Newsletter · · Ashutosh Agarwal

Software Earnings Bury the SaaSpocalypse as the AI Pricing Scramble Begins - Weekly SaaS / Software Podcast Recap - Week of September 6, 2026

Weekly SaaS / Software Podcast Recap for the week of August 30 to September 6, 2026. Podcast synthesis on the earnings week that declared the SaaSpocalypse over, Snowflake's blowout comeback, the scramble over how to price AI agents, the Claudeforce partner-or-predator fight, and the concentration risk sitting under the whole AI boom.

Weekly SaaS / Software Podcast Recap

Week of September 6, 2026: Software Earnings Bury the SaaSpocalypse as the AI Pricing Scramble Begins

The one-line version: After six months of doom, the podcasts declared the "SaaSpocalypse" officially over this week. Software stocks that were left for dead in February came roaring back on earnings, Snowflake became the poster child of the comeback, and the whole conversation shifted from "will AI kill software?" to "how do software companies charge for AI, and can they do it before the AI labs eat their lunch?" Underneath the victory lap, a loud minority kept pointing at a scarier question: what happens to everyone if OpenAI stumbles?

A quick glossary, since a few terms come up constantly below:

  • SaaS ("software as a service"), software you rent by subscription instead of buying, like Salesforce or Adobe.
  • ARR, annual recurring revenue, i.e. subscription sales annualized. NRR, net revenue retention, how much more (or less) existing customers spend a year later; above 100% means they are spending more.
  • Seat-based pricing, you pay per user ("seat"). Consumption/usage-based, you pay for how much you use. Outcome-based, you pay only when the software actually delivers a result.
  • System of record, the trusted database a company runs its business on (your CRM, your payroll system).
  • Hyperscaler, the giant cloud providers: Amazon AWS, Microsoft Azure, Google Cloud (and increasingly Oracle).

1. Dominant themes

Theme 1: the earnings week that buried the SaaSpocalypse

This was the week's biggest story, and podcasters said it plainly. Back in February, when AI tools were launching weekly and everyone decided traditional software was toast, the US software index (the IGV) fell more than 30%. This week, the survivors took a victory lap.

On Prof G Markets, Ed Elson ran a full "we were right" segment, having publicly bought the four most beaten-down names in February:

"Adobe has risen 6%. Salesforce has risen 34%. Microsoft has risen 34%. And ServiceNow has risen 37%. Meanwhile, if you had purchased the software index as I had suggested, you would now be up more than 30%... SaaS's death was greatly exaggerated and software will continue to live on."

His broader point: markets are usually good at pricing, but "there are moments where the market does lose its mind," and the SaaSpocalypse was "a perfect case study in herd mentality and groupthink."

On The Six Five, Daniel Newman and Patrick Moorhead framed the entire episode around it, the title literally called it "the earnings week that buried the SaaSpocalypse." Newman's verdict on the whole scare:

"Nobody ever left their enterprise software. It was a dream. It was a great ploy. It was a tabloid headline that created massive and unfortunate selling that hurt a lot of companies... SaaS isn't dead, it never was."

Why it matters: the rotation was real money moving. On The KE Report, Joel Elconin described money rotating "out of overvalued semiconductor stocks into beaten-down software names," with Salesforce and ServiceNow moving higher on Microsoft's earnings catalyst.

Theme 2: AI agents and the great pricing scramble

The deepest, most repeated theme all week was pricing: as software companies bolt AI "agents" (software that does tasks on its own) onto their products, nobody agrees on how to charge for them. The industry is drifting away from paying per user toward paying for usage, or paying only when the AI actually works.

Salesforce sat at the center of it. On Tech Brew Ride Home, CEO Mark Benioff was quoted directly from the earnings call:

"Customers want to buy and want to price in different ways. This is something I've learned really aggressively recently."

