Newsletter · 2026-09-13 · Ashutosh Agarwal

Oracle Soars on Backlog, Adobe Falls Despite a Beat, and Seat Pricing Cracks - Weekly SaaS / Software Podcast Recap - Week of September 13, 2026

Weekly SaaS / Software Podcast Recap for the week of September 6 to September 13, 2026. Podcast synthesis on Oracle's giant AI backlog and cash burn, Adobe falling despite a beat, the shift from per-seat to usage and outcome pricing, vibe coding's security bill, and the is-SaaS-dead debate as venture investors pull back from classic software.

Weekly SaaS / Software Podcast Recap

Week of September 13, 2026: Oracle Soars on Backlog, Adobe Falls Despite a Beat, and Seat Pricing Cracks

This week the podcasts stopped arguing about whether AI would change software and started arguing about who gets paid when it does. Two big earnings reports, Oracle and Adobe, split the room. A pricing revolution that has been building for years finally got a name from nearly every guest: the shift away from charging "per seat" toward charging for what customers actually use, or for the results they get. And the oldest question in the sector, "is SaaS dead?", reached a genuine fever pitch, with venture capitalists openly admitting they have stopped writing checks to ordinary software startups.

Here is what people were actually saying, in their own words.

Top of Mind This Week

• Oracle's earnings were the single biggest story. Its order backlog exploded to $664 billion, but the podcasts spent as much time on the fine print (half of that backlog is one customer, OpenAI, and the company is burning cash) as on the headline number.
• Adobe beat expectations and the stock fell anyway. It became the poster child for the week's central fear: that even a great, profitable software company can be punished if investors think AI is coming for its customers.
• "Per-seat" pricing is dying, and everyone said so. From a 20-year software analyst to the sales chief of Databricks, guest after guest argued that charging by the number of user logins no longer makes sense when the "worker" is now an AI agent.
• Venture investors are quietly walking away from classic software. One VC said his firm's software deals have "fallen off a cliff." Another said the business of writing custom code for money "will not exist two years from now."
• The private-market fireworks were extraordinary. Reports of Anthropic aiming for a $2 trillion IPO, a coding startup (Cognition) raising money at a $48 billion valuation, and Nvidia buying Hugging Face for $12.93 billion all landed in the same seven days.

1. Dominant Themes

Theme 1: Earnings Week, Oracle Soared on Backlog, Adobe Was Punished for a Beat

The week's calendar was dominated by two software heavyweights reporting results, and the reaction to each tells you exactly what the market is rewarding and fearing right now.

Oracle (ORCL) delivered what Zaid Admani of Public.com, on The Rundown, called a "pretty strong quarter." The numbers he laid out: revenue up 30% from a year ago to $19.3 billion, adjusted earnings of $1.92 a share versus the $1.74 expected, and, the eye-catcher, its cloud infrastructure business (the data centers Oracle rents out for AI computing) up 121% to $7.4 billion. Strip that division out, Admani noted, and "the rest of Oracle grew just 3%. So almost all of Oracle's growth is coming from AI." Oracle signed more than $30 billion in new AI cloud contracts in the quarter, pushing its total backlog (the value of work it has been hired to do but hasn't delivered yet) to $664 billion.

But Admani was careful to walk through the worrying details, and this is where the podcast added value over the headline. "Almost half of Oracle's backlog is from a single customer, which is OpenAI," he said, a big concentration risk if OpenAI ever stumbles. Oracle's spending on building data centers ("capex") exploded to $28.5 billion in the quarter, up from just $8.5 billion a year earlier, dragging free cash flow (the cash left after running and investing in the business) to negative $5.4 billion. The company now carries $125 billion in debt, and analysts don't expect it to generate positive free cash flow again until 2030.

The clever wrinkle Admani flagged: Oracle is now asking customers to prepay for computing capacity, or even to buy the Nvidia chips themselves and hand them over, what he called a "bring your own chips" model. Oracle collected $14.4 billion in customer prepayments in the quarter alone. "Instead of them going out and borrowing billions of dollars and buying a ton of Nvidia chips and then hoping someone rents them, the customer just signs the contract first and helps pay for the hardware up front."

