H1 Tech 2025: Trends That Materialized and Those That Didn't

The reality check

Six months ago, trend reports for 2025 converged on a clear narrative: generative AI everywhere, autonomous agents as the next big thing, cloud costs under control thanks to FinOps, and European regulation (AI Act, DORA, NIS2) transforming operations. Now that half the year has passed, it’s time to review what actually happened.

Not everything played out as predicted. Some trends accelerated faster than expected. Others remain PowerPoint slides.

What materialized

Generative AI normalized (but not as expected)

The prediction was right: generative AI has integrated into the workflows of most technology companies. But the form wasn’t what was forecast. “AI copilots” embedded in existing tools (GitHub Copilot, Notion AI, Salesforce Einstein GPT) won, not standalone AI applications.

The pattern we observe: companies extracting the most value from generative AI apply it to specific, bounded problems (email draft generation, document summarization, code assistance) rather than attempting ambitious “AI transformation” projects. Generic customer service chatbots still have mediocre resolution rates (35-45% without human escalation). Specialized domain-specific assistants work much better.

The most telling data point: according to a McKinsey survey published April 2025, 72% of companies use generative AI in at least one business function, but only 28% report measurable impact on the bottom line. The gap between usage and value remains enormous.

Cloud spending rose (again)

Against predictions of “optimization and stabilization,” global public cloud spending grew 22% year-over-year in H1 2025 according to Synergy Research. AWS, Azure, and GCP all reported double-digit growth. The reason: AI workloads. GPU clusters for model training and inference have become the fastest-growing line item in cloud budgets.

FinOps, which was supposed to be the discipline controlling spending, has seen real adoption but hasn’t slowed absolute growth. It’s made each cloud euro more efficient, but the total number of euros keeps rising because organizations run more workloads every quarter.

For SMEs and mid-size companies, cloud AI costs are the new “surprise on the AWS bill.” We’ve seen companies whose monthly cloud bill jumped 40% in a single quarter just from adding LLM inference endpoints via Bedrock or Vertex AI.

European regulation bit

DORA (Digital Operational Resilience Act) took effect in January 2025 for financial entities. NIS2 transposition began across member states. The AI Act established its first obligations for high-risk AI systems.

The real H1 impact: financial sector companies invested significantly in operational resilience, disaster recovery testing, and third-party risk management. Technology audits saw a demand spike, especially from entities needing to demonstrate compliance.

NIS2 is generating the most concern in the mid-market segment. It affects far more companies than the companies themselves realize (essential and important sectors, which include logistics, manufacturing, food, and waste management, among others). Most aren’t prepared, and penalties are significant.

What didn’t materialize

Autonomous agents didn’t take off

The most overhyped prediction of H1. “2025, the year of AI agents” said Gartner, Forrester, and half the industry. The reality: AI agents operating autonomously in real enterprise environments (not demos) remain experimental. Reliability, cost, and governance problems haven’t been solved.

Where there is progress: semi-autonomous agents executing predefined workflows with human oversight. Our own agent deployments confirm that the model working in production is agent + human-in-the-loop, not autonomous agent.

The main obstacle isn’t model capability. It’s trust. No company wants an AI agent sending an email to a client’s CEO or executing a financial transaction unsupervised. And building appropriate guardrails is significantly more complex than demos suggest.

Edge computing didn’t explode

Every year since 2020, edge computing has been “about to explode.” In H1 2025, it remains relevant for specific use cases (industrial IoT, autonomous vehicles, CDNs) but hasn’t seen the mass adoption analysts predicted. The reason: for most enterprise applications, public cloud latency (10-50ms) is perfectly acceptable, and the complexity of managing distributed edge infrastructure doesn’t compensate.

Mass Kubernetes migration didn’t happen in SMEs

Another perennial favorite. SMEs and mid-size companies are not migrating to Kubernetes. They’re migrating to modern PaaS platforms (Railway, Render, Fly.io) or to serverless (Lambda, Cloud Functions), which offer containerization benefits without K8s operational complexity. Kubernetes continues consolidating as the enterprise standard, but the mid-market is bypassing it.

Unpredicted trends

”Vibe coding” became institutional

Nobody predicted that “vibe coding” (using LLMs to generate complete code from natural language descriptions) would become a serious practice in development teams. Cursor, Windsurf, and similar tools have moved from toys to productivity tools that some teams report saving 2-3 hours daily.

The tension: senior teams use it as an accelerator. Junior teams use it as a crutch. The quality of LLM-generated code without expert review is concerning, and we’re starting to see AI-generated technical debt in production codebases.

Tool fatigue worsened

The number of SaaS tools an average mid-size company uses grew to 130 (Productiv data, Q1 2025). The fatigue is real: teams spending more time integrating tools than using them, data fragmented across 15 platforms, and subscription costs that combined exceed the IT personnel budget.

The reaction: an emerging consolidation trend. Companies actively reducing their tool stack and prioritizing platforms covering multiple functions over best-of-breed solutions for each niche.

What this tells us about H2

If H1 has a dominant theme, it’s the divergence between narrative and reality. Generative AI works, but tangible value remains concentrated in specific use cases. Cloud keeps growing, but cost control isn’t keeping pace. Regulation is biting, but most companies are late to compliance.

For H2, the pragmatic bet: invest in concrete AI applications with measurable ROI (not in “AI strategy”), strengthen compliance for NIS2 and AI Act, and review the cloud bill before the CFO does it for you. If you need an assessment of your infrastructure and operations, our consulting team conducts technology diagnostics that identify priority action points.