Newsletter · · Ashutosh Agarwal

Wall Street Debates Whether SaaS Is Dead as Anthropic Sets the Growth Bar - Weekly SaaS / Software Podcast Recap - Week of August 23, 2026

A plain-English digest of what the software and SaaS podcasts argued for the week of August 16 to 23, 2026, from the is-SaaS-dead debate and the death of seat-based pricing to bull and bear stances on Microsoft, Oracle, and the rest of enterprise software.

Weekly SaaS / Software Podcast Recap

Week of August 23, 2026: Wall Street Debates Whether SaaS Is Dead as Anthropic Sets the Growth Bar

This is a plain-English digest of what the podcasts said this week about SaaS (software sold as a subscription), application software, and the "infrastructure" software underneath it: cloud, databases, security, developer tools, and data platforms. A quick glossary for terms that keep coming up: seat-based pricing means you pay per user; consumption (or usage-based) pricing means you pay for how much you actually use; NRR/NDR (net revenue retention / net dollar retention) measures how much an existing customer's spend grows or shrinks year over year, above 100% means customers spend more over time; a system of record is the software where a company's official data lives (its CRM, its HR system), which is historically very hard to rip out.

Top of mind this week

  • The "is SaaS dead?" argument was everywhere, and it split cleanly. Software founders and operators mostly said the panic is overblown as long as you have a real moat; several investors said the opposite, that AI competition drives prices toward zero and there are "no growth assets" left.
  • Seat-based pricing is being pronounced dead by the people who build billing systems. The new consensus is hybrid models: a recurring base plus usage, or prepaid "credits." As one guest put it, "the unit of work is no longer a human."
  • The gross-margin question got specific. Multiple guests said the classic 80% software gross margin drifts to roughly 65–70% (some said as low as 25–60%) once you're paying model providers for AI usage inside your product.
  • "Open source is going to win it all" was the loudest single claim of the week (Jason Calacanis), triggered by legal-AI startup Harvey building its own model. The counterpoint: frontier labs still own the hardest tasks and "deserve those economics."
  • Anthropic's numbers dominated the tape of company chatter, described as the fastest-growing software company in history, alongside a genuine debate about whether it has now passed OpenAI, and what a ~$2 trillion IPO would imply.
  • On individual stocks, the clearest bear calls were Oracle (a "proxy for OpenAI") and Microsoft (its pricing shift is "aimed like an arrow to the heart" of its own model). The clearest relative bull was ServiceNow over Salesforce.

1. Dominant Themes

The "SaaS-pocalypse" debate moved from theory to stock prices. Nearly every software podcast referenced the idea that AI lets companies "vibe code" (build software with plain-language prompts instead of engineers) their own tools and skip paying for SaaS. On Software Spotlight (Aug 19), Biller Genie founder Thomas Aronica argued the narrative is "mostly wrong, but only if you know what your actual moat is," because "people pay for convenience… you could pay $50 a month to Anthropic and build something yourself, but then you have to maintain it… or you could pay $300 a month to someone that just runs it for you." His filter: features, code, and clever UI are getting commoditized fast; distribution, proprietary data, domain expertise, and switching costs are not.

AI capex spilling into the software conversation, as both threat and opportunity. The macro shows framed the roughly $1 trillion-a-year hyperscaler build-out as a risk that eventually pressures software. On Bloomberg Tech (Aug 20), Martin Orton said hyperscaler AI spending creates "existential risk" for some companies but also "an opportunity for investors to sort which companies will survive." Several credit and macro shows flagged the financing side: The Banker Next Door (Aug 20) cited nine tech companies carrying ~$3 trillion in off-balance-sheet commitments (mostly AI-related) growing faster than reported capex.

The death of seat-based pricing. On Run the Numbers (Aug 20), Stripe/Metronome's Wisam Hirzalla was blunt about which pricing models are gaining share: "Certainly not seat-based. Certainly not seat-based." The winners are hybrids (a recurring base plus usage) and "credits with auto top-up." His reasoning: "It used to be seats were a really good proxy for value… you could charge one user $20 a month and they run five prompts… another user the same $20 and they run hundreds of thousands of prompts… the unit of work is no longer a human." He noted AI-native companies now change prices more than three times a year.