Salesforce is now letting customers negotiate custom Agentforce contracts that charge based on how much revenue the AI grows or how much cost it cuts. The same podcast laid out how widespread the shift has become:

  • OpenAI has started letting some big customers pay only when its AI completes a task.
  • Sierra and Finn (which Salesforce is acquiring for $3.6 billion) charge only when the AI resolves an issue without a human.
  • Cognition (the coding startup) is promising enterprise customers up to $10 million in credits if its work fails to deliver value worth what they paid.
  • Adobe, HubSpot, and Zendesk have already moved toward charging for AI "only when it works."

On 20VC, Jason Calacanis called Salesforce's move the real headline, not the marketing:

"They said... we're going to do more outcome-based deals, which is a BFD. So I think the marketing was great... but the real things, the commitment to multi-service and the beginning commitment to outcome-based pricing, are huge changes for a $45 billion run rate company."

He tied it to Salesforce's just-announced purchase of Intercom (whose support product charges per resolution) as proof Benioff wants to "play the outcome game" and not just be a back-end database.

How big is the agent business already? The numbers cited varied by show but all pointed up and to the right:

  • The Canadian Investor: Agentforce at $1.5B ARR, up 205% year-over-year, with agent customers running 2 to 4x higher contract values, and pricing rising, not falling.
  • Telltales: Agentforce $1.5B ARR up more than 240%, and Agentforce plus Data360 combined at nearly $3.9B ARR, up 210%.

Why it matters: many software companies launched AI tools that haven't yet lifted overall revenue growth, so how they price agents is now the whole ballgame for whether AI is a tailwind or a margin trap.

Theme 3: "Claudeforce," is the AI lab a partner or a predator?

The week's marquee product event was Claudeforce, Anthropic's Claude wired directly into Salesforce, with Salesforce data, workflows, and governance living inside Claude's interface. Dario Amodei (Anthropic's CEO, the loudest "AI will disrupt everything" voice) sat next to Benioff on CNBC to bless it. Salesforce stock popped ~20%.

Podcasters read it as the physical embodiment of the "AI is the UI" idea: increasingly, workers won't open Salesforce at all, they will ask Claude, and Claude reaches into Salesforce behind the scenes ("headless" software). On The Six Five, Newman explained the logic: enterprise software needs to be deterministic (exact, auditable), while AI models are probabilistic (same question, slightly different answers), so pairing Claude's reasoning with Salesforce's governed data "can be really powerful." But he also flagged the catch: "once the UI moves, the seat pricing moves." (More on the partner-vs-predator fight in the Debates section, it was the sharpest disagreement of the week.)

Theme 4: hyperscaler capex and "concentration risk"

The bull case for software met a persistent bear drumbeat about how the AI boom is financed. The recurring worry: hyperscalers are spending ~$700 billion on capex this year (about half of 2026's projected 2% US GDP growth, per The Real Eisman Playbook), and that spend increasingly leans on debt.

  • Research @ Citi: over $200 billion of hyperscaler debt issued year-to-date at twice last year's pace, plus ~$100B in data-center project financing, in a period Citi's credit strategist called one of "greatest opacity."
  • Monetary Matters: for the first time in the speaker's lifetime, tech is being financed with debt rather than equity, Google now paying ~6% coupons versus 2 to 2.5% in 2021.

Why it matters for software: if the AI ROI disappoints and that capex slows, the money currently flooding into infrastructure and data-platform software slows with it.

Theme 5: the frontier-model arms race (and the Anthropic IPO)

The AI-lab layer had a blockbuster week that spilled straight into software sentiment:

  • OpenAI launched GPT-6 "Astra," with president Greg Brockman claiming it qualifies as AGI (Daily Tech News Show), pricing floated around $10 per million input tokens / $50 per million output tokens, matching Anthropic.
  • Anthropic is building a war chest ahead of an IPO: Bloomberg Tech said it crossed $100B ARR and 10x'd in a year, while Bloomberg Intelligence reported a $15 billion pre-IPO credit facility being finalized. On All-In, the hosts described OpenAI and Anthropic as a "frontier intelligence duopoly" and called Anthropic's coming IPO "a massive wealth generation event."