Not everyone was convinced Oracle is out of the woods. On The Information's TITV, RBC Capital Markets analyst Rishi Jaluria said Oracle has "the most aggressive roadmap" of any cloud provider, with spending peaking in 2027–2028, but warned the company is burning cash and may not turn cash-flow-positive soon, so he would "rather own Microsoft at these levels over Oracle," since Microsoft is already generating cash. On CNBC's Fast Money, Dan Ives called Oracle a "penalty box" stock that needs several more quarters to prove it can actually deliver on that $664 billion backlog. More bullish voices included Jim Cramer on Squawk on the Street, who praised management's discipline and noted AI agents in production doubled quarter-over-quarter to 3.5 million, and Citi's Tyler Radke on the same show, who kept a buy rating, pointing to a $20 billion equity raise that removed a cloud hanging over the stock.

Adobe (ADBE) was the mirror image: good results, bad reaction. On The Rundown, Admani reported revenue up 13% to $6.76 billion, an earnings beat, annual recurring revenue from its newer AI products up more than 150% from a year ago, and over 1 billion monthly active users. And yet the stock fell. "Wall Street still wasn't convinced that Adobe will survive the AI era," he said, citing pressure from newer design tools like Figma and Canva. On Schwab Network's earnings panel, analysts pinned the sell-off on a miss in Adobe's "remaining performance obligations" (a measure of contracted future revenue): investors wanted proof that Adobe can hold onto customers and its pricing power, not just show flashy AI adoption numbers.

The other name that kept coming up as the counter-example was Snowflake (SNOW), the data-warehousing company. On The Six Five, the hosts noted Snowflake was up 111% over six months after beating on revenue and raising its guidance, with roughly half the beat driven by AI products, prompting a wave of analysts to lift their price targets. Their framing was pointed: Snowflake "successfully executed on the narrative that software and data workflows would not be eradicated by models." On Telltales, the hosts contrasted Snowflake (which raised full-year guidance to about $6.07 billion, with 60% of accounts adopting AI coding tools, and jumped) with Broadcom, which beat but guided next-quarter revenue slightly below hopes (~$34.8 billion) and got sold off: evidence, they argued, that the market is now trading on guidance revisions, not on AI demand itself.

Theme 2: The Pricing Revolution, From "Per Seat" to "Per Use" to "Per Outcome"

If one idea united the operator and investor podcasts this week, it was that the classic software pricing model (charge a fixed fee per user, per month) is breaking down. The logic is simple: if an AI agent, not a human, is doing the work, then "number of human logins" is a strange thing to bill for.

The clearest overview came from software analyst Matt Hedberg on Run the Numbers, who has covered the sector for 20 years. He framed it as the third great shift: "Cloud really changed where software runs, where AI is really kind of changing what software does." Just as cloud computing moved the industry from per-seat licenses toward consumption-based pricing (he cited Snowflake, which charges for how much you compute), AI is pushing pricing toward "per token, per query, or per outcome rather than by seat," a fundamental change in how software makes money and where the durable competitive advantages ("moats") will sit.

The most emphatic version came from Ali Ghodsi, co-founder of Databricks, and its sales chief on The Peel with Turner Novak. The company recounted how, in the early days, it tried adding per-user fees on top of usage-based pricing, and found that charging per user actually restricted adoption and dragged down usage. So it scrapped user fees entirely and tied all pricing to consumption. Ghodsi's prediction: this model is superior to seat-based pricing and will become universal across software as user-based models prove insufficient. (For scale on the stakes: the episode was framed around Databricks going from $1 million to more than $7 billion in annual recurring revenue in ten years.)