Margin compression is now a number, not a worry. On The Information's TITV (Aug 20), PwC's Dallas Dolan said that as SaaS platforms embed model costs, "if 80% was the old gross margin that software companies loved," it likely settles "closer to like a 70 or a 65." A CFO on Recruiting Future (Aug 18) went further, saying traditional 80%+ SaaS gross margins are being replaced by "AI services with 25–60% margins and variable consumption-based costs."

Open source vs. frontier models, and enterprises building their own. This was the week's hottest infrastructure theme, sparked by Harvey (see Debates and Names below). The through-line: big customers increasingly don't want to hand their proprietary data to a frontier lab, so they route routine work to cheaper open-weight models and keep sensitive work in-house.

Anthropic's growth as the reference point for all software. On FYI – For Your Innovation (Aug 19), ARK's Sam Korus framed it vividly: "Salesforce is a roughly $45 billion run-rate revenue business and they are over 25 years old. And Anthropic added more than a whole Salesforce in run-rate revenue in less than a year," going from ~$9B at the start of 2026 to a rumored $70–80B run rate.

Agentic security and identity as a new software category. Multiple security shows argued that AI agents acting inside enterprise systems create a fresh identity/access problem. On the SourceForge Podcast (Aug 21), Auth0/Okta executives said agents must operate with "fine-grained, runtime-controlled permissions rather than broad access," pitching "agents as principals" products; the Cloud Security Podcast (Aug 18) described an "agentic control plane" for shadow-AI sprawl.

2. Key Debates

Is SaaS dying, or just changing? Software operators vs. macro investors.

  • "Mostly fine, moats survive": On REWORK (Aug 19), Basecamp's David Heinemeier Hansson argued most small and mid-market customers won't rebuild their own software with agents because of the operational burden. The real "danger zone" is high-end enterprise SaaS: "the kind of companies who might pay Salesforce $600,000 a year for a system that they hate," citing Shopify building its own CRM layer because "what they were using was crap." Biller Genie's Aronica and even the hosts of The Chad & Cheese Podcast (Aug 21), discussing the rumored Silver Lake/Workday buyout, reject the "SaaSpocalypse," calling depressed valuations a buying opportunity.
  • "No growth assets left": On The Pomp Podcast (Aug 22), Jordi Visser made the bleakest case: hyper-competition from AI drives "prices to zero… There are no growth assets in my viewpoint anymore." He pointed back at software's Q1 sell-off: "Remember what happened to software companies in the first quarter of the year and how it was spreading like a disease across the market… It's happening in the S&P 500 because multiples are compressing." A guest on Something More with Chris Boyd (Aug 20) warned of a 2027–2028 refinancing crisis for private-equity-backed software credits as valuations compress.

Seat-based vs. consumption pricing. Metronome's Hirzalla (above) and GitLab's CEO both say seats are giving way to hybrids. On Tech Disruptors (Aug 22), GitLab CEO Bill Staples/Shuby Badrinath explained the split: seats stay for human builders (premium/ultimate licenses), but agent work gets its own consumption charge because "one engineer can start an agent or 100 agents… the work, the cost and the value scale independent from the seats." Notably, he pushed back on the "engineers are done" narrative: headcount keeps growing and engineers are "busier than ever."

Open source will win it all vs. frontier labs deserve the economics.

  • Open source wins: On This Week in Startups (Aug 21), Jason Calacanis declared: "Open source is going to win it all… the overwhelming majority of tokens in corporate America will not be on the frontier models." His trigger was Harvey releasing an in-house model built on the open-weight Kimi K3. His warning to founders about frontier partnerships: "There's no free pizza… If you partner with any of these people, they will slit your throat and take your business wholesale." He estimated Harvey was spending ~$10M/month with OpenAI and that such companies will move "99% of their spend" to open source, keeping only a sliver for "distillation."
  • Frontier still wins the hard stuff: On 20VC (Aug 22), investor Jerry (Fireworks/FirstMark discussion) argued open source is great for specialized tasks but "there's never been a time yet where the open source model has trumped a frontier model on innovation." He was bullish on the infrastructure layer riding on the labs ("Fireworks… four years to two billion") but skeptical of the app layer: "I'm not buying it for the legal and I'm not buying it for the app layers yet."

