For the past two years, enterprise AI strategy has largely revolved around one question: which is the best model? As models proliferate, costs rise and geopolitical risks increase, that is becoming the wrong question.
The next phase of enterprise AI will be less about choosing a single winning model and more about knowing which model to use for which task, and being able to switch when circumstances change.
The shift from model selection to workload orchestration
Enterprise AI is becoming less about model selection and more about intelligent workload orchestration.
The first wave of enterprise AI focused on proving that large language models could solve business problems. The second wave is about making those solutions economically sustainable. Organisations are becoming more disciplined about how tokens are consumed, treating them as a managed operational cost rather than an unlimited cloud resource.
The most effective strategy is matching the model to the task. Not every query requires the largest or most expensive model. Routine document classification, extraction and summarisation can often be handled by smaller, cheaper models, reserving premium models for genuinely complex reasoning.
Organisations are also investing in prompt optimisation, retrieval-augmented generation (RAG), caching repeated requests, limiting unnecessary context windows, and introducing governance around who can access premium models.
Perhaps most importantly, enterprises are beginning to measure cost per business outcome rather than cost per token. If a workflow reduces manual compliance effort by 80% or improves decision quality, token expenditure becomes easier to justify. The focus is shifting from simply reducing AI costs to maximising value delivered per token consumed.
The end of vendor loyalty
This is also changing how enterprises think about AI providers.
A year ago, many enterprises selected a single AI provider based largely on model performance. Today, procurement conversations increasingly revolve around economics, flexibility and long-term resilience.
Rather than remaining loyal to one vendor, organisations are adopting a portfolio approach. Different models excel at different tasks, and businesses are recognising there’s little commercial sense in paying flagship prices for workloads that lower-cost models can perform just as effectively.
Vendor decisions are becoming more strategic. Enterprises are evaluating governance capabilities, deployment options, data privacy, integration with existing infrastructure, and explainability, alongside inference costs.
As orchestration layers mature, switching providers is becoming easier, reducing the risk of lock-in. The result is a more competitive market where providers must demonstrate business value alongside benchmark performance. Cost remains important, but enterprise buyers increasingly view flexibility and governance as equally valuable.
Open-weight models change the equation
Lower-cost and open-weight models have consequently become credible options for many enterprise workloads. For internal document search, knowledge management and workflow automation, organisations can often achieve strong results without relying exclusively on premium frontier models.
Open-weight models are becoming increasingly attractive because they give organisations far greater control over how AI is deployed. Unlike closed APIs, enterprises can run open-weight models within their own environments, customise them for specific domains, optimise performance and avoid vendor lock-in.
That flexibility is particularly valuable in regulated industries where data sovereignty, security and governance are non-negotiable.
The discussion should not be framed as open versus closed models, but around giving organisations the right deployment options. Closed frontier models will continue to lead in some areas, while open-weight models are rapidly closing the capability gap at a significantly lower cost.
Security and governance remain critical, however. Enterprises still need to consider where data is processed, how models are updated, regulatory compliance and explainability, often placing these concerns above raw inference costs.
That is driving hybrid architectures. Organisations use open-source or lower-cost models for routine, high-volume tasks while reserving premium models for advanced reasoning or specialist applications. This provides a better balance between performance and economics while reducing dependency on any single provider.
AI is becoming a geopolitical supply-chain risk
The case for flexibility becomes even stronger as AI becomes increasingly geopolitical.
The previous US restrictions on access to frontier AI models reflect a broader shift in how governments view advanced AI. These models are no longer seen purely as commercial technology; they are increasingly regarded as strategic assets with implications for national security, economic competitiveness, and cyber capability.
It is clear that AI governance is moving beyond privacy and ethics into the realm of geopolitical control.
For UK organisations, the immediate concern is less about losing access to one particular model and more about growing uncertainty around AI supply chains. Businesses that build critical workflows around a single provider or jurisdiction may find themselves exposed to policy changes beyond their control.
This reinforces the need for AI strategies that prioritise portability, governance, and explainability over dependence on any one frontier model.
We are also likely to see greater fragmentation of the global AI ecosystem, with different regions applying different rules to access, deployment and oversight. For multinational organisations, that could make compliance more complex and increase the importance of transparent, auditable systems.
For CISOs, AI risk therefore extends beyond model security. They should understand where AI is embedded across the organisation, what external models or services those systems rely on, and what contingency plans exist if access changes unexpectedly.
AI should be governed like any other critical enterprise capability, with clear visibility over data provenance, model dependencies, human oversight, and decision accountability.
Efficiency becomes the advantage
Rising energy costs are unlikely to derail enterprise AI adoption, but they will increase scrutiny around where and how AI is used. Most organisations now view AI as part of their operating infrastructure rather than an experiment, with investments increasingly assessed against cost, governance and business-value metrics.
Higher costs are pushing organisations to focus on efficiency, prioritising AI deployments that reduce manual effort, improve decision-making speed and deliver measurable operational gains.
Enterprises are also paying closer attention to inference costs, consumption-based pricing, and the total cost of ownership, including data preparation, governance and human review. Many are scrutinising AI contracts more closely and demanding greater transparency around usage and costs.
The organisations best positioned to navigate this environment will be those with clear visibility into AI consumption, performance and business outcomes. AI adoption will continue, but the focus is shifting towards disciplined operational management.
The orchestration race
The AI model arms race is entering a new phase.
The question is no longer simply which company has the most powerful model or which model tops the latest benchmark. The competitive advantage will increasingly come from building an architecture capable of using multiple models intelligently.
That means routing workloads according to cost, complexity, security requirements and business value. It means being able to move between providers when economics, capability or geopolitics change. And it means governing the entire system rather than treating the underlying model as the only source of risk.
Ultimately, the conversation has moved beyond choosing the “best” model.
Enterprise AI is becoming an exercise in intelligent workload orchestration, where organisations continuously balance cost, performance, governance and security to optimise both operational efficiency and business value.
The winners will not necessarily be the organisations using the biggest or most expensive models. They will be the ones that know when to use them, and when not to.
