. This, or something like it, is going to be how you build every application in the age of LLMs.\n\nEvery SaaS app ran on the same LAMP stack, and they were not remotely the same app. Same thing here. Most AI apps get built the same way, reasoning over a model and distilling it into something a business user can use, and they’ll all describe themselves the same way. But how they get instantiated varies enormously depending on whether you’re a legal product, a customer support product, a coding product, or something with hardware attached. Sameness of architecture is not sameness of business.\n\n## The Moats That Actually Hold\n\nThe real question underneath all of it: what can Claude or ChatGPT roll over, and what can’t they? Rory’s firm has been sorting this into moats. The ones that hold:\n\n**Software plus sensors.** If your product combines vision, touch, or other real-world sensors with reasoning to drive a business outcome, the foundation models aren’t going to start shipping sensors. Eminently defensible.**Marketplaces and network effects.** If you’ve built a platform where every side wants to be there and more users make it more valuable to everyone, a single person with a Claude Code app can’t recreate that. A recruiting marketplace they recently backed is a clean example.**Proprietary, non-public data.** If your business is built on data the model has never seen and can’t see, you don’t even have to explain the moat. The model simply can’t do what you do.**Full-stack.** Instead of selling software to a business, you become the business. Rory has a portfolio company doing wealth management with LLMs. OpenAI isn’t getting into wealth management. They’ll hire wealth managers for their own engineers, but that’s different.\n\nThen the two hardest, closest to the model providers, where you’re still fundamentally a software company:\n\n**Data flywheel.** You start with a simple app, and as you learn how your users actually work, you compound into a specialist, differentiated product over time.**Forward-deployed engineers.** Sometimes the software only works with people on the ground to capture context. That can differentiate you, though note the foundation model companies are signaling they’ll do this too, partnering with PE on one hand and making clear they’ll build it on the other.\n\nThe takeaway: there are subtle architectural moats and obvious business-model ones, and they point the same way. The model companies start with a trillion dollars and will be fine. There’s still room for everyone else.\n\n## Three Versions of “Is Software Dead”\n\nThe phrase means three different things. Keep them separate.\n\n**Version one: will the foundation models take it all from the new AI companies?** Nobody knows, and most VCs are betting both ways: writing checks into Anthropic at a few hundred billion pre, while also funding companies that only make sense if there’s room outside the model layer. You have to form a theory of the case on what survives.\n\n**Version two: is plain vanilla SaaS dead?** For companies that just automate a workflow, this is the real threat. Move and survive, or get rolled.\n\n**Version three: what happened to everything built before 2022?** This is the question LPs ask. Rory’s rough take on a large pre-GPT portfolio:\n\n- About\n**10% was effectively dead on arrival** when ChatGPT shipped, because they’d solved an AI problem that suddenly cost $5 per million tokens instead of $30M to build. - The remaining 90% splits into rough thirds:\n**A third is insulated.** Different enough, often with data effects or predictive AI, that GenAI is orthogonal to it.**A third is additive.** You bolt AI on top of a strong core product and increase its value. Think a drone management product adding AI-driven safety and progress reviews. No way AI replaces the underlying value.**A third is flat-out threatened.** The plain vanilla SaaS bucket. Quick or dead.\n\nAnd yes, they may have been squinting a little optimistically on that last third. Be honest about which bucket your company is in.\n\n## Do You Need Your Own Model?\n\nOne way to compete with the foundation models is to own a model yourself. Sorting through a long list of ~65 “neolabs” produced 10-15 categories of non-pure-foundation-model bets.\n\nThe clearest are the frontier LLM replacements going straight at the next generation. More interesting for founders: voice, video, and image all had model options that weren’t pure LLMs, and far less compute- and capital-intensive than a foundation model. ElevenLabs built a real business on a speech model. Coding-specialist models, chemistry and physical-science models, and more all count. You don’t need foundation-model money to own a defensible model in a narrower domain.\n\n## The Multiples Reset Is Real, and It’s a Growth Story\n\nThis hits fundraising directly. Public software multiples crashed, and they crashed for a reason: the growth rate collapsed. The average company used to grow ~30%. Now it’s closer to 10, sometimes under. Wall Street isn’t being irrational. Companies are getting marked down because they stopped growing.\n\nThat doesn’t change the compounding math, but it changes the price. If you’re compounding out to a 4x revenue multiple, it’s harder to make that deal work than it was even last November. Everything eventually trades at a rational, growth-adjusted multiple, so you can’t pay up the way you used to.\n\nSo category conviction is everything. If you think your company blends into the undifferentiated mass of low-growth software, there’s no point doing the deal. A 3x on the way to a 2x on the way to a 1.5x is on the way to a 1x return. Some deals Rory looks at and knows he’ll make money. Others, in his words, might just be the white noise of mediocre.