Venture & AI

About this video

Marc Andreessen is a legendary software engineer, entrepreneur, and investor who co-created the Mosaic web browser, co-founded Netscape, and now co-leads Andreessen Horowitz (a16z), one of Silicon Valley’s most influential venture capital firms.

The important AI shift for businesses may not be that intelligence is becoming more powerful. It is that useful intelligence is becoming cheaper, more available and increasingly difficult to monopolise.

Open-source models, aggressive Chinese optimisation and competition between AI providers are already putting pressure on the cost of accessing capable models. If that continues, simply having access to advanced AI will become less of an advantage. The more durable advantage may move toward businesses that can apply abundant intelligence better — through proprietary workflows, trusted customer relationships, specialised knowledge, distribution and execution.

For SME operators, that changes the question from “Which AI should we buy?” to “What do we own that becomes more valuable when intelligence becomes cheap?”

Full Video at the end of page

Core Insight (Plain English)

Much of the AI investment story has assumed that enormous technical capability will produce enormous economic value for the companies building it. That relationship may be less straightforward.

Marc Andreessen points to an important historical precedent: technologies that initially appear scarce and enormously valuable can eventually become infrastructure. Linux did not need to capture the operating-system profit pool to reshape it. By providing a credible open alternative, it helped destroy much of the scarcity supporting proprietary Unix economics.

AI could develop along a similar path.

Open-source models do not necessarily have to become the dominant models. They only need to become good enough alternatives to constrain what proprietary providers can charge. Chinese AI development in this regards, is emerging as a distinct force in the global race. Rather than simply competing at the frontier, Chinese developers are aggressively optimising models to deliver comparable capability at substantially lower cost. Scarcity of leading hardware may even be accelerating this push toward efficiency, turning constraint into advantage.

This creates an unusual competitive dynamic.

American frontier laboratories can continue producing increasingly capable models while open-source and Chinese alternatives follow behind, reducing the price of yesterday's breakthrough. The leader can therefore remain technologically ahead without necessarily preserving the same economic advantage.

That distinction matters far beyond the AI industry.

If model capability becomes increasingly accessible, AI itself becomes less of a differentiator. What businesses build around it becomes more important.

What this means for operators

1. Do not confuse access to AI with competitive advantage
Buying access to a leading model may improve productivity, but competitors can increasingly buy similar capability. The competitive advantage is more likely to come from how AI is embedded into the business: internal knowledge, customer history, operating processes, specialist expertise and decision-making.

Operators should therefore ask what becomes difficult for a competitor to reproduce after both companies have access to similar AI.

2. Avoid building unnecessary dependence on one model provider
The AI market remains unusually unsettled, this raises possibilities ranging from proprietary model dominance to open-source commoditisation, hardware capturing more value, or specialised applications becoming the important layer.

SMEs do not need to predict which outcome wins. They can instead preserve flexibility.

Where practical, workflows should be designed so that changing model providers does not require rebuilding the entire process. The cheaper and more interchangeable models become, the more valuable that flexibility becomes.

3. Existing software should be judged by the workflow it owns, not how many AI features it adds
Adding AI to an existing product does not automatically protect that product. The more important question is whether AI strengthens the existing workflow — or eliminates the reason customers needed that workflow in the first place.

This applies well beyond creative software. Operators reviewing their software stack should distinguish between systems deeply embedded in how the business operates and productivity tools whose main function AI may increasingly perform directly.

That could eventually change where software budgets deserve to go.

4. Cheap intelligence could disproportionately benefit smaller companies
SMEs historically operate with fewer specialists than large organisations. A large company may have analysts, developers, researchers, designers, lawyers, marketers and operational specialists available internally. Smaller businesses often cannot justify that overhead.

AI potentially lowers the cost of accessing parts of those capabilities. That does not eliminate the need for professional expertise, particularly in regulated or high-risk decisions. But it could narrow part of the capability gap between large and small organisations.

The opportunity for SMEs is therefore not simply reducing headcount. It is attempting work that previously required resources the company could not economically justify.

5. Leadership and organisational speed may matter more than technology selection
Technology does not determine outcomes by itself. Some incumbent software companies may fail because they respond too slowly. Others may use AI to strengthen existing products and restart growth.

