Newsletter · · Ashutosh Agarwal

Software Sorts Into Moats and Wrappers as the SaaSpocalypse Reprices Defensibility - Weekly SaaS & Software Podcast Recap - Week of July 5, 2026

SaaS and software newsletter for the week of July 5, 2026. With the SaaSpocalypse sell-off still fresh, operators and PMs spent the week sorting software into defensible systems of record and disposable wrappers, while a loud new backlash urged enterprises to stop feeding proprietary data to the frontier labs.

Weekly SaaS & Software Podcast Recap

Week of July 5, 2026: Software Sorts Into Moats and Wrappers as the SaaSpocalypse Reprices Defensibility

Window: June 28 – July 5, 2026 (trailing 7 days)

Top of mind

The dominant frame this week was defensibility over growth: with the "SaaSpocalypse" sell-off still fresh, operators, buyers and PMs spent the week sorting software into winners (systems of record, regulated verticals, mission-critical workflow) and losers (thin wrappers, horizontal point solutions). Two structural debates ran underneath everything, seat-based vs. consumption pricing as AI automates seats away, and who captures the value from AI agents, incumbents who own the system of record or the frontier labs. And a genuinely new fault line opened up: Alex Karp, Chamath and others made a loud, specific case that enterprises should stop feeding proprietary data to OpenAI/Anthropic and run open models behind their own control plane, framed around Anthropic's move to vertically integrate against its own customers (Figma, Cursor). On the tape, the software complex (IGV) has been the funding source for the memory/semis rally, and a handful of names, ServiceNow (bull), Adobe and Uber (bear), Palantir (bull), got worked over in detail.

1. Dominant Themes

Defensibility is the new valuation driver, "systems of record" vs. everything else. The clearest framework of the week came from Myles Lacey of Vista Point Advisors, who laid out a three-tier hierarchy buyers now use: at the top, "systems of record with proprietary data… ideally in highly regulated industries" where "you need a human in the loop"; a strong middle of "verticalized software that's more or less an ERP for a particular category"; and a bottom tier of "thin wrappers, BI dashboards, analytics, point solutions… a lot of those things are starting to be replaced by the frontier models themselves." He dated the shift precisely: "the SaaSpocalypse was precipitated really by the launch of cloud code agents" in late January/early February, sending investors to "a defensive posture… looking at software through the lens of a defensive moat as opposed to one of growth." The gut-check test he and the host use: "What would a customer lose if they turned the software off?" (The Path to Exit, Jun 30). Note the retention bar has ratcheted up, "we're even seeing a lot of pushback on gross retentions in the mid 80s."

Software has been the funding source for the memory/semis melt-up. On the tape, "software continues to be under pressure" while memory names "got stratospheric," a dynamic several hosts expect to normalize (Schwab Network, Jun 29). The IGV software ETF "is down 11%" while the SOXX is up "82% year to date"; ServiceNow specifically "trades now at 23 times forward estimates, and the 5-year average is 54 times" (Halftime Report, Jul 2).

Seat-based vs. consumption pricing, the model is genuinely in flux. The cleanest natural experiment: Help Scout's then-CEO tested a fully usage-based (per-contact) model for 12 months and reverted. "Even when this new usage-based business model would benefit a customer, even when they would pay less, they just didn't want to do it… there's a perception that people have more control over their costs when they pay per seat," despite per-contact pricing being "actually 30% less variable than per seat." They landed on "a hybrid, which is kind of where the whole market has now landed… seats plus you pay for AI resolutions" (Startups For the Rest of Us, Jun 30). ServiceNow's CTO described the emerging incumbent template, "We're a hybrid pricing model… When people are involved, I charge per seat. When AI is involved, it's consumption" (alphalist.CTO, Jul 2).

"Own your alpha", the anti-frontier-lab backlash went mainstream. The loudest theme of the week. Palantir's Alex Karp: enterprises are "livid… I am paying for tokens that create no value. These people are stealing the weights and alpha of my business," arguing critical infrastructure "does not run these models without an application layer" (Squawk Pod, Jul 1). All-In extended the argument to Anthropic's vertical-integration playbook (Claude Design/Science/Security/Legal/Financial/Code), calling it "very Microsoft-like… dominate the model layer… then use that position to capture the most lucrative verticals" (All-In, Jul 3).