The same message came from a startup at the other end of the size spectrum. On The Official SaaStr Podcast, the ElevenLabs team described moving from fixed annual platform fees to "pay-as-you-go" usage pricing, which reduced friction and drove more demand, and endorsed experimenting with outcome-based pricing to lower the barrier to getting started. And on Tech Disruptors, a speaker from Amazon flagged the flip side of consumption pricing: it makes bills unpredictable. He said 69% of companies exceeded their AI budgets, and urged businesses to forecast costs by component (users, workflows, token costs, request volume) rather than assume a fixed monthly number.

Theme 3: Hyperscaler Spending Is Now Colliding With the Bond Market

A recurring macro thread: the enormous sums the big cloud companies ("hyperscalers") are spending on AI data centers are now large enough to move interest rates, which in turn affects every software valuation.

On The Compound and Friends, Erik Norland noted that hyperscalers are now issuing bonds in amounts comparable to US Treasury issuance at the long end of the market, as their spending outstrips the cash they generate, helping push the 10-year yield toward 4.9% and the 30-year to multi-decade highs. Two bond managers on Excess Returns put a number on it: hyperscalers are issuing roughly $400 billion of the ~$2 trillion in investment-grade corporate debt this year to finance data centers, "a phenomenon that barely existed three to four years ago." On On The Tape, Cameron Dawson observed that after everyone expected hyperscaler capex to be flat, it is now projected to grow 90–100% in 2026, with the companies increasingly leaning on outside capital rather than their own cash flow, the definition of a "crowding out" risk for other borrowers.

There was also a healthy dose of skepticism about how the spending is being reported. On both Excess Returns and Two Quants and a Financial Planner, speakers described "accounting games": Nvidia extending payment terms to help customers flatter their cash flow, Microsoft reclassifying leases to lower its reported capex, and additional spending parked in off-balance-sheet vehicles.

Theme 4: "Vibe Coding" Goes Mainstream, and the Security Bill Comes Due

"Vibe coding" (using AI to write software from plain-English descriptions, often by people who can't code) was one of the most-discussed topics of the week, and the tone had shifted from wonder to caution.

On Everyday AI, the host walked through the boom: Cursor (the coding tool from Anysphere) at $2 billion in annual recurring revenue with over 1 million paying customers, alongside GitHub Copilot, Anthropic's Claude Code, OpenAI's Codex, Replit, and browser-based builders like Lovable and Bolt. The striking statistic: 63% of "vibe coders" have never written code, and companies are starting to replace bought software entirely; the host cited a $40,000 Salesforce setup replaced by a three-hour Lovable build. But he was cautious: developer trust in AI-written code remains shaky.

That caution was the whole point on the AWS for Software Companies podcast, where David Hsu, CEO of Retool, warned that AI tools now let non-engineers ship applications, producing (per GitHub) 10x more code but at lower quality. Retool found 7–8 customers had independently published proprietary company data openly on the internet, and its audits reveal "90–95% of companies have major security problems." A study cited on The Pacesetter Pod put it starkly: 45% of AI-generated code introduces known security vulnerabilities (per Veracode). ServiceNow's CEO gave this its best label of the week: "vibe slop" (more below).

Theme 5: The "SaaSocalypse" M&A Wave and the Agent Land-Grab

Even as pundits debated whether software is doomed, the deal-making told a more nuanced story. On Next in Tech, Bren Daly argued the "SaaSocalypse" fears have moderated, pointing to continued big software M&A (Salesforce–Contentful, ServiceNow–Autodesk) as evidence the incumbents are still buying, not dying, though he warned hyperscalers' massive upfront AI spending is exhausting the cash that might otherwise fund acquisitions. In infrastructure software specifically, Software Engineering Daily flagged two notable moves: Temporal raising $500 million at a $12 billion valuation (more than double its valuation six months earlier), riding the wave of AI agents needing reliable long-running infrastructure, and Dynatrace acquiring Arize for $915 million, a sign that "AI observability" (tools for monitoring how AI systems behave) is being folded into the bigger monitoring vendors.

2. Key Debates

Debate 1: Is SaaS Dead, or Just Changing Shape?

This was the argument of the week, and the two sides could not have been further apart.