Do the model companies sell through SaaS, or eat it? On TITV, PwC's Dolan argued the model companies' most consistent revenue will come through enterprise software integration (e.g., Salesforce Agentforce) rather than direct sales, because enterprises want "consistent buyers" and system-of-record control. The opposing worry, raised in the same segment: the labs eventually build their own applications that "leapfrog the role of the existing enterprise software companies."

Anthropic vs. OpenAI, who's ahead? On More or Less (Aug 21) and across CNBC's shows, the consensus was that Anthropic's enterprise-first strategy is "crushing it" (roughly $11.5–11.6B in quarterly revenue, reportedly with positive operating income) while OpenAI's consumer-first path (~$40B ARR, wider losses) is now "catching up" in enterprise. The counter-view, repeatedly voiced: leads aren't secure because "model switching" is easy.

Is the AI trade a bubble that pressures everything? On 20VC, the framing was blunt, "half the neoclouds will die," with warnings that private-equity software portfolios levered "four to six times" are exposed if EBITDA drops and churn rises in a market dislocation.

3. Specific Names: Bull/Bear Stances From the Podcasts

Public software

Microsoft (MSFT). On Invest Like the Best (Aug 18), Ben Thompson gave a nuanced bear-lean: Microsoft's new E7 enterprise plan ($100/user/month plus usage overage) is "a fraught position," because the old magic was that a license was "a thoughtless revenue stream… untethered from headcount." Once customers watch a monthly usage bill, "you start thinking about what am I paying for… should I actually just start… spraying this out?" His bull case is the "IBM/middleware playbook," Microsoft as the dependable layer that helps enterprises adopt AI "without giving away the crown jewels," but he warned that agent products like Codex, Claude, and Cowork are "aimed like an arrow to the heart of what Microsoft has" (the user interface). Verdict: "Their strategy is sound. It's also desperate in an existential way."

Oracle (ORCL). Clear bear on Barron's Streetwise (Aug 21). Barron's tech writer Adam Levine noted Oracle's ~$640B multi-year backlog, but "almost half of that is one contract with OpenAI," making the company "a proxy for OpenAI." With sentiment on OpenAI down, negative free cash flow, rising debt and equity issuance, and Anthropic leapfrogging OpenAI, his bottom line: "I wouldn't touch it." (He still thinks Oracle is a genuine software winner in AI, the concern is the cloud/OpenAI exposure.)

Salesforce (CRM). Recurring bear-lean on displacement risk. On InvestTalk (Aug 19), Justin Clont ranked it middle-risk: "CRM… it's basically a spreadsheet in the back end," with earnings decelerating from +23% last year to ~+10% next year. REWORK's DHH used a hated "$600k/year" Salesforce install as the archetype of enterprise SaaS most exposed to being rebuilt.

ServiceNow (NOW). The relative bull on InvestTalk: "more deeply integrated into the process of big corporations," with earnings +26% last year, +16% this year, +23% next year, "relatively cheap" at ~24x forward earnings and "a more protected economic moat." Separately, ServiceNow's own podcast highlighted its 25-language localization edge.

Adobe (ADBE). Mixed debate on The Intrinsic Value Podcast (Aug 16): one host bullish on Adobe's creative-software leadership, ~10% revenue growth, and a P/E compressed into the low-20s/8x forward; the other bearish, having sold most of his position over AI commoditization of creative work.

Snowflake (SNOW). CEO Sridhar Ramaswamy made a measured bull case on Minus One (Aug 20): Snowflake's edge is model neutrality, running customer workloads across OpenAI, Anthropic, and open-source models to cut cost and lock-in. He flagged enterprise "trepidation about being too reliant on a single model source," and argued "the harness is the new browser," which is why Snowflake is building user-facing products (its Coco agent and "Cowork" front end) rather than sitting "buried under someone else's harness."

