AI News Today: What Makes AI Companies Worth Investing In Today

Two years ago, artificial intelligence companies could raise capital on narrative alone. A compelling vision, an elegant demo, and a promise to “transform everything” were often enough to unlock venture funding. Today, that playbook is obsolete. Investors are no longer underwriting ambition they are underwriting operational proof.

This shift is not cosmetic; it reflects a deeper economic transition. In the Financial Times, Erik Brynjolfsson, director of Stanford Digital Economy Lab, highlights a striking macroeconomic pattern in the United States: slowing job growth alongside resilient GDP. He describes this decoupling sustained output with fewer labor inputs as a hallmark of productivity growth.

If that interpretation holds, we are entering a phase where AI is no longer a speculative technology. It is becoming an economic instrument.

Against this backdrop, the way capital is allocated to AI companies is undergoing a structural recalibration. Conversations with Patrice Mesnier, Founding Partner at Oldenburg Capital Partners, and Julio Martínez, CEO of Abacum, reveal how dramatically investor expectations have evolved and what now separates durable AI businesses from fragile ones.

From Demos to Dependencies

Due diligence has shifted from theatrical to forensic.

Where investors once evaluated model sophistication and interface polish, they now ask a harder question: What breaks if this AI disappears tomorrow?

Mesnier describes a new focus on production usage rather than presentation quality. It is no longer about how impressive a model looks in a demo environment; it is about how deeply it is embedded in a customer’s workflows. The key metric is dependency.

One underappreciated signal is customer behavior under pricing pressure. Many AI businesses face rising inference and compute costs. When companies introduce usage caps or increase prices, investor scrutiny intensifies:

• Do customers meaningfully reduce usage?
• Do they churn?
• Or do they absorb the cost because the product is mission-critical?

If usage collapses under modest price increases, the product was never essential. If customers remain despite higher costs, the AI has crossed a critical threshold: it has become operationally sticky.

This distinction exposes a weakness in parts of the market. Martínez notes that some high-profile AI companies boast extraordinary top-line growth while quietly suffering significant churn. Revenue expansion can mask fragility. If retention fundamentals are weak, long-term value is uncertain regardless of headline valuation.

Growth is no longer enough. Durability is the metric.

The Defensibility Imperative

Another structural shift in investor thinking concerns defensibility.

If a competing model with minimal disruption can replace an AI feature, then the company’s moat is shallow even if it’s branding and UX are strong. The commoditization of foundation models has accelerated this reality. When underlying intelligence is interchangeable, advantage must come from integration and infrastructure.

Mesnier argues that defensibility emerges when AI is embedded into high-risk, high-friction environments: compliance systems, payment rails, logistics platforms, core decision engines. In such contexts, replacement carries operational, regulatory, and financial risk. Switching costs are not marketing-driven; they are structural.

For Abacum, Martínez describes two sources of defensibility.

1. AI-Native Architecture

Companies built natively around AI — rather than retrofitting it into legacy stacks — can design workflows, automation, and data structures from first principles. The result is compounding efficiency. Martínez suggests that modern AI-native companies can ship better products 30% faster with half the resources compared to incumbents constrained by technical debt.

2. Data Integrity and the “Single Source of Truth”

The second moat is data architecture. AI performance is only as reliable as the data it processes. Clean, normalized, centralized data creates institutional trust. Without this foundation, AI becomes a black box layered over fragmented information.

The true value lies not just in model performance, but in data harmonization. Cleaning, reconciling, and structuring enterprise data allows organizations to surface internal dynamics, forecast accurately, and collaborate across departments. In financial planning and analysis, this “single source of truth” is not optional — it is existential.

Companies that own this layer are far more defensible than those that merely wrap generative interfaces around unverified inputs.

Where Capital Is Actually Moving

Despite media focus on consumer AI applications, a significant share of serious capital is flowing toward enterprise and infrastructure.

Mesnier highlights “enabling technologies” as a core investment focus:

• Data pipelines and orchestration layers
• Model infrastructure and deployment tooling
• Proprietary datasets and IP
• Compliance and governance systems

These businesses often lack consumer visibility, but they benefit from long-term contracts, regulatory friction, and deep technical integration. Once embedded, they are difficult to displace. Their defensibility compounds over time.

This dynamic echoes previous technology cycles. Infrastructure appears expensive before it appears obvious.

Oversaturation and the Coming Selection Event

There is widespread discussion of an AI bubble, particularly in consumer-facing verticals. Mesnier does not predict a dramatic collapse. Instead, he anticipates gradual compression:

• Valuations normalize.
• Capital becomes more selective.
• Weaker players fade without spectacle.

The technology itself remains intact. What is changing is pricing discipline.

Martínez raises a sharper concern about vertical oversaturation. Consider customer service AI. Dozens — perhaps 50 or 60 — venture-backed companies pursue similar problems with valuations that far exceed revenue realities. Market mathematics suggests that only a handful can achieve meaningful scale. The rest face consolidation, acquisition at discount, or quiet dissolution.

This pattern mirrors the dot-com era. The internet did not fail; mispriced expectations did. Infrastructure survived. Durable operators survived. Excess evaporated.

The New Standard for AI Companies

The AI market is not collapsing. It is maturing.

The new criteria for survival are clear:

• Demonstrable productivity gains
• High retention under pricing pressure
• Deep workflow integration
• Data ownership and normalization capabilities
• Regulatory or operational switching costs

Narrative capital has been replaced by operational capital.

As infrastructure cycles historically show, early investment phases appear inflated until their necessity becomes undeniable. The strategic error is not investing early — it is assuming that every early investment will endure.

AI is not disappearing. But the era of funding every intelligent interface at venture multiples is ending. Capital is concentrating around companies that move from experimentation to indispensability.

The retrenchment ahead will not destroy the sector. It will define it.