The bear case: classic software is being commoditized. On Venture Declassified, an investor named Mike (DeveloperTown) put it bluntly. He said his firm views the future cost of producing software as effectively zero, so the first question he now asks of any startup is: "How hard would it be for Bank of America to vibe code this? … Why would they even consider using your crappy startup software versus just building their own crappy startup software?" His verdict on the classic model: "Why would any company on the planet pay millions of dollars in licensing fees to Salesforce when they could just build a custom CRM for themselves in probably a week?" And on the money: "This is why we have not invested in very many B2B SaaS companies in the last two years… we have fallen off a cliff with SaaS investments. Almost everything we're doing now is going to med device, real estate, other stuff." He added that the business model of "trading dollars writing code for money… will not exist two years from now." Rob Walling, on Startups For the Rest of Us, sharpened the threat: the real danger isn't AI replacing a feature, it's competitors using AI to rapidly build competing modules into existing "system-of-record" software, commoditizing specialized niche tools.

The bull case: the data and the workflows are the moat, not the code. The counter-argument, made repeatedly, is that software's value was never really the code; it was trust, data, governance, and the messy business processes wrapped around it. On The So What from BCG, Brian Gouch argued that CRM (customer-relationship software) will change but not disappear: agents are replacing the interface and the workflows, not "the underlying data backbone and governance." Notably, even the bears half-conceded this point: in that same Venture Declassified exchange, when one host scoffed that a CRM is "just a database," another shot back that trust and integration are exactly why enterprises still pay. Matt Hedberg's Run the Numbers case was that AI is more likely to expand the total market for software (by automating workflows that were never automatable before) than to shrink it, the same way cloud did. And Snowflake's results were held up across multiple shows as living proof that a data-software company can grow because of AI, not despite it.

The wedge: whether the durable value of software sits in the code (which is being commoditized) or in the data, distribution, trust and governance around it (which is not).

Debate 2: Seat Pricing vs. Usage vs. Outcome Pricing

Everyone agreed the model is changing; they disagreed on where it lands. Usage-based is winning the argument: Databricks' Ali Ghodsi flatly predicted consumption pricing "will become universal" (The Peel), and ElevenLabs credited pay-as-you-go with driving demand (SaaStr). The skeptics' wedge is predictability: as the Amazon speaker on Tech Disruptors noted, usage pricing means customers can't forecast their bills, and 69% are already blowing past their AI budgets. The frontier of the debate is outcome-based pricing (charge only when the software delivers a result), which several guests called the logical endpoint but admitted is hard to measure fairly.

Debate 3: Oracle, Generational AI Winner or Over-Leveraged Penalty Box?

Oracle split the investors cleanly. Bulls (Jim Cramer on Squawk; Citi's Tyler Radke on Squawk; Anurag Rana on Bloomberg Intelligence) pointed to a $664 billion backlog, scarce-compute pricing power, an equity raise that de-risked the balance sheet, and improving operating margins. Bears/skeptics (Rishi Jaluria on TITV, who prefers Microsoft; Dan Ives on Fast Money, "penalty box") focused on negative free cash flow, $125 billion of debt, no positive cash generation expected until 2030, and the fact that ~half the backlog rides on a single, unprofitable customer (OpenAI). The wedge: do you trust Oracle to convert a giant backlog into cash before its debt and capex catch up with it?

Debate 4: Does Your Valuation Come From Growth, or From Your "AI Story"?

On Topline, investor Tomasz Tunguz of Theory Ventures argued the market has stopped rewarding growth alone and now prices companies on the credibility of their AI narrative, hence the episode's title, "Forget growth. Your valuation depends on your AI story." He described AI vendors deploying "forward-deployed engineers" (essentially embedded consultants) to help enterprises actually adopt the software, calling them a permanent feature "here to stay." The wedge with the traditional camp: is an "AI story" a real driver of durable value, or a temporary premium that reverses the moment growth slows, exactly what happened to Adobe this week.