"Vibe coding" and AI coding tools were the single largest cluster. Multiple angles: model-agnostic coding harnesses ("software factories"), the Cursor/Anysphere acquisition by SpaceX/xAI reportedly at ~$60B and its new iOS app (Everyday AI, Jul 2), and a security wrinkle, the "DuneSlide" prompt-injection vulnerabilities (CVE-2026-50548/50549, 9.8 CVSS) in Cursor, "used by over half of Fortune 500 companies" (The Cybersecurity Defenders, Jul 3).

Hyperscaler capex digestion and its read-through to software/infra. The narrative that hyperscaler free cash flow is being consumed by AI buildout, and that memory suppliers are capturing the pricing power, ran across several finance shows (see Debates and Names below).

2. Key Debates

Is traditional SaaS a melting ice cube, or a value opportunity?

  • Bear: Synthesia's CEO, "if you're not like an AI native company, the multiples have just gone down by an astronomical amount" (The Business, Jul 3). Aileen Lee: traditional SaaS is "trading at such crappy multiples, they haven't been very acquisitive," and pre-AI portfolio companies increasingly have to "burn the boats" and rebuild AI-native (This Week in Startups, Jul 1).
  • Bull: The market is overreacting, while AI could theoretically replace software (justifying skepticism on "why anyone would pay 15x earnings for CRM companies"), strong incumbents have durable advantages and the sell-off created value at depressed multiples (Mind of a Millionaire, Jun 30). The Halftime desk noted the median S&P software forward P/E is just 16.4x and IGV "bottomed on the same exact day as the [Mag] 7… up 11% versus a 7% return for the Mags" (Halftime Report, Jul 2).

Can incumbents monetize AI agents, or do they cannibalize their own seats?

  • Incumbents win: ServiceNow's CTO, the automation revolution favors "the people who currently own the systems… the ServiceNows, the Workdays, the SAPs," because "we are the system that orchestrates the workers. That is very hard or… impossible for a third party to do." He calls the seat-cannibalization fear "extremely overblown," but concedes "the proof point is when it shows up in our numbers" (alphalist.CTO, Jul 2).
  • Skeptics: The market "doesn't know what to make of it right now", the strong-form bear thesis is "you're just going to vibe code SAP in a weekend." Same source flags ServiceNow "lost 40% of its value" and dropped 7% in a single day (Feb 3) alongside Salesforce on an Anthropic Claude-plugin launch.

Seat-based pricing, durable or dead? Covered above; the live disagreement is whether customers' preference for seat-based predictability (Help Scout's finding) outlasts the economics of AI automating away 20% of unused seats.

Do frontier labs deserve your data, and are they a duopoly worth fighting?

  • Protect your alpha / go open-source: Chamath ran a live benchmark on his 8090 "software factory", wrapping an open model in their harness came out "16.4x cheaper" than Anthropic Opus 4.8 alone (though 3x slower); wrapping Claude itself was "1.4x cheaper and 1.5x faster." His conclusion: continuing to hand data to frontier labs "is kind of becoming derelict and irresponsible" (All-In, Jul 3). Factory's Matan Grinberg: "the biggest threat to OpenAI, Anthropic… is not each other. It's open models," and criticized Anthropic's Fable 5 rollout for mandated data retention and quietly degrading performance on disfavored use cases (The Generalist, Jun 30). 20VC framed the week as "Dario and Anthropic Declare War on Open-Source" (20VC, Jul 2).
  • Frontier labs still win the platform: All-In's own read is that Anthropic's integration flywheel (watch where value accrues on your model, then build the vertical app) is a textbook Microsoft/Google monopoly-adjacent strategy that keeps working.

Hyperscaler capex: rational rerating or malinvestment?

  • Rational: Animal Spirits, the memory rally "is rational. The free cash flow projections for these hyperscalers going to zero… the company is getting rerated and now trades at a forward P/E the same as the market." Not a bubble: "when NVIDIA is trading at 24 times forward earnings, is that expensive? Yeah, fine, whatever. But is it a bubble?… Look what the market is doing to the hyperscalers. It's debubbling them" (Animal Spirits, Jul 1).
  • Skeptic: Fund managers remain "skeptical about hyperscaler AI capex malinvestment and uncertain whether hyperscalers will ever get paid back… while software will get disrupted" (Monetary Matters, Jul 2); Microsoft's FCF deceleration from capex framed as "a long-term risk" (InvestTalk, Jul 2).

