AI prices need to drop 90 percent, warns Arora

Introduction: A wake-up call from the heart of the tech industry

Nikesh Arora, CEO of Palo Alto Networks, one of the most influential cybersecurity companies globally, recently released a statement that sparked heated debates in the international technology community. According to him, Current prices of services and solutions based on artificial intelligence must decrease by at least 90 percent for large-scale adoption to become a sustainable economic reality. This statement does not come from a vacuum, but from the deep perspective of a leader who manages billions of dollars in budgets every day and who sees firsthand how high costs block digital transformation in organizations of all sizes. His message is clear: AI is technologically fantastic, but financially inaccessible for most market players.

The economic context of current AI prices

To understand why Arora's statement has such great relevance, we need to analyze the cost structure of AI solutions available on the market today. Large Language Models (LLM), such as GPT-4, Claude 3 or Gemini Ultra, involve huge infrastructure costs: from the initial training which can cost hundreds of millions of dollars, to the inference costs, i.e. processing requests in real time, which accumulate quickly at enterprise scale. Companies today pay per token, per API call, per processing session, and these costs add up exponentially when we talk about industrial-scale applications.

In addition, hardware costs associated with high-performance GPUs, especially those produced by NVIDIA (the H100 and H200 series), have kept prices prohibitive. A cluster of GPUs required to run a mid-sized model can cost between $500.000 and $5 million, not including the costs of electricity, cooling, and maintenance. These economic realities mean that AI adoption is currently the preserve of large corporations and governments with generous budgets, leaving SMEs and startups in developing countries behind.

Why 90 percent? The math behind the statement

Inference cost analysis

The claim that prices need to fall by 90% is not arbitrary. Arora is referring to a well-known principle in the semiconductor industry and technology in general: Moore's Law and its modern equivalent applied to AIHistorically, processing costs have fallen dramatically over time. In the case of AI, analysts at ARK Invest and other research firms have projected that the cost of training a top AI model will drop by about 70-80% every two years, thanks to advances in chip architecture, algorithmic optimizations, and streamlining of distributed training processes.

However, these cost reductions are not automatically passed on to end consumers. Large service providers cloud – AWS, Microsoft Azure and Google Cloud – they maintain their margins high, partly due to explosive demand that exceeds supply, but also due to pricing strategies that favor customer lock-in to their proprietary ecosystems. Arora suggests the market needs to reach an inflection point where fierce competition and the democratization of access to hardware will force suppliers to reduce prices dramatically – and this point could be closer than we think.

Comparison with the evolution of other technologies

A relevant example is the evolution of prices in the industry cloud computingIn 2006, when Amazon launched AWS, the cost of storing a gigabyte of data was about 15 cents per month. Today, that same gigabyte costs less than a cent, a reduction of more than 90%. The same trajectory is expected for AI, and Arora believes that this price compression is not only possible, but necessary to unlock the next wave of economic value generated by artificial intelligence.

Implications for the cybersecurity industry

Arora's perspective is not disinterested. As the leader of Palo Alto Networks, he operates in a field where AI has become an essential tool for both defenders and attackers. AI-based cybersecurity It involves real-time processing of huge volumes of network data, detection of behavioral anomalies, analysis of advanced persistent threats (APTs) and automated response to security incidents. All of these capabilities consume massive computational resources.

If AI prices remain high, smaller organizations will have to choose between security and other business priorities, an unacceptable choice in an era where cyberattacks are becoming increasingly sophisticated and frequent. Arora argues that a substantial reduction in the costs of AI would allow the democratization of advanced security, protecting not only large corporations but also hospitals, schools, government agencies and small businesses that are today easy targets for cybercrime groups.

Factors that can accelerate the decline in prices

The emergence of competitive open-source models

A major factor putting pressure on AI prices is the rise of high-performance open-source models. Meta AI with the Llama series, Mistral AI with its European models, or DeepSeek from China have demonstrated that it is possible to build extremely high-performance models at a fraction of the cost of proprietary models. These models can be run on-premise, completely eliminating the dependency on APIs cloud costly and giving organizations total control over data and costs.

In 2025 and 2026, we are witnessing an accelerated maturation of the open-source AI ecosystem. Techniques such as quantization (reducing the numerical precision of a model's weights), pruning (removing redundant neurons), and distillation (transferring knowledge from a large model to a smaller one) allow capable models to run on general consumer hardware, dramatically lowering the barrier to entry. A 4-bit quantized model can run on a regular laptop with an integrated GPU, which was unthinkable two years ago.

