Goldman Sachs: AI Agent Wave Reshapes SaaS Landscape, Widening Gap Between Winners and Losers
Claire Weston
Goldman Sachs calls Q2 2026 the inflection quarter when AI agent activity begins producing measurable impacts on software fundamentals, naming explicit longs and shorts while warning that traditional SaaS models face structural erosion.
Who made Goldman's long and short list?
Turning incrementally positive: Snowflake (SNOW) and Palo Alto Networks (PANW). Turning incrementally negative: Adobe (ADBE), Intuit (INTU), and Workday (WDAY).
Conviction longs — Microsoft, Shopify, Cloudflare, and Twilio — are seen as sitting at the start of a multi-quarter fundamental inflection.
This means → Goldman is not saying "AI lifts all software boats." It is drawing a clear line: platform companies eat; legacy subscription vendors get squeezed.
Why is agent traffic the first domino?
Cloudflare data shows AI agent traffic now accounts for over 50% of all internet traffic, up from just 20-30% in 2025 — hitting that threshold roughly 18 months ahead of Cloudflare's own forecast.
Developer platforms are the first beneficiaries: Cloudflare added nearly 2 million developers in Q2 alone (versus 1.5 million for all of 2025); Vercel's ARR — annual recurring revenue, the subscription income a company can count on each year — re-accelerated to $340 million as of late February 2026.
Shopify's Catalog product tags merchandise attributes with agent-readable parameters. In plain terms = it lets long-tail brands show up disproportionately in AI-powered search, a trend that accelerated sharply in Q2.
SEO is fading — who gets hit hardest?
Search behavior is shifting from traditional SEO to AIEO — AI engine optimization, the practice of making content discoverable by AI models rather than legacy search crawlers — and companies highly dependent on search traffic for customer acquisition are first in the line of fire.
HubSpot is partly pressured by this shift; the company also disclosed accelerating downgrades in its core products. EverCommerce flagged evolving search behavior as a headwind to organic acquisition in H1 2026.
Goldman is incrementally negative on Adobe: the SEO-to-AIEO transition compresses the top of its acquisition funnel, while tools like Gemini with built-in image generation intensify competition. This means → the conversion funnel Adobe built on search traffic now faces a squeeze at both the intake and the product end.
Where is the AI budget coming from — and why is legacy software bleeding?
Goldman's May 2026 CIO survey: only 33% of enterprise AI token budgets come from new allocations; 66% come from reallocating existing budgets — primarily from labor and application-software line items.
In plain terms = companies are not spending new money on AI. They are pulling it out of existing software subscriptions, directly eroding the revenue base of traditional SaaS.
The market has also formed a baseline expectation that new tools should be free to try, with inference costs borne by the software vendor. After Figma Make began charging on March 13, its MCP connection count fell roughly 25% from the May 1 peak to June 1 — once the free tier ended, user behavior turned notably more rational.
How is consumption-based billing changing the forecasting game?
The key driver behind Cloudflare's beat: customers on flex-credit pool plans accelerated spending in Q2. Pool revenue already exceeded 20% of net-new annual contract value by Q4 2025.
This reflects a structural shift — consumption-based billing widens the standard deviation of results, making traditional guidance ranges less useful as reference points.
Snowflake's Cortex Code product is still in early ramp; guidance already bakes in a discount, leaving room for a beat in the early-September print. Microsoft Azure partly beat thanks to GitHub Copilot's shift to variable pricing on June 1 — Goldman notes some customers saw GitHub bills rise 100-200% year-over-year in Q2.
Where is the opportunity in cybersecurity?
Palo Alto Networks is Goldman's top security pick — upside in firewall and Chronosphere, plus targeted M&A positioning against AI-specific risks.
CrowdStrike's fundamental inflection will likely arrive later this year; current valuation expectations are elevated. Okta has explicitly said its agent-identity security product cycle probably won't begin until 2027. This means → Goldman views any pullback in either name as a buy-the-dip opportunity.
Check Point underperformed due to an aging product portfolio and sales-force restructuring; Fortinet beat on firewall price-performance and pre-price-hike demand pull-forward — this reflects a sector where the pace of product iteration is dictating individual stock outcomes.
Content is for reference only, not financial advice.