Debate 5: The AI Labs' Concentration and Price War (OpenAI vs. Anthropic)

On Prof G Markets, RAMP economist Ara Karazian laid out a striking data point: OpenAI and Anthropic are now in a price war, both cutting prices on their top models while pushing cheaper "standard" and "light" tiers, and businesses are increasingly choosing the cheaper models because the return on investment is good enough. More important for the software world, he said these two labs drive "70 or 80% of the AI revenues" of the big cloud companies like Microsoft, Amazon and Google, describing "a level of concentration risk unseen in any other software category that we track." The debate: is this concentration a fatal fragility (if a few big customers pull back at once, the whole AI revenue stack wobbles) or a normal feature of a young technology? Karazian himself landed in the middle, bullish that OpenAI and Anthropic will keep dominating US enterprise AI, but expecting the growth to come from lower-margin cheaper models.

Debate 6: Is Anthropic's Rumored $2 Trillion IPO Sane?

On Rich Habits, the hosts dug into reports that Anthropic is targeting a mid-October IPO at a $2 trillion valuation, more than double its ~$965 billion valuation in May. The math they walked through: it would require revenue to grow from a $47 billion annual run-rate to $120 billion by end-2026 and $200 billion by 2028, implying roughly 16x and 10x forward revenue. The catch: Anthropic is reportedly burning $5.2 billion in cash on $9 billion of revenue with only ~30 million monthly users, versus OpenAI's larger user base. The debate is whether these valuations reflect a genuine platform shift or the top of a cycle, a question echoed on The Acquirers Podcast, where Zeke Baker compared today's AI valuations unfavorably to the 2000 dot-com peak, citing TAM (total addressable market) claims (SpaceX at $28 trillion, Anthropic at $30 trillion) against a global GDP of only ~$120–130 trillion.

3. Specific Names, the Bull and Bear Angles

Public Companies

Oracle (ORCL): Mixed to bullish. $664B backlog, cloud infrastructure +121% to $7.4B, revenue +30% to $19.3B, EPS $1.92 vs $1.74 expected; but negative $5.4B free cash flow, $125B debt, ~half the backlog is OpenAI, and a smart new "bring your own chips"/prepay model ($14.4B prepayments collected). Bulls: Cramer, Citi's Radke, Bloomberg's Rana. Bears: RBC's Jaluria ("rather own Microsoft"), Dan Ives ("penalty box"). (The Rundown, TITV, Fast Money)

Adobe (ADBE): Bearish reaction despite a beat. Revenue +13% to $6.76B, AI-first ARR +150%, 1B+ monthly users, but the stock fell on a contracted-revenue (RPO) miss and fears that Figma and Canva are eroding its core. The market wants proof of customer retention and pricing power, not adoption stats. (The Rundown, Schwab)

Snowflake (SNOW): Bullish, the week's "AI winner" in software. Up 111% over six months / ~41% on the week; FY guidance raised to ~$6.07B; ~60% of accounts adopting AI coding tools; analysts lifting targets (e.g., toward $437). Held up as proof that data software grows because of AI. Caution from IBD: wait for it to hold above ~$342 given broad software weakness. (The Six Five, Telltales)

Microsoft (MSFT): Relative-safety bull. Repeatedly cited as the safer AI infrastructure bet because it's already cash-flow positive (Jaluria prefers it over Oracle). Planning to roughly triple data-center capacity toward ~$200B annual spending. (TITV, Bloomberg Intelligence)

ServiceNow (NOW): Bullish, "the execution layer." CEO Bill McDermott argued on Squawk on the Street that "the next big AI investment opportunity is moving away from infrastructure spending" toward enterprise software that governs and secures AI. His pitch: "94% of the companies out there have zero to very little AI visibility"; ServiceNow provides a "control tower," a "kill switch" for rogue agents, and ROI tracking. He coined "vibe slop" for the technical debt created by AI-written code, claiming a 10x lower total cost of ownership versus building from scratch.

Salesforce (CRM): Debated. The archetypal "why pay millions when you can build it yourself?" target for the bears (Venture Declassified), but defended by BCG as owning the data backbone and governance agents can't replace (BCG); also cited (Salesforce–Contentful) as proof the SaaS M&A machine is alive (Next in Tech).

