The Protocol Rewriting Software

Here’s the part most investors are skipping. In May 2026, the iShares Expanded Tech-Software Sector ETF (IGV) outpaced the Semiconductor ETF (SMH) on a monthly return basis — SMH logged roughly 18% in May, and software kept pace. That’s notable. But the more important story isn’t the monthly return. It’s what actually caused it.

The “SaaSpocalypse” trade — the idea that AI agents would render per-seat software licensing obsolete — hit a wall in Q1 2026 earnings. And underneath the earnings recovery, something structural is happening that most buy-side models haven’t fully accounted for yet.

Two open protocols are quietly becoming the connective tissue of the entire AI agent economy. One is Anthropic’s Model Context Protocol (MCP), which standardizes how AI agents connect to external tools, data sources, and APIs. The other is Google’s Agent-to-Agent protocol (A2A), which handles how agents communicate with each other across vendor boundaries. Together, they’re doing for AI agents what HTTP did for the web — and the software companies that sit at those connection points are the ones accelerating right now.

What Just Became Infrastructure

MCP launched in November 2024. By late 2025, it had reached 97 million monthly SDK downloads and over 10,000 deployed servers. Anthropic then donated it to the Linux Foundation’s new Agentic AI Foundation — co-founded with OpenAI and Block, with backing from Google, Microsoft, AWS, Cloudflare, and Bloomberg. That’s not a startup protocol anymore. That’s infrastructure.

Forrester projects 30% of enterprise software vendors will ship their own MCP integrations in 2026. Google’s A2A protocol hit v1.0 in production at over 150 organizations, with launch partners including Salesforce, Atlassian, SAP, and ServiceNow. At Google Cloud Next 2026, Thomas Kurian renamed the entire Vertex AI platform the Gemini Enterprise Agent Platform — the whole rebranding was organized around agents, not models. OpenAI now counts enterprise revenue at 40% of total, with Codex reaching three million weekly users.

Slight tangent, but it reframes everything: Gartner predicts 40% of enterprise applications will embed task-specific AI agents by the end of 2026, up from less than 5% in 2025. The multi-agent systems market is growing at a 48.5% compound annual rate through 2030. The AI agents market overall is expected to hit $11.78 billion this year, up from $8.03 billion in 2025 — a 46% jump in a single year. These aren’t rounding errors in some analyst’s model. This is a platform shift moving at speed.

Why This Changes the SaaS Recovery Story

The original bear thesis on software was that AI agents would replace SaaS applications. What’s actually happening is almost the opposite for the right companies. AI agents need Cloudflare’s network to run. They need Snowflake’s data warehouse to query. They need Twilio’s communication APIs to act. MCP has already named Cloudflare as a core infrastructure partner with managed MCP servers running across its edge network. Snowflake’s enterprise marketplace is building a Claude-powered tooling ecosystem. These aren’t coincidences — they’re the protocol layer picking its winners.

The Q1 2026 earnings reflected exactly this split. Cloudflare posted $639.8 million in revenue, up 34% year over year, an acceleration from prior quarters, with full-year guidance targeting roughly 30% growth. Snowflake jumped 29% in a single session after results came in well ahead of expectations, with net revenue retention at 126%. Twilio reported 20% revenue growth year over year — its highest rate in over three years — and added 43,000 net new accounts in a single quarter. MongoDB delivered re-accelerating revenue of 22% with Atlas growing 26%.

These are not the numbers of a sector in structural decline. They’re the numbers of a sector that found out it was infrastructure before the infrastructure wave arrived.

The Three-Way Split

What the market is still pricing as one trade has fractured into three.

• MCP/A2A infrastructure layer (Cloudflare, Snowflake, Twilio): These companies are named partners in the open agent protocol ecosystem. Agents run on their rails. Re-accelerating on the numbers, and structurally positioned for every new AI agent deployment that follows.
• AI-attached application layer (HubSpot, Atlassian): Subscription models being rebuilt around AI as a pricing expansion lever. HubSpot posted 23% revenue growth in Q1 2026. Atlassian has over 600 customers above $1 million in ARR, up nearly 40% year over year. Holding, but dependent on continued AI feature monetization.
• Seat-model incumbents without a moat: The original disruption risk was always here. Pure workflow automation plays without data infrastructure or protocol positioning. These are the ones where the bear thesis still applies — and they may not participate in the same recovery.

Gartner’s global software spending forecast sits at $1.44 trillion for 2026, growing 15.1% year over year — revised upward three times in six months. Enterprise AI agent development costs in 2026 range from $60,000 for midscale pilots to over $300,000 for regulated production implementations, with governance consuming up to 60% of project budgets. That spend goes somewhere. Most of it flows through the infrastructure layer.

What the Price Action Shows

IGV bottomed in early April and staged a meaningful rally off those lows. The fund remains down roughly 11% year to date as of mid-June — semiconductors have lapped the sector on a full-year basis, with SMH up around 57% YTD. That gap is real. But it’s also why the May reversal mattered: for one month, software finally pushed back, and the earnings gave it a reason to.

Cybersecurity names led the charge within software. CrowdStrike and Palo Alto Networks both gained roughly 60% year to date heading into their June reports. Both pulled back sharply after results — a reminder that even strong earnings can fall short when the stock has already moved that far. The Global X Cybersecurity ETF is up about 27% year to date after its strongest single month since launch in May.

Key resistance for IGV sits near the September 2025 highs. A clean break above that on volume would signal a new leadership phase — not just a relief rally. That hasn’t happened yet. But the conditions that would make it happen are showing up in the data more clearly than they were three months ago.

Three Scenarios From Here

• Bull: MCP/A2A adoption accelerates enterprise agent deployment at scale. Infrastructure-layer software companies become the de facto picks-and-shovels of the AI economy. Snowflake, Cloudflare, and Twilio emerge as the platform winners of this cycle. IGV reclaims all-time highs by Q4 2026 as the market re-rates the infrastructure layer correctly.
• Base: The three-way split continues. Infrastructure names extend gains. AI-attached application companies hold growth but don’t expand multiples significantly. Seat-model incumbents without protocol positioning stabilize but don’t recover at the same velocity. Investors get more selective and the index grinds higher without a broad re-rating.
• Bear: Governance complexity and integration costs slow enterprise agent deployments. Over 40% of agentic AI projects are already expected to fail by 2027, per Gartner, primarily due to governance gaps and underestimated scale costs. If the failure rate materializes broadly, agent spending pulls back, infrastructure demand softens, and the sector revisits April lows.

The base case is probably the most intellectually honest. But the bull case has more evidence behind it right now than most models anticipated six months ago.

What investors may be getting wrong: they’re still asking whether AI will kill SaaS. The real question is which software companies become the plumbing of the agentic economy — and whether you own them before the market fully prices that in. The protocol layer already named its partners. The earnings are starting to confirm it. Whether the broader market catches up is the open question nobody can answer cleanly yet.