Software eats AI, Act II: The vertical feast begins

What happened

Q3 earnings from big AI players revealed a sharp divergence between infrastructure promises and application delivery. While none of the companies break out pure AI revenue, markets rewarded IBM's vertical consulting prowess, Alphabet's search resilience and Amazon's AWS growth as proxy indicators of AI success, while Meta and Microsoft faced pressure despite robust results, primarily due to escalating capital commitments without clear near-term ROI visibility. 

Key developments included Google crossing $100bn quarterly revenue with continued AI integration across search. IBM hitting a record $9.5bn AI book of business. Amazon's AWS exceeding 20% growth, with CEO Jassy highlighting vertical-specific solutions and developer productivity tools. Microsoft reporting sector-specific cloud solutions growing 25-30% vs. 17% for Azure overall. Meta's AI-driven Reels hitting $50bn annual run rate but facing scrutiny over $70-72bn capex guidance.

Impact on our Investment Case

Infrastructure spending hits the wall of diminishing returns

Q3 earnings revealed the harsh reality of AI infrastructure economics. Hyperscalers committed an unprecedented $380bn+ in combined capex for 2025-2026, yet Microsoft saw margins compress despite strong AI metrics, while Meta faces $70-72bn in spending with no clear monetization path. More telling: Amazon emphasized vertical solutions without massive margin compression, while Google monetized AI through existing products. The message is clear: infrastructure alone doesn't create value, applications do.

The math is straightforward: AI's real opportunity isn't some arbitrary software TAM, but the whole labor force awaiting transformation (worth an estimate $11tn in the U.S. alone). According to McKinsey, AI-enabled vertical software can automate 25–50% of the tasks performed by humans in certain roles, enabling vendors to capture a meaningful portion of labor savings through value-based pricing. The actual percentage of economic value captured varies widely by industry, workflow complexity, and pricing power.

Vertical applications outperform despite market noise

The earnings transcripts tell a consistent story: every successful AI deployment is vertical. Amazon emphasizes industry-specific solutions. Google monetizes AI through specialized search applications. Microsoft's healthcare vertical shows 5x growth while general Copilot struggles for adoption. IBM's Arvind Krishna admits "to get real value from AI, people have to be able to integrate their existing applications" and that regulated industries won't use public AI. 

This validates our May thesis: vertical AI will dominate because it solves real problems with domain expertise.

Unlike conversational AI tools like ChatGPT, which aim to be versatile generalists, vertical AI solutions are designed to excel at specific tasks within particular industries, from contract analysis for lawyers to automated medical documentation for healthcare providers.

Will vertical players survive?

Vertical players are mostly still "small" compared to the big horizontal players, and there are legitimate fears that they would be the loosers in direct competition. Indeed, OpenAI recently announced plans to aggressively expand beyond its foundational role as a model provider. The company is leveraging its lead in large language models to build a comprehensive ecosystem of consumer and enterprise products, positioning it as a direct competitor into various verticals, raising the question about the sustainability of the vertical AI pure players.

Despite AI's capabilities, horizontal players face significant barriers, which would require massive investments to be overcome for any single vertical segment: 

• Domain Expertise Moat: when Microsoft's Dragon processes medical records, it must understand that "SOB" means "shortness of breath," not what it means in casual text. This isn't trainable from generic data as it requires years of healthcare-specific development, regulatory compliance, and clinical validation. Access to vertical-specific data is key. No amount of compute power substitutes for this knowledge.

• Trust Barrier: IBM's consulting margins improved precisely because enterprises won't trust generic AI with critical workflows. A legal AI must understand attorney-client privilege. A healthcare AI must be HIPAA-compliant. A financial AI must handle SOX requirements. Horizontal players can't "prompt engineer" their way past regulations.

• Integration Reality: every vertical has unique systems (e.g., Epic in healthcare, SAP in manufacturing, Salesforce in sales. Vertical AI players build native integrations; horizontal players offer APIs and hope. When Abridge connects directly to hospital EMRs while ChatGPT requires copy-paste, the winner is obvious.

• Specific Data: the most valuable training data never touches public clouds. Patient records, legal precedents, financial transactions - this data trains vertical AI models that horizontal players can never replicate. It's not about having more data; it's about having the right data.

General-purpose models will continue to advance and grab headlines, but the real value creation will happen where AI meets industry-specific challenges and workflows.

2026: The vertical feast begins 

History may not repeat, but it definetly rhymes: back in the end '90s - early 2000 all the capex burden was on the telecom operators à la Verizon, Nortel, Marconi - but in the end the real winners of the internet revolution turned out to be the Google, Netflix, Facebook who leveraged the infrastructure to provide applications. 

As horizontal infrastructure commoditizes, three forces will drive massive capital rotation:

1. Horizontal Margin Compression: the diminishing returns of hyperscalers facing massive infrastructure costs will push investors toward more capital-efficient business models.
2. Enterprise Reality Check: as companies realize generic AI doesn't solve specific problems, vertical solutions will see adoption accelerate.
3. Vertical AI IPOs/Acquisitions: Harvey, Abridge, and other vertical AI leaders that are mostly private today will likely go public or be acquired,

Our portfolio is positioned to capture this rotation. On one side we have built exposure to the enablement layer (with names like Datadog, Snowflake, Mongodb, Confluent, Dynatrace): these companies power every vertical AI deployment without picking winners. They're selling blueprints and concrete to everyone building vertical AI skyscrapers.

On the other hand, we are scouting for early vertical leaders, like Pegasystems, or Palantir which in our view is the poster child: already showing 48% growth by combining platform capabilities with increasing vertical focus, thus proving the model works at scale. 

Our Takeaway

Recent Q3 earnings validate our thesis: as AI matures, value flows to those who solve specific problems efficiently. 

Horizontal and vertical AI will coexist, serving different needs. Consumer applications, creative tools, and general productivity benefit from horizontal scale. Enterprise workflows, regulated industries, and specialized tasks require vertical depth. 

Our portfolio captures this evolution perfectly. The enablement layer monetizes all vertical AI deployment today, while positioning for tomorrow's vertical leaders. As infrastructure spending peaks and margins compress, capital will flow toward efficient, specialized solutions.

By early 2026, this rotation will be obvious - and too late.