Is AI the internet, or the Industrial Revolution? Kleiner's Mamoon Hamid firmly the latter: "It is like the industrial revolution… the railroads… the printing press. It is not the internet," pointing to a "$30–35 trillion" white-collar labor TAM and Anthropic going "from zero to $45 billion revenue run rate… in less than three years" (Masters in Business, Jul 3). ARK's Brett Winton sized frontier models as a "$15–20 trillion… opportunity by 2030" with "likely multiple winners" (Closing Bell, Jul 2). Counterweight: OpenAI's unit economics can't survive public scrutiny, one show pegged OpenAI as needing "$700 billion in revenue by 2035 with 55% gross margin (up from 43% in 2025) to deliver only 10% IRR" (Never Sell, Jun 28), and OpenAI "hitting the brakes" on its IPO was read as a tell (Elon Musk Podcast, Jun 29).

3. Specific Names, bull/bear stances

Public

ServiceNow (NOW), Bull (highest-conviction long of the week). Upgraded to buy at Guggenheim (Jul 1); Stephanie Link added to it. Thesis: "23 times forward estimates" vs. a "5-year average of 54 times," with "total revenues of 20%, subscription revenue of 21%, gross margins in mid-70s, earnings growth of 20%… mission-critical workflow automation" (Halftime Report, Jul 2). Reinforced structurally by its own CTO on the incumbent-orchestration advantage (alphalist.CTO, Jul 2).

Adobe (ADBE), Bear (most-criticized name of the week). "Yet another quarter of ARR growth decelerating… I believe this marks the 10th consecutive quarter this number has gone down." Management is "delaying it [price increases] until after the current CEO leaves," which raises the question "are they worried about churn?… do they feel like they don't have the pricing power they once did?" Trades "10 to 12 times" cash flow/earnings, hit a multi-year low near "$190 a share," now ~$220; $25B buyback authorization; the open question is "is this a melting ice cube?" (The Synopsis, Jul 3). Jay Woods was blunter: "the earnings have been horrible… it could be the next Kodak or Polaroid," citing the CEO and CFO both departing, "those are red flags" (Schwab Network, Jun 29).

Uber (UBER), Bear. Down ~25%; the AV threat is now existential, Waymo pulled out of the Uber app, and "even one competitor, but let alone two competitors [Waymo, Tesla] who want to gain share… could restart the sort of competitive cycle… very destructive to margins," reminiscent of the Uber/Lyft promo wars (The Synopsis, Jul 3).

Palantir (PLTR), Bull. Karp positions the "ontology" application layer as the mandatory safe bridge between LLMs and enterprise/defense data, sees "$15, $18 billion in free cash flow" two years out, and "much more demand than we can supply." The new Nvidia partnership is about customers wanting "control over their compute, their models, their data stack. And their alpha" (Squawk Pod, Jul 1); the Palantir–Nvidia deal was a headline topic on All-In (Jul 3).

CrowdStrike (CRWD), Bull / rich. "Cyber has been crushing it… they're not being treated like software stocks anymore." 4-for-1 split took effect Jul 2. But valuation dispersion is extreme: CRWD, Datadog (DDOG) and Palo Alto (PANW) trade at "127, 91, and 85 times respectively" forward, "that's not the same as 16 times" (Halftime Report, Jul 2).

Oracle (ORCL), Bear / value-trap watch. Woods is "concerned" on "lack of leadership" and the "dual CEO structure" plus debt issuance (Schwab Network, Jun 29); Halftime flagged Oracle at "13 times" as a value-tier software name, and separately noted Oracle "walking back all the reasons that their buildout might not work" (Halftime Report, Jul 2).

Microsoft (MSFT), Mixed. Bought as a 5-year "close my eyes" momentum hold on Animal Spirits, and preferred over memory names for a 3-year horizon ("it's not even close. The hyperscalers"), but flagged as "technically… very concerning" at the $350 level (Schwab Network, Jun 29) and for FCF deceleration/capex intensity limiting buybacks (InvestTalk, Jul 2).

Salesforce (CRM), implied Bear read-through / defensive. Cited as the canonical "system of record" (the defensible archetype) but also the name that fell 7% alongside ServiceNow on the Anthropic Claude-plugin launch (alphalist.CTO, Jul 2); it is acquiring Intercom for $3.6B (below).

Rubrik (RBRK), cautious. A reverse-DCF walk-through showed "growth rate expectations being cut in half as the stock fell," with margin compression from AI investment spend (Chip Stock Investor, Jul 2).

Alphabet (GOOGL) & NVIDIA (NVDA), Bull (relative). Woods' two favorite Mag-7 names: "the two stocks that I personally own and would want to own going into the second half… best potential to continue to grow" (Schwab Network, Jun 29). NVDA "at 24 times forward earnings" framed as not-a-bubble on Animal Spirits (Jul 1).

