Is Software Dying?

The reset

Investors and CIOs are having the same argument from different chairs. On one side, big SaaS names are still growing customers and revenue. On the other, a torrent of AI hype and a rush into infrastructure budgets have knocked roughly $1 trillion off software stocks in a single month — it has been reported that this was the largest single-month drawdown since the 2008 financial crisis. Cuts at companies such as Atlassian and Block — both of which cited AI-driven efficiency or the need to “self‑fund AI investments” — look like panic. At the same time, it has been reported that startups like Anthropic saw dramatic revenue jumps and that NVIDIA (英伟达) CEO Jensen Huang reportedly urged an “OpenClaw” strategy at GTC while previewing an enterprise product called NemoClaw. One group is laying off people. The other is hiring to build AI. Both are right about different things.

Why this feels existential

Western readers should know the core of the story: for two decades the most lucrative enterprise model has been SaaS — cloud software sold on monthly or per‑seat subscriptions. Think Salesforce, Adobe or Feishu (飞书). That per‑seat model is now under pressure. It has been reported that listed SaaS firms’ revenue growth has decelerated quarter after quarter since 2021, and valuation multiples have slid from roughly 35x to about 20x, a compression that wipes out years of froth. Yet many enterprise vendors still post healthy growth — ServiceNow, Salesforce, Snowflake and Oracle all reported expanding revenues and renewals — which shows demand hasn’t vanished. What has changed is investor confidence in re‑accelerating growth.

What’s actually changing — and what isn’t

Ask the simple question: is AI making software useless? No. But AI is changing how companies buy it. Per‑seat billing is loosening. It has been reported that the share of products billed strictly per head has fallen and that 70% of enterprise customers now press vendors for usage‑ or outcome‑based pricing. CIOs are redeploying budgets from bundled tools into AI infrastructure — analysts estimate global cloud providers may spend on the order of $470 billion on AI hardware and related capex this year — and that reallocation is the proximate cause of the valuation haircut. The other reality check: “vibe coding” — AI that writes code from prompts — is hyped, but most AI‑generated projects never reach production; data integration, permissions, compliance and scale remain hard engineering problems. Geopolitics matters too: export controls and chip access shape who can build and who can buy the accelerators powering this shift. So is software dying? No — but a decade of business models and valuations is being re‑priced, fast.