SAP Freezes Hiring and Travel: AI Costs Are Devouring Corporate Budgets

When the AI Bill Starts Crushing a Software Giant

Software giant SAP recently made a highly symbolic move: in response to surging artificial intelligence-related expenditures, the company suspended most travel and hiring activities. As one of the world's leading enterprise software providers, SAP's decision is not an isolated incident—it reflects the cost dilemma facing the entire industry in the AI arms race.

SAP (Systems, Applications and Products in Data Processing) was founded in 1972 and is headquartered in Walldorf, Germany. It is one of the world's largest enterprise application software providers. Its core ERP (Enterprise Resource Planning) system is used by companies involved in over 77% of global transaction revenue, with customers spanning virtually every industry from manufacturing to financial services—more than 100,000 clients worldwide. In recent years, SAP has been driving a comprehensive transformation from traditional on-premise deployment to cloud-based SaaS models, with its flagship product S/4HANA Cloud becoming a core platform for enterprise digital transformation. It is precisely this massive customer base and product portfolio that makes the scale of AI integration costs particularly staggering.

For a company with tens of thousands of employees and global operations, freezing travel and hiring is no small matter. Such decisions typically indicate that management needs to free up substantial cash flow in the short term to cover a rapidly expanding expenditure item. And right now, the only thing capable of creating such pressure is the astronomical investment in AI infrastructure and R&D.

Why AI Costs Are So Staggering

Computing Power and Infrastructure Investment Continue to Climb

The costs of building and running large-scale AI systems come primarily from three areas: procurement or rental of high-end computing hardware like GPUs, ongoing compute consumption for model training and inference, and engineering talent compensation for the teams surrounding these systems. A single high-performance AI accelerator currently costs tens of thousands of dollars, while the cluster scale needed to train a frontier large model often numbers in the thousands.

Specifically, the core hardware for AI training and inference is currently dominated by NVIDIA. Its H100/H200 series GPUs are priced between $25,000 and $40,000 per card, with the latest B200 series commanding even higher prices. A typical large model training cluster may require thousands or even tens of thousands of such GPUs, with hardware costs alone reaching hundreds of millions of dollars. Beyond NVIDIA, AMD's MI300 series, Google's TPUs, and various custom ASIC chips are also competing for market share. However, due to the deep ecosystem built around CUDA, NVIDIA still holds over 80% market share in AI acceleration, and supply-demand imbalances causing shortages and premiums further drive up enterprise AI infrastructure costs.

More critically, AI costs are not a one-time investment but an ongoing operational expense. Every model inference, every AI feature invocation, is backed by a real compute bill. When enterprises embed AI capabilities throughout their products, these costs accumulate in a near-exponential fashion.

From an economics perspective, AI system computational costs are divided into two phases: Training and Inference. Training is a one-time large-scale computational process—for example, training a GPT-4-class model is estimated to have cost over $100 million. Inference is the computational consumption each time the model processes a user request after deployment. While the per-request cost is far lower than training, the total scales linearly or even super-linearly with user volume. According to industry analysis, inference costs currently account for 60%-70% of total AI compute spending, and this proportion continues to rise as AI features are embedded into everyday products. For a company like SAP with a massive enterprise user base, it may face billions of AI inference calls daily, making cumulative costs extremely substantial.

A Structural Shift from Capital Expenditure to Operating Expenditure

Enterprise software companies like SAP are integrating extensive AI capabilities into their cloud product lines. This means AI costs are shifting from traditional one-time capital expenditures to operating expenses that rise in tandem with user consumption. When customers increase their usage of AI features, SAP's payments to underlying compute providers increase proportionally—and these costs may not be fully passed on to customers.

The difference between Capital Expenditure (CapEx) and Operating Expenditure (OpEx) has profound implications for corporate financial statements. CapEx can be amortized over multiple years through depreciation, creating a smaller impact on current-period profits; OpEx is charged directly to current-period profit and loss, immediately affecting margins. In the traditional software era, R&D investments largely fell under CapEx. But in the cloud computing and AI era, usage-based payments to AWS, Azure, Google Cloud, and others for compute are OpEx—meaning revenue growth must outpace cost growth to maintain margins. This structural shift is also why Wall Street analysts have begun closely monitoring a new metric: "AI gross margin" for tech companies.