Meta (META), Mixed / frustrated bull. "Tired of them spending and not showing the results," but the new cloud-business report and Reality Labs restraint could drive a "20%" pop if they pivot to shareholder return (Halftime Report, Jul 2).

Memory complex (MU, SNDK, WDC), the other side of the software trade. "Micron's earnings since January 2025 are up 1,400%… the stock is up 1,400%. It's rational." A "clean transfer" of ~$500B market cap in a single day from Mag-7 to memory suppliers, who now hold "all the pricing power" (Animal Spirits, Jul 1). Woods expects memory to "come back to normal over the coming months" (Schwab Network, Jun 29).

Private

Databricks, Bull (structural). Last valued at $134B, "powering the AI behind something like 70% of the Fortune 500," 20,000+ customers. New "Genie One" agent routes across Anthropic/OpenAI/Gemini/Grok + open models via "Unity AI Gateway" to the "most cost-efficient but effective model," lifting accuracy "north of 85%." Positioned as the neutral platform layer: customers "want choice, control, context, and cost optimization across multiple models", explicitly "not winner-take-all" (The Rundown, Jun 29).

Snowflake (private-context, public SNOW), Bull (repositioning). Its VP of AI framed the pitch as "your data doesn't need a new home, it needs new context", access data where it sits rather than consolidating, via the "Cortex Sense" auto-updating context layer, "Horizon" catalog, and "Coco" coding agent (Zero Prime, Jun 30).

Anthropic, Bull on trajectory, but the week's antagonist. "$45B revenue run rate… in less than three years"; Kleiner invested at the "$900 billion valuation" round (Masters in Business, Jul 3). Bear/antagonist framing: it "blindsided" Figma with Claude Design (its CPO sat on Figma's board, resigning "three days before the launch"), and is running a Microsoft-style vertical land-grab against its own ecosystem (All-In, Jul 3).

OpenAI, Bear (valuation/unit economics). IPO "hitting the brakes" read as an admission its "infrastructure burn rate… eclipses revenue" (Elon Musk Podcast, Jun 29); the "$700B revenue by 2035 for 10% IRR" math on Never Sell (Jun 28).

Cursor / Anysphere, Bull (momentum). Reportedly acquired by SpaceX/xAI at ~$60B, giving it "unlimited compute"; new iOS app puts it "in the top two alongside Codex" (Everyday AI, Jul 2). Risk flag: the "DuneSlide" CVEs (9.8 CVSS) affecting all versions before Cursor 3.0 (The Cybersecurity Defenders, Jul 3).

Figma, Bear (collateral damage). "Figma's stock has fallen something like 50% this year" after Anthropic launched a competing design product (All-In, Jul 3); the CEO is trying to "beat back" the narrative that AI throws the software baby out with the bathwater (The Synopsis, Jul 3).

Harvey, Windsurf, Ambience, Open Evidence, Rogo, Hippocratic, Bull (labor-pyramid theses). Kleiner's map of "AI for [profession]" bets: Harvey (legal, "in most of the AmLaw 100"), Windsurf (dev tools, "acquired by Google"), Ambience/Open Evidence (medicine), then down-pyramid to Rogo (finance), Hippocratic (nursing), Nooks/Revo (sales). The pitch: these sell "actual labor," not software seats (Masters in Business, Jul 3).

Sierra, Bull. Clay Bavor made the case for forward-deployed engineers as "the future of enterprise AI" and the "$100,000 token budget every engineer will need" (20VC, Jul 4).

Intercom, M&A exit. Being acquired by Salesforce for $3.6B (Boardroom Club, Jun 28); its "Finn" product uses outcome-based pricing (The Way of Product, Jul 2).

Bending Spoons, Bull (the SaaS roll-up template). IPO'd this week: priced at "$29 per share, up from its range of 26 to 28… worth about $18.5 billion… last valued at $11 billion," with doubled YoY revenue and GAAP-positive operating and net income in Q1. Flagged as a liquidity vehicle and roll-up model for "pre-AI" SaaS companies (This Week in Startups, Jul 1; also 20VC's "smartest IPO of 2026" framing, 20VC, Jul 2).

Also mentioned: Coinbase "slashed AI spend by 50%," Kalshi at a "$40BN valuation and impending IPO" (20VC, Jul 2); enterprise AI pricing shootout, ChatGPT Enterprise "$60/user/month with uncapped usage" vs. Claude Enterprise "$20/user/month plus per-token," where token costs can balloon "from $25,000 to $60,000–$70,000 annually" (Elon Musk Podcast, Jun 30).