Theme 6: vibe coding, and OpenAI cutting off Cursor

"Vibe coding" (building software by describing what you want to an AI) stayed red hot, but the week's story was a power play: after SpaceX acquired Cursor (the AI coding tool from Anysphere), OpenAI cut off Cursor's access to its models (AI Daily Brief, AI Chat). It ended what one show called a "$1 billion annual partnership," though OpenAI models were only ~5% of Cursor's traffic. The broader takeaway on 20VC: enterprises are getting nervous about vendors that can cut them off "just because you don't like that other guy," and will start demanding openness.

2. Key debates

Debate 1: Claudeforce, genuine partnership vs. Trojan horse

This was the sharpest exchange of the week, staged explicitly as a debate on The Six Five.

Side A, it's a partnership; SaaS wins (Daniel Newman): Anthropic needs the software incumbents as its distribution channel and doesn't want the data, permissions, or compliance liability.

"If you're plotting to kill the system of record... I don't think you'd marry it on national television. You marry it because you need to... Frontier Labs have models. SaaS has 150,000 enterprise logos... Killing SaaS is basically killing their own channel."

Side B, it's a Trojan horse; incumbents are being hollowed out (Patrick Moorhead):

"Dario's public reassurances is exactly what a smart displacement play looks like... Salesforce and Claude means sellers never open Salesforce... once the UI moves, the seat pricing moves."

The wedge: whether becoming the invisible back-end behind Claude is a durable, high-value role, or the slow removal of the customer relationship (and pricing power) that made SaaS valuable in the first place. Even the bull, Newman, conceded it is "a re-rating," the software UI "probably isn't going to be the destination it once was."

Debate 2: seat-based pricing vs. usage/outcome-based pricing

Side A, the subscription model is superior and here to stay. On The Information's TITV, a VC argued SaaS's predictability, near-zero marginal cost, especially as smaller models run locally, makes subscriptions economically better than usage or outcome pricing, and predicted founders migrating back to it.

Side B, customers now want to pay for outcomes, and incumbents must follow or shrink. On 20VC, Calacanis put it bluntly:

"Customers want outcomes now. That is why Palantir is growing 90-something percent... can you deliver outcome from a system of record, or will agents or other systems deliver outcomes? If they deliver the outcomes, you will shrink over time."

The wedge: whether AI's unpredictable cost-to-serve breaks the tidy per-seat model, and whether incumbents can charge for results without cannibalizing their own recurring revenue.

Debate 3: "Is SaaS dead?", commoditized vs. moated

Even with the stocks recovering, the underlying disruption question stayed live, and the answer landed as "it depends what you own."

  • Vulnerable: On The Intelligence Age, the host argued bloated, feature-stuffed SaaS is genuinely exposed, citing a company that replaced a $600K/year Salesforce implementation with an agent-built custom tool. TECHtonic's Michael Speranza argued the moat is no longer code (AI commoditized it) but deep industry expertise and proprietary data, so horizontal SaaS is pressured while vertical specialists prevail.
  • Defensible: On The Josh Bersin Company, Bersin called the narrative flat wrong, pointing to Workday generating $600M in new AI-related ARR in Q2 with 200% quarter-over-quarter growth. On RiskReversal, Paul Kedrosky dismissed the "SaaSpocalypse" as naive extrapolation, enterprises simply don't want to build and maintain their own mission-critical CRM. And Engineering Culture by InfoQ argued the "build it yourself with AI" dream collides with reality: token bills that climb "from $20/month to $40 to 70k annually," plus maintenance and security burdens that non-software companies can't sustain.
  • The synthesis, from The Learning Corner: software has already recovered 60%+ from the lows, but real disruption is happening, legacy UI, databases, and systems of record have lost value, and interoperability (MCP servers, REST APIs) is now table stakes.

Debate 4: the AI boom's foundation, solid, or one stumble from a crisis?