\n\nOn the triple-triple-double-double question: 10x year-on-year is unprecedented for software, and all else equal you want the faster company. But read that signal carefully. Some hypergrowth is unnatural adoption, driven by a corporate mandate to buy AI now, or by users who’ll get fired if they don’t burn tokens. A category leader tripling nicely and boring as crap, at an attractive price, can compound for a decade the way great SaaS always did. The split: some categories have a huge corporate itch to scratch (legal, where there was nothing before) and grow explosively. Others ship an obviously better next-gen app customers will buy, just without the board-level urgency. That one won’t spike, but it compounds.\n\nIf you’re not going to 5-10x in the hypergrowth phase, you’d better be capital efficient, category-leading, and differentiated.\n\n## Compute Intensity: The Standard Deviation Went Way Up\n\nThe single biggest change for an investor is the spread, not any one number. Every SaaS company used to look roughly the same on cost structure. Now you walk into one company committing a billion dollars to cloud, and the next one is spending 10%.\n\nRough map of where LLM/compute cost lands as a share of the business:\n\n**Model companies:** 70-80%**Coding companies:** 50-60%**App companies:** often ~10% of COGS on LLMs, building the other 90% of value on top**Traditional B2B adding AI features:** roughly 8-20%; if you’re at 20 you need to get more efficient, target 5-8\n\nAn app company spending 10% isn’t competing on compute. It’s using a slice of revenue for the model and building its differentiation everywhere else. That’s a healthy place to be.\n\nOne heuristic to keep: AI can be the new sales and marketing. If the product is so good it sells itself, leaps into users’ hands, and grows on its own, you can afford to spend 60% on compute because you’ve gutted your go-to-market. But the product really has to sell itself. If it doesn’t, and you’re still carrying 700 people in sales and marketing plus a 50% compute bill, you’re dead. You get compute that sells itself, or people who sell it. You can’t fund both.\n\n## Where This Leaves Founders: It Got Harder. That’s The Job\n\nSoftware isn’t dead. The bottom of the stack, making AI, is the biggest spend in the history of venture and mostly not yours to win. But just as Microsoft owned the client-server era and hundreds of apps companies still went public on top of it, just as AWS owned cloud and a generation of SaaS companies won on top of that, the foundation models will own their layer and there will be room for a large set of software companies on top.\n\nThe job now is to know exactly which layer you’re on, which moat you’re actually building, whether your growth is real or borrowed, and whether your product sells itself. It got harder. It got trickier. It isn’t dead.\n\n## Top Mistakes Founders Make Right Now\n\n**Pitching a pre-2022 company as if it’s brand new.** You can’t pretend you just started, and investors see through it instantly. Drop the SaaS framing, show an agent-first architecture, and prove the new product is the thing growing.**Growing at old-SaaS speed against AI-native peers.** If the category has AI-native competitors putting up 5-10x, a “healthy” 2x looks like falling behind. Know what growth rate your specific category demands before you set the plan.**Being token-inefficient with no differentiation to show for it.** Spending 40% of revenue on the model while building little value on top is the worst of both worlds. Aim to keep LLM cost a slice of the business and put the rest into what actually differentiates you.**Trying to fund heavy sales and marketing and heavy compute at the same time.** You get a product that sells itself, or a team that sells it. Carrying 700 people in go-to-market plus a 50% compute bill is how you die.**Paying up on the assumption multiples hold.** Public software multiples reset because growth reset. Underwriting to a 4x revenue multiple that no longer exists is how a 3x round becomes a 1x return.**Building without category conviction.** If your product blends into the undifferentiated mass of low-growth software, no amount of execution saves the return. Have a real reason your category compounds, or don’t do the deal.**Reading borrowed adoption as real demand.** Some AI growth is unnatural: corporate mandates to buy now, or users who get fired for not burning tokens. Separate durable pull from a temporary itch before you extrapolate the curve.**Solving a problem the models now do for $5 per million tokens.** A meaningful share of pre-GPT companies were dead on arrival because their hard-won capability became a cheap API call. Ask honestly whether your core is now a commodity primitive.**Not knowing which layer or moat you’re actually on.** Founders who can’t say whether they’re insulated, additive, or threatened tend to be the threatened ones. Name your bucket, name your moat, and build against it deliberately.", "url": "https://wpnews.pro/news/is-software-dead-no-it-just-got-a-lot-harder-to-win-the-saastr-ai-deep-dive-with", "canonical_source": "https://www.saastr.com/is-software-dead-no-it-just-got-a-lot-harder-to-win-the-saastr-ai-deep-dive-with-rory-odriscoll/", "published_at": "2026-07-15 18:48:48+00:00", "updated_at": "2026-07-15 19:09:40.713991+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-infrastructure", "ai-policy", "ai-startups", "ai-products"], "entities": ["Rory O'Driscoll", "SaaStr AI", "Nvidia"], "alternates": {"html": "https://wpnews.pro/news/is-software-dead-no-it-just-got-a-lot-harder-to-win-the-saastr-ai-deep-dive-with", "markdown": "https://wpnews.pro/news/is-software-dead-no-it-just-got-a-lot-harder-to-win-the-saastr-ai-deep-dive-with.md", "text": "https://wpnews.pro/news/is-software-dead-no-it-just-got-a-lot-harder-to-win-the-saastr-ai-deep-dive-with.txt", "jsonld": "https://wpnews.pro/news/is-software-dead-no-it-just-got-a-lot-harder-to-win-the-saastr-ai-deep-dive-with.jsonld"}}