Cloudflare (NET). Bear-lean on Chip Stock Investor (Aug 20). Nicholas Rossolillo argued hyperscalers building their own CDNs (content delivery networks) commoditizes standalone providers: despite 33%+ revenue growth, Cloudflare's GAAP operating margins stay low and growth "is only sustainable if it continues spending a lot of money" on stock-based sales comp, "not this critical choke point… a flag for Cloudflare."

Palantir (PLTR). Bull. On The Rundown (Aug 16) it was called "the best story in all of software," with ~93% revenue growth acceleration and an orchestration-layer role that insulates it from frontier-lab competition, the caveat being a ~90x forward-earnings multiple. On Stock Market Today with IBD (Aug 20), a holder kept it because "it just hasn't done anything wrong yet."

CrowdStrike (CRWD) & Datadog (DDOG). Both bullish on Investing Unscripted (Aug 19): CrowdStrike as a durable performer with AI-driven addressable-market expansion; Datadog "a fundamentally sound business" capable of market-beating returns over 5–10 years even if it looks ~60% overvalued near-term. (Separately, Brew Markets (Aug 20) flagged CrowdStrike's CTO departure.)

Palo Alto Networks (PANW). Bull mention on Network Break (Aug 17) for its new "Frontier AI Defense" service that finds exposures beyond traditional scans, "over a third of discoveries aren't tied to CVEs."

GitLab (GTLB). Neutral-constructive on Tech Disruptors: the hybrid seat-plus-consumption model (above), model-neutral positioning, and a much larger revenue opportunity now that it can "monetize based on the outcomes of that work" rather than just headcount.

Workday (WDAY). In the news for buyout rumors, The Compound and Friends (Aug 18) and The Chad & Cheese discussed a possible Silver Lake take-private, using it as evidence that depressed SaaS multiples make such deals possible.

Shopify (SHOP), HubSpot (HUBS). Mostly referenced as platforms rather than analyzed as stocks this week; the Everyday AI (Aug 17) show cited a Gartner prediction that 40% of enterprise apps will include AI agents by year-end, "agentifying" tasks in tools like Salesforce, ClickUp, and HubSpot.

Private companies

Anthropic. The week's headline name. ARK (FYI) called it the fastest-growing software company ever ($9B to ~$70–80B run rate in under a year); Bloomberg Intelligence (Aug 21) said it's grown "from $10 billion to $65 billion in ARR over eight months" and uses special-purpose vehicles to lease Broadcom chips. On valuation, ARK noted a rumored ~$2 trillion IPO would be "20 times run-rate revenue… lower than premium public software companies like Cloudflare and Palantir trading at 30x." The forward-revenue debate: Gavin Baker floated $400–500B for 2027 while Brad Gerstner countered with $250B.

OpenAI. Framed as the challenger to Anthropic's momentum: ~$40B ARR, wider losses, cutting API prices to defend share, and (per CNBC) planning to IPO in 2027 or sooner. Also drew attention for pausing frontier training over safety incidents.

Cursor / Anysphere. The week's most-repeated deal: acquired by SpaceX for $60B (all-stock), per numerous shows including the Elon Musk Podcast (Aug 17), which cited 64% of the Fortune 500 as users and ARR "scaling to $8 billion by end of year." Cursor also launched "Origin," a GitHub competitor, during a GitHub outage, widely read as smart competitive timing.

Stripe. Very active: its ~$7B–$7.5B acquisition of OpenRouter was framed as owning the "tollbooth" for AI token routing (Tech Brew Ride Home, Aug 20, Patrick Collison: tokens are "the central currency for AI"). On The a16z Show (Aug 17), Will Gaybrick said Stripe uses AI to "build more, not less," with its internal agents generating 7,000 pull requests weekly (30% of all PRs).

Harvey. The open-source flashpoint, released its first in-house model (post-trained on the open-weight Kimi K3) so clients' proprietary legal data stays out of frontier labs' hands.

Databricks. Mentioned mostly in passing (e.g., a DOJ antitrust probe into shared board seats), not deeply analyzed this week.

Owner (private vertical SaaS). A useful operator case study on SaaStr (Aug 19): CEO Adam Guild described rebuilding a ~$100M restaurant-software business around "extremely opinionated" agentic products, 83% of new customers now start with AI products, up from 0% two years ago.