Side A, concentration risk is catastrophic. On Better Offline, Ed Zitron laid out data from Ramp:

"80% of OpenAI's enterprise revenues come from 1% of their customers, a number that hasn't improved over the last 3 years."

His punchline: many of those top customers are themselves VC-funded AI startups burning subsidized tokens, so "the only people the tech industry can get to spend money on AI is the tech industry." The Real Eisman Playbook located the "Achilles heel" specifically at OpenAI: June-quarter revenue of $6.7B, up only 18% versus March, while costs surged to $12.3B (up $3B in one quarter), versus Anthropic's $11B+, up over 100%. With ~70% of hyperscaler AI revenue tied to those two labs, Eisman argued the whole US economy now leans on them.

Side B, don't call the top yet. Notably, Eisman himself refused to predict doom: "the data does not support the end of the AI story, at least not yet... if AI generates new businesses and large returns on this CapEx investment, then the hyperscalers will be okay." And the software bulls (Prof G, Six Five, The Compound) argued this week's earnings proved demand is real and durable.

The wedge: is the AI economy a resilient flywheel, or a tower balanced on two private companies (one of them wobbling)?

3. Specific names

Public companies

Snowflake (SNOW), the week's biggest bull story. The standout comeback name after a blowout quarter.

  • Halftime Report: Josh Brown bullish, citing 126% net revenue retention and product revenue growth accelerating to 37%, justifying the move from $205 to $371 in one quarter; Jason Snipe bullish on the 37% growth and expanding margins.
  • The Compound and Friends: the host is long (with a trailing stop), calling Snowflake "one of the most important layers of the whole AI layer cake" with 9,000+ corporate customers running agentic workflows inside the data warehouse; noted the stock is up 164% from the April low. The bear counter he raised himself: rival Databricks is growing ~65%, roughly twice Snowflake's rate, and recently passed it on product revenue.
  • Squawk on the Street / Bloomberg Tech: CEO Sridhar Ramaswamy pointed to $1.49B product revenue (+37%), ~400bps operating-margin expansion to 15%, 2,000 new accounts, and full-year guidance raised to 36%.
  • Motley Fool Hidden Gems: bullish but flagged valuation, ~20x forward sales, ~80x free cash flow.

Salesforce (CRM), bull, with the pricing/headless overhang. Central to Claudeforce and outcome-pricing (above). The Morning Filter put a $280 fair value on it (bull, ~7% discount) on AI-driven acceleration and the Anthropic partnership, while noting valuation is "becoming full" at 18.5x 2026 earnings. On 20VC, Calacanis is holding a "pretty large slug," saying "they've weathered the apocalypse," but cautioned growth is now only ~11 to 12%.

Microsoft (MSFT), bull. On Halftime Report, the committee liked Azure's 40%+, supply-constrained, high-margin growth at "low 20s" multiples. The Rundown more neutral, noting Copilot reached 30M paid subscribers but the stock is only up ~4% YoY.

Oracle (ORCL), bull into earnings. On CNBC Fast Money, DA Davidson's Gil Lauria called Oracle undervalued on its ability to convert AI capex into revenue, citing 34% revenue guidance and a large backlog getting less credit than peers, with a $20+ FY2030 EPS target, while flagging Cerner-debt constraints and a New Mexico facility's gas-access issue.

Adobe (ADBE), bear/skeptical, on the CEO change. Bloomberg Intelligence's Anurag Rana is bearish: stock down ~7% on leadership-change concerns, viewed as vulnerable to free AI tools (though disruption "hasn't materialized yet"); investor disappointment that the new CEO ran the smaller marketing-analytics unit, not Creative Cloud. On Chit Chat Stocks, a shareholder was neutral-to-negative, wanting "a rational capital allocator like Tim Cook," not a reinvention.