The same principle applies to SMEs. Two competitors can receive access to essentially the same technology and produce very different outcomes. Management still decides which workflows to redesign, where experimentation is acceptable, where human judgement remains necessary and how quickly successful experiments spread through the organisation.

As AI becomes more widely available, the bottleneck may increasingly move from access to execution.

6. Falling AI costs could matter more than headline capability improvements
Businesses naturally pay attention when a new model performs something that the previous generation could not.

Operators should pay equal attention to another curve: how much yesterday's capability now costs. A model becoming 10% better may have little immediate impact on an SME.

A sufficiently capable model becoming dramatically cheaper could make thousands of previously uneconomic workflows viable. That may ultimately be the more important commercial signal.

Practical watchpoints

Operators should watch five developments particularly closely.

Model price versus capability.
Track whether acceptable AI performance continues becoming cheaper. Falling cost can change the economics of automation even when capability improvements appear incremental.

Open-source performance.
Open source does not need to lead the market to influence it. The important question is whether open alternatives remain sufficiently close to proprietary models to constrain pricing.

Software pricing and bundling.
Watch whether existing SaaS providers charge significant premiums for AI functionality or increasingly include it within standard subscriptions. Bundling would be another indication that AI features themselves are becoming commoditised.

Workflow displacement rather than feature adoption.
Do not only monitor which applications introduce AI. Watch whether customers begin bypassing applications entirely because AI can accomplish the underlying task directly.

Competitor operating behaviour.
The most meaningful competitive signal may not be competitors announcing AI strategies. Watch for shorter response times, smaller teams handling larger workloads, faster product development or significantly different service economics.

Summary & Reflections

There is good reason to be cautious about taking the commoditisation argument too far.

Frontier AI remains extremely expensive to develop. Leading laboratories may preserve substantial advantages through scale, infrastructure, proprietary technology, distribution and continual improvements in capability.

Open-source alternatives may also remain dependent on innovations originating from expensive frontier development. There is therefore no certainty that AI models become interchangeable commodities. But operators do not need that extreme outcome for the business implication to matter.

If credible alternatives consistently emerge at lower prices, even with some delay, they can limit the economic scarcity of frontier capability. Open source does not have to defeat proprietary AI; it only has to remain credible enough to influence its price.

There is another reason for caution. Andreessen's broader argument about regulation suppressing productivity is strongly shaped by his own investment and political perspective. His historical explanation should therefore be treated as an argument rather than settled economic fact.

The more useful business signal is narrower: technical leadership, pricing power and value capture are not necessarily the same thing.

That distinction is worth watching.

Regional Consideration — Southeast Asia
For Southeast Asian SMEs, cheaper and more accessible models could be particularly important because the region contains large numbers of businesses operating without deep internal technology teams.

At the same time, Southeast Asia is fragmented across languages, regulation, labour costs and levels of digital adoption. The economics of replacing or augmenting human work will therefore differ substantially between Singapore, Indonesia, Vietnam, Malaysia and other markets.

Cheap AI capability may spread globally. Its business value will remain local.

Who should watch the full video

The full discussion is particularly relevant for:

• SME owners deciding where AI belongs in their operations

• Founders building software or AI-enabled businesses

• Technology decision-makers reviewing long-term platform dependencies

• Product leaders assessing whether AI strengthens or threatens existing products

• Business leaders evaluating SaaS expenditure and automation

• Investors and operators thinking about where value may migrate as AI infrastructure matures

Decision Rating

Decision Usefulness — ★★★★☆
The discussion does not provide a ready-made AI implementation playbook, but it offers a useful strategic lens for technology decisions: avoid assuming today's model leaders, pricing structures or software categories will remain stable.

Technology Relevance — ★★★★★
The signal cuts directly across model selection, SaaS dependency, open-source alternatives and AI architecture. It is particularly useful for operators making technology commitments that may need to survive rapid changes in price and capability.

Timing Sensitivity — ★★★★☆
Operators do not need to restructure their businesses immediately, but the economics are moving quickly enough to justify regular reassessment. Decisions that create expensive long-term dependence on today's AI landscape deserve particular scrutiny.

Until next time,
The SME Signal editorial Team

Contact

Questions? Reach out anytime.

Email

hello@smesignal.com

© 2026. All rights reserved.