Competition in the AI ​​hardware space

NVIDIA's dominance in the AI ​​GPU segment is starting to be seriously challenged. AMD with the Instinct MI300 series, Intel with Gaudi 3, Google with the v5 TPUs and startups like Groq, Cerebras and Tenstorrent offers increasingly viable alternatives. In addition, chip manufacturers in Asia, especially in Taiwan and South Korea, are investing heavily in developing their own AI accelerators. This diversification of hardware offerings will significantly contribute to reducing inference and training costs in the medium term.

Streamlining algorithms and architectures

Advances in fundamental AI research are also contributing to cost reductions. Mixture of Experts (MoE) architectures, which activate only a subset of the neural network for each request instead of using the entire computational capacity, significantly reduce resource consumption. Similarly, techniques such as Flash Attention, speculative decoding and continuous batching optimizes the use of GPU resources in large-scale inference scenarios. Every algorithmic advance of this type translates directly into cost reductions for end users.

The reaction of the market and the big players

Arora's statement comes at a time when OpenAI, Anthropic and Google DeepMind are facing increasing pressure from investors to demonstrate that their business models are viable in the long term. The operating costs of these companies are astronomical: OpenAI is estimated to spend over a billion dollars a year on computing infrastructure alone. In this context, reducing prices by 90% would mean either a major technological revolution in computational efficiency or a drastic consolidation of the market that would eliminate inefficient players.

Microsoft, which has invested heavily in OpenAI and is integrating AI across its entire product suite Office 365 and Azure, has already started adjusting the prices of some AI services to attract reluctant enterprise customers. Co-pilot for Microsoft 365, which initially launched at $30 per user per month, has seen a series of pricing adjustments and new packages that reflect increasing competitive pressure. This trend at least partially confirms Arora's prediction.

Romanian Perspective: What does this mean for companies in Romania?

For the Romanian business environment, reducing AI prices by 90% would represent a tremendous opportunity for digital transformation. Currently, the adoption of AI solutions in Romanian companies is still at an early stage, partly due to high costs, but also due to the lack of local experts capable of implementing and optimizing these technologies. A dramatic reduction in prices would remove one of the main barriers and allow mid-sized companies to access advanced AI capabilities.

Sectors such as industrial production, precision agriculture, digital health and public administration could benefit enormously from affordable AI. Automating repetitive processes, predictive analytics of operational data, fraud detection in financial systems, and optimizing supply chains are just a few of the applications that would become economically feasible for Romanian organizations. However, to capitalize on these opportunities, it is essential that the workforce is properly trained and that companies invest in professional training in the field of AI.

The risks associated with a rapid price decline

Not all the consequences of a drastic reduction in AI prices are positive. A rapid democratization of access to powerful AI It also raises serious issues related to security, disinformation and misuse. If synthetic content generation models (deepfake video, audio, text) become accessible at minimal cost, the potential for manipulation of public opinion and cyber fraud will increase exponentially.

A cheaper AI market can also accelerate job displacement in sensitive sectors: customer support, data analysis, content writing, translation, even certain aspects of software programming. Governments and international organizations will need to respond with appropriate social policies, professional retraining programs, and regulations that ensure a fair transition to the AI ​​economy.

Conclusion: The inevitability of falling AI prices

Regardless of the perspective from which we view the problem, Nikesh Arora's conclusion reflects an economic and technological reality that is difficult to dispute.: Current prices for AI solutions are unsustainable in the long term and represent a major barrier to global adoption. Market forces – increasing competition, advances in hardware and algorithms, the rise of open source – are combining to create inevitable downward pressure on prices.

The question is not whether AI prices will fall, but how quickly will this happen and who will benefit most from this transformation. Companies and individuals who prepare today, investing in education, a deep understanding of AI technologies, and the development of practical skills, will be best positioned to capture the extraordinary value that AI promises. Arora’s statement should be a wake-up call to anyone who still views AI as a fad of the distant future: the future is already here, and the price of entry is about to plummet.

You have certainly understood what is new in 2026 related to artificial intelligence. If you are interested in deepening your knowledge in the field, we invite you to explore our range of courses structured by roles and categories in AI HUBWhether you're just starting out or want to brush up on your skills, we have a course for you.