CrowdStrike (CRWD), mostly bull, one valuation bear. The Compound host is long, calling it furthest ahead in AI-era cybersecurity with "mandatory" spending. Business of Tech noted Falcon platform usage doubled year-over-year. The bear: The Morning Filter sees it overextended (fair value $152, two-star).

Palo Alto Networks (PANW), split. Closing Bell: Saket Kalia bullish, the AI-security narrative "can still work despite high multiples of 45 to 50x free cash flow." Bear: Chip Stock Investor's Nicholas Rossolillo is skeptical of value creation given the acquisition binge, the $25B CyberArk deal (Feb 2026), Chronosphere, and Console, despite 31 to 34% post-CyberArk growth, worried about dilution.

Zscaler (ZS), bull. On Squawk on the Street, CEO Jay Chaudhry said the company is "built for this moment," citing record Q4 results, 25% ARR growth to $3.8B, 24% operating margin, and positioning zero-trust as the answer to agent-based threats.

Okta (OKTA), bull. Business of Tech: strong beat, with new AI-related products now nearly a third of bookings.

Palantir (PLTR), bull, on rich valuation. The Intrinsic Value Podcast: trades at ~100x trailing sales, but the analyst argues it is justified by ~90% growth and ~65% operating margins (a "rule of 40" score of ~155% vs. a healthy 40); on 2027 assumptions the P/S compresses toward 20x.

Intuit (INTU), relative loser. The Canadian Investor framed AI as a tailwind for agent-integrators like Salesforce but a headwind for Intuit.

Private companies

OpenAI, the week's biggest question mark. Launched GPT-6 "Astra" (claimed AGI), but the financial picture drew scrutiny: $6.7B June-quarter revenue, +18% QoQ, costs at $12.3B (Eisman); 80% of enterprise revenue from 1% of customers (Zitron); ~900M weekly active users and ~$25B annualized revenue (Everyday AI); ads business hit a $1B run rate. Also cut off Cursor after the SpaceX deal.

Anthropic, the strong one. ~$100B ARR, 10x'd in a year, finalizing a $15B pre-IPO credit facility, with the highest revenue-per-token and enterprise traction (Claude Code cited at ~$2.5B revenue). Shipped Fable 5.1 (cheaper tokens) and restricted the more capable Mythos 5.1 to cybersecurity/life-sciences partners. Anchored Claudeforce.

Cognition, outcome-pricing pioneer, huge round. 20VC noted a raise at $46 billion, and its guarantee of up to $10M in credits if it under-delivers.

Replit, coding-tool bull. The Smashi Business Show: valued at $9B (up from $3B six months earlier), second only to Lovable at $13.3B among AI coders; praise for founder Amjad Masad's international expansion.

Cursor (Anysphere), the disruption target. Acquired by SpaceX; OpenAI pulled model access. Framed mostly through the OpenAI to Elon rivalry; OpenAI models were only ~5% of Cursor traffic.

Glean, enterprise-search bull. On Revenue Builders, Daniel Simon highlighted a deal quantified at 50 to 100x ROI and Glean Protect's security edge over ChatGPT and Copilot.

Sierra & Finn, being bought by Salesforce for $3.6B; pure outcome-based pricing (pay per resolution). Intercom, also acquired by Salesforce, resolution-based pricing.

Stripe + OpenRouter, the week's big fintech/AI-infra deal. Market Maker and AI to ROI: Stripe acquired OpenRouter for $7.5 billion, a bet on the emerging "AI gateway" market (OpenRouter processes ~55 trillion tokens weekly across 400+ models); the startup had raised at $1.3B post-money just months earlier (May 2026).

Databricks, Snowflake's fast-growing rival, ~65% growth, AI-native architecture, expected to go public later this year (via The Compound).

Also raising: Clay at $7B, Linear at $2.5B (20VC).

Xero (public, defensive-moat view). On TITV, the CEO addressed a 47% one-year stock decline on SaaSpocalypse fears, arguing its regulated infrastructure, compliance rails, and bank/tax-authority integrations can't be replicated by AI alone.