Palantir (PLTR): Held up as the exception. In the Venture Declassified bear case, Palantir's proprietary data "ontology" was cited as the kind of "magical" advantage that survives commoditization. Also named among software stocks Schwab clients took profits in after a 20–25% rally. (Venture Declassified, Schwab STAX)

Cloudflare (NET): Structurally bullish. CEO Matthew Prince said on Empire that 80% of major AI companies (including OpenAI and Anthropic) run on Cloudflare's infrastructure, and argued Google has an unfair data advantage, seeing "twice as much internet traffic as OpenAI, four times more than Anthropic."

MongoDB (MDB): Bullish product pitch. CPO Ben Safalo argued on Boardroom Club that MongoDB's JSON-native design plus integrated search/vector/embeddings makes it well-suited for AI agents, with higher retrieval accuracy that reduces token costs versus bolting on a separate vector database.

Datadog (DDOG) / Dynatrace / observability: Consolidation theme. Dynatrace's $915M acquisition of Arize framed as larger monitoring vendors absorbing AI-specific observability. (Software Engineering Daily)

Broadcom (AVGO): Beat but sold off on soft next-quarter guidance (~$34.8B), a "guidance revision" reaction. (Telltales)

Arm (ARM): Bullish. CEO said its new server-CPU venture saw demand jump from ~$1B in March to over $2B by May, with customers including Cloudflare, Oracle and Meta, on power-efficiency claims. (Big Boss Interview)

Pure Storage (PSTG, "EverPure"): Cautionary. Big hyperscaler deals (Meta and another top-five cloud) but free cash flow went negative (~-20% margins) due to $500M in prepaid NAND/storage components. (Chip Stock Investor)

AppLovin (APP): Value debate. A "30-bagger down more than half," examined on its business model and valuation assumptions. (The Intrinsic Value Podcast)

Private Companies

OpenAI: Central to the whole AI-software economy: ~half of Oracle's $664B backlog; ~70–80% (with Anthropic) of hyperscaler AI revenue; in a price war with Anthropic; launched a new image model aimed squarely at Canva. (The Rundown, Prof G)

Anthropic: Rumored $2T IPO targeting mid-October; ~$47B run-rate but burning ~$5.2B on ~$9B revenue; also the subject of a viral researcher-resignation over AI safety that dominated the general-tech podcasts and could show up in its IPO risk factors; walked away from a $6B acquisition of Israel's Descartes. (Rich Habits, Squawk)

Databricks: $1M → $7B+ ARR in ten years; the flag-bearer for consumption-based pricing becoming universal. (The Peel)

Cursor / Anysphere: $2B ARR, 1M+ paying customers; the leading "vibe coding" tool, though one report claims Elon Musk acquired it to train Grok, opening the door for rivals. (Everyday AI)

Cognition (Devin): Raised $2B at a $48B valuation; released a coding model (SWE-2) hitting ~50% on a hard benchmark at ~64% lower cost, read as VCs betting there's room for multiple winners in AI coding. (Equity)

Replit: Revenue reportedly leapt from $2.5M to $250M within a year after launching its AI coding agent, now targeting $1B ARR. (The Smashi Business Show)

Temporal: $500M raise at a $12B valuation, riding demand for reliable infrastructure to run long-running AI agents. (Software Engineering Daily)

ElevenLabs: Moved to pay-as-you-go pricing to reduce friction and drive demand; one 20VC guest flagged its ~$22B valuation as potentially rich if voice AI commoditizes. (SaaStr, 20VC)

Hugging Face: Nvidia's $12.93B acquisition (~86x revenue), framed as Nvidia moving to control the distribution layer for open-weight AI models. (Market Maker)

Stripe, Figma, Canva, Glean: Recurring references: Stripe as the archetype of a trusted brand-as-moat (Venture Declassified); Figma and Canva as the competitive threat weighing on Adobe (The Rundown); Glean discussed via its head of sales on go-to-market differentiation (Revenue Builders).