This change in cost structure is forcing many enterprises to restructure their financial models. R&D investments that could previously be amortized over many years have now become monthly operational bills that keep growing.

The Strategic Trade-offs Behind Freezing Travel and Hiring

Here's a telling detail: SAP chose to freeze travel and hiring rather than directly cutting AI investment. This trade-off itself speaks volumes—in management's eyes, AI has become a non-negotiable strategic priority, while other routine operational expenses are what needs to give way.

In other words, enterprises are compressing traditional costs to fund their AI transformation. Travel and hiring are relatively easy-to-adjust expense items with manageable short-term impact, making them the first sacrifices when executives balance budgets. This "robbing Peter to pay Paul" approach reflects how the rigidity and urgency of AI investment has exceeded many companies' original financial planning.

The Tech Industry Broadly Faces AI Cost Anxiety

SAP's case is not unique—it's a microcosm of the anxiety pervading the current tech industry. From cloud service giants to software companies, virtually every player is bearing cost pressures far exceeding expectations in their pursuit of AI capability deployment.

The tech industry's current AI investment exhibits classic "arms race" characteristics. Microsoft's capital expenditure in fiscal year 2024 exceeded $50 billion, while Google parent Alphabet and Meta each plan to invest over $30 billion in AI infrastructure. This scale of investment creates powerful competitive pressure: not investing means falling behind in product capabilities, but investing carries uncertain returns. For enterprise software companies like SAP, they must not only integrate AI capabilities into their own products but also compete with Microsoft Copilot, Salesforce Einstein, and other direct competitors for enterprise customers. This dual pressure makes AI investment an unavoidable strategic imperative.

Hidden behind this is a business model puzzle that remains unsolved: AI feature costs are high, but customers' willingness and capacity to pay for AI has a ceiling. When companies race to position AI as a differentiator but struggle to effectively monetize it, the scissors gap between costs and revenue continues to widen, ultimately translating into internal operational belt-tightening.

Short-Term Pain or Long-Term Reality

Optimists believe that as AI chip supply improves, model efficiency increases, and inference costs decline, the current cost pressure is merely short-term pain before the technology matures. Historically, virtually all computing costs have fallen dramatically with scale effects and technological progress.

From a technology roadmap perspective, computing cost reduction has clear historical precedents: cloud storage costs have dropped approximately 95% over the past decade, and CPU computing cost per unit has steadily declined following Moore's Law. For AI inference costs, multiple technical pathways are currently driving reductions: model quantization (reducing parameter precision from FP32 to INT8 or even INT4) can reduce inference costs by 3-4x; knowledge distillation (having small models mimic large model behavior) can dramatically reduce computational requirements while retaining most capabilities; MoE (Mixture of Experts) architectures reduce actual computation per inference through sparse activation. Additionally, the emergence of dedicated inference chips and ongoing algorithmic optimization all suggest that inference costs could drop by an order of magnitude within the next 2-3 years.

But the pessimistic view points out that the capability race for frontier models has no finish line, and leading companies will continue to increase investment to maintain competitiveness, meaning cost pressure may persist long-term. For companies like SAP that need to prove AI's value in the enterprise market, finding a sustainable balance between investment and returns will be the core challenge for years to come.

Practical Implications for Enterprises and Practitioners

For enterprises currently pursuing or planning AI strategies, SAP's case offers several real-world warnings:

• Establish clear cost accounting and revenue models: AI investment must be linked to commercial returns—avoid falling into the "AI for AI's sake" money pit
• Take the continuity of operational costs seriously: Ongoing inference and maintenance expenses deserve more vigilance than one-time investments, requiring long-term financial planning in advance
• Anticipate organizational resource reallocation: AI transformation often comes with unexpected budget squeezes on other departments and roles

For tech professionals, this trend sends complex signals: AI-related positions may be red-hot, but overall corporate hiring may tighten due to AI cost pressures. