Accel just dropped their 2025 Globalscape report, and it’s packed with real data on where AI and B2B are actually heading. I’ve been through the whole thing, and here are the 10 takeaways that matter most if you’re building or investing in B2B software.
1. The Great Divergence: AI Infrastructure vs. Enterprise Software Performance
The Real Story: While NASDAQ is up 114% (2020-2025), there’s a massive split in how different software categories are performing.
Market Cap Changes (Oct 2024 vs Oct 2025):
The AI Infrastructure Winners (+$4.9T combined):
- Nvidia: +$1.6T (largest single gain)
- Microsoft: +$0.8T
- Apple: +$0.4T
- Alphabet: +$1.2T
- Amazon: +$0.5T
- Meta: +$0.5T
The Enterprise Software Isn’t Going Away — But It Is Split:
Winners (companies with >10% market cap growth):
- Oracle: +$313B to $785B (+151%) – AI infra play
- Palantir: +$371B to $472B (+27%) – bridging models to business
- IBM: +$94B to $288B (+206%) – AI distribution to enterprise
- Shopify: +$129B to $233B (+80%) – AI-powered commerce
- CrowdStrike: +$61B to $137B (+125%) – AI cyber threats
Losers (negative market cap growth):
- ServiceNow: $197B → $189B (-4%)
- Salesforce: $286B → $214B (-25%)
- Adobe: $239B → $197B (-18%)
What This Actually Means:
The market is pricing in a replacement risk for traditional SaaS. Salesforce lost $72B in market cap despite being profitable and cash-flow positive. Why? Investors are betting that Agentforce and Copilot Studio won’t be enough to defend against AI-native vertical apps.
Meanwhile, companies positioned as infrastructure (Nvidia, Oracle) or distribution (Palantir, IBM) are being rewarded. The belief is: if you control the picks and shovels OR you’re the one helping enterprises deploy AI, you win. If you’re a legacy application, you’re at risk.
For Founders:
Don’t build “SaaS with AI features.” Build AI infrastructure (tooling, orchestration, security) or AI-native applications that make legacy SaaS irrelevant. The middle ground is where market cap goes to die.
The Question to Ask: Is your product more like Palantir (helping enterprises deploy AI) or more like Salesforce (defending existing workflows with AI features)? The market has made its bet on which survives.
2. The Efficiency Gap: AI-Native Companies Operate at 3-12x Revenue per Employee
The Data:
- Cursor: $6.1M ARR per FTE
- Lovable: $3.4M ARR per FTE
- OpenAI: $1.5M ARR per FTE
- Anthropic: $1.2M ARR per FTE
- Traditional SaaS (Atlassian, Datadog, ServiceNow, Salesforce): $0.46-0.54M ARR per FTE
What It Means: AI-native companies are achieving 6-12x better revenue efficiency than traditional SaaS. This isn’t incremental improvement—it’s a complete reimagining of what’s possible with minimal headcount.
For Founders: If you’re not building with this efficiency mindset from day one, you’re leaving massive value on the table. The companies winning today have revenue-per-employee metrics that would have been impossible five years ago.
3. Bottoms-Up Adoption Is Hitting Unprecedented Velocity
The Data:
- Lovable: $100M ARR in 8 months
- Cursor: $500M ARR in 30 months (as of June 2025)
- n8n: 10x YoY revenue growth
- 11ElevenLabs: $200M ARR (2x in 10 months)
- 90% of developers now use AI coding assistants (up from 36% in 2023)
What It Means: Distribution advantages are compounding. Viral adoption through developer communities and social media is creating user bases of millions without traditional enterprise sales motions.
For Founders: The wedge is everything. If your product can go viral with individual users or small teams, you can build massive revenue before you ever need a sales team. Focus on product-led growth and community from day one.
4. But Gross Margins Are Still a Problem (For Now)
The Data: Emerging AI application leaders show gross margins ranging from 7-40%, compared to 76% for the Globalscape Public Cloud Index average.
The Reality Check: Inference costs are eating into margins, but this is temporary:
- GPT-4 pricing dropped 97% from March 2023 ($75/1M tokens) to October 2025 ($2/1M tokens for GPT-5 Mini)
- OpenAI’s GPT-4o is 2x faster, half the price, and has 5x higher rate limits than GPT-4 Turbo
For Founders: Don’t let current margins scare you off. Model costs are plummeting, and by the time you hit scale, your unit economics will look dramatically different. Focus on revenue velocity and product-market fit now; margin expansion will follow as inference costs continue dropping.
5. Enterprise Cloud Growth Rates Are Under Sustained Pressure
The Data: Public cloud companies’ average quarterly growth rates have declined from 47% (Q2 2021) to 15% (Q3 2025).
What’s Happening: Traditional SaaS companies are struggling while AI-native players are thriving:
Mixed Results:
- Winners (lifted by AI): Oracle (+$639B to $785B), Palantir (+$371B to $472B), Shopify (+$94B to $233B)
- Flat/Down (waiting for AI adoption): ServiceNow, Salesforce, Adobe
For Founders: The market is bifurcating. If you’re building traditional SaaS with incremental AI features, you’re in the wrong category. The wins are going to companies that are AI-native from the ground up.
6. Venture Funding Hit All-Time Highs Despite Macro Uncertainty
The Data:
- Total EU/IL/US venture capital in Cloud & AI reached $184B in 2025E (60% of which is AI model funding)
- Cloud & AI application funding (excluding models): $30B combined for EU/IL
- Horizontal AI Apps, Vertical AI, and Developer Tooling are the most well-funded categories
The Split:
- US Model Funding: $106B (2025E)
- EU/IL Model Funding: $4B (2025E)
- But on applications, EU/IL represents 66% of US funding
For Founders: Capital isn’t the constraint—at least not for the right ideas. If you’re building in the application layer (horizontal tools, vertical AI, or developer infrastructure), there’s plenty of capital available globally. The US dominates model funding, but Europe and Israel are competitive in applications.
7. The Race for Compute: $4.1T in AI CapEx by 2030
The Math:
- 117 GW of incremental data center capacity needed by 2030
- ~$4.1T total CapEx required (2026-2030)
- 55%+ of forecasted capacity already committed/announced
- Hyperscalers’ operating cash flow can finance the buildout
The Bottleneck: It’s not capital—it’s electricity. The US faces a 36 GW power shortfall (2025-2028), equivalent to 35 nuclear reactors or solar panels covering an area larger than Los Angeles.
For Founders: Infrastructure constraints are real, but they’re being addressed. Plan for compute to be available but potentially constrained in specific regions. Consider multi-cloud and edge strategies.
8. 45% of Businesses Plan To Increase AI Budgets Further Next Year
The Data: 45% of businesses plan to increase AI budgets by 10-25% over the next 12 months due to interest in agentic AI (per PwC survey).
Early Winners:
- UiPath x Fiserv: 98% end-to-end automation, 12K hours saved
- Decagon x Duolingo: 80% ticket deflection rate, 500M users
- Celonis x Cosentino: >1.8K daily blocked orders reviewed by AI, up to 28-day cycle time reduction
Agentic AI and Computer-Use Models Are the Next Enterprise Frontier
For Founders: Agents are moving from demos to production. The opportunity isn’t just in building agents—it’s in building the guardrails, orchestration layers, and vertical-specific workflows that make agents trustworthy at scale.
9. Tech IPO Market Is Reopening, But We’re Still Far Off The 2019-2021 Pace. For Now
The Data:
- 8 software/AI IPOs in 2025 (vs. 1 in 2023, 4 in 2024)
- Notable 2025 IPOs: Navan, Circle, CoreWeave, Netskope, Via, Figma, Innoscripta, SailPoint
- Public market multiples back above pre-COVID levels: 7.8x EV/NTM revenue (vs 7.1x pre-COVID average)
- The Globalscape Public Cloud Index is up 25% YoY (recovering from the 2022-2023 trough)
What Changed:
After two years in the wilderness (2022-2023 saw almost zero tech IPOs), the public markets are finally rewarding AI execution. But here’s the key: not all software companies are benefiting equally.
The IPOs that are working share three characteristics:
- AI-driven efficiency (revenue per employee >$500K)
- Clear path to profitability (or already profitable)
- Demonstrated enterprise traction (not just consumer viral growth)
Companies like CoreWeave (AI infrastructure) and Netskope (security) are getting premium valuations. Traditional SaaS companies without an AI story are staying private.
For Late-Stage Founders:
The exit window is opening, but it’s selective. If you’re approaching $100M ARR with strong unit economics and a clear AI differentiation story, 2026 could be your year. If you’re a traditional SaaS company waiting for multiples to return to 2021 levels (20x+), you’ll be waiting a while.
The Real Signal: Public market investors are willing to pay up for companies that prove AI actually improves the business model (margins, growth efficiency, customer acquisition costs). They’re not paying for “AI-washed” features bolted onto legacy software.
10. AI Unicorns Are Being Built in Record Time—65% Are 0-3 Years Old
The Data:
- 65% of breakout AI companies are 0-3 years old
- US AI companies average just 2.4 years since founding
- EU/IL AI companies average 4.1 years since founding
- These companies are hiring at 122% YoY (EU/IL) and 213% YoY (US)
New 2025 Unicorns (representing the speed of this shift):
- US: Cursor, Perplexity, Harvey, Sierra, Cognition, Gamma, Filevine, Abridge
- EU/IL: Lovable, n8n, Tines, Framer, Poolside, Parloa, Legora
The Funding Reality: These 200+ breakout AI companies raised $10.3B combined, with funding concentrated in:
- Horizontal AI Apps: $2.2B (21%)
- Vertical AI: $2.2B (21%)
- Developer Tooling: $1.8B (17%)
- Security: $1.4B (13%)
What’s Different This Time:
In the cloud era, it took 5-7 years to build a unicorn. In mobile, 4-6 years.
Now? Companies are hitting $100M+ valuations in 18-36 months. Cursor got to $500M ARR in 30 months. Lovable hit $100M ARR in 8 months.
For Founders:
This is both opportunity and warning. The time window to establish category leadership is compressed. If you’re still in stealth mode after 18 months, you’re already late.
The winners are:
- Launching in public (building in the open)
- Shipping weekly (not quarterly)
- Going horizontal first (or very focused vertical)
- Optimizing for viral loops (not enterprise sales cycles)
The average age of 2.4 years means most of these companies launched after ChatGPT (Nov 2022). They’re not pivots from existing companies—they’re ground-up AI-native builds.
The Real Insight: If you started building in 2023 and you’re not seeing exponential traction by 2025, the market is telling you something. The winners are obvious within 24 months now.
5 More Data Points That Matter
11. AI Security Is Now a Top CISO Priority
- 39% of CISOs rank “Securing AI Agents” as their #1 painpoint (up from essentially zero in 2023)
- AI-augmented SOC automation is #3 priority at 33%
- New attack surfaces (prompt injection, data exfiltration, model poisoning) require entirely new security infrastructure
12. M&A Is Accelerating—But It’s Strategic, Not Financial
- Strategic M&A hit $150B annualized in 2025 (vs. $12B in 2022)
- Biggest deals: Google/Wiz ($32B), Palo Alto/CyberArk ($26.2B), Emerson/Aspentech ($17B)
- PE activity remains strong for legacy software ($22B in 2025E), but the real action is big tech consolidating AI capabilities
13. Foundation Models Show 3% Performance Delta—Video/Computer-Use Models Show 70% Delta
- Text LLMs have converged (Google, Anthropic, OpenAI all within 3% on benchmarks)
- Video generation models show 29% performance gap between leaders (Google) and followers (Genmo AI)
- Computer-use models show 70% gap between Claude Sonnet 4 and UI-TARS-1.5 7B
- Implication: The defensibility is in specialized models, not general-purpose LLMs
15. Vertical AI Is Capturing Services Revenue, Not Just Software Budgets
- Companies like Harvey (legal), Abridge (healthcare), and PermitFlow (construction) are replacing $200K/year professionals with $20K/year AI workflows
- Industries with heavy documentation requirements (legal, healthcare, finance, construction) seeing fastest disruption
- The shift: AI applications aren’t just competing with other software—they’re competing with human labor
15. Hyperscalers Have $5.5T+ in Combined Op Cash Flow to Fund the AI Build-Out
- Amazon, Microsoft, Google, Meta, Apple combined 2026-2030 op cash flow: $5.5T+
- Estimated AI CapEx required 2026-2030: $4.1T
- They don’t need to raise debt—they can self-fund the entire infrastructure build
- For founders: Your compute will be there. The bottleneck is power/data centers, not hyperscaler budgets
The Bottom Line
We’re 24 months into the most significant platform shift since cloud. The winners are moving with velocity we’ve never seen before—$100M ARR in months, not years. Efficiency metrics are 10x better than traditional SaaS. And the market is rewarding execution with unprecedented valuations.
But here’s what’s different from past cycles: this isn’t just about better software. It’s about fundamentally reimagining what’s possible when intelligence is available on-demand, at near-zero marginal cost.
If you’re building, the playbook is clear:
- Go AI-native from day one (not AI-enabled)
- Optimize for revenue-per-employee (not just growth)
- Build for bottoms-up adoption (viral > sales)
- Focus on vertical workflows (not horizontal features)
- Move fast (the window is now)
The infrastructure is being built. The capital is available. The talent is there. The question is: are you building the right thing, fast enough?
Full report available at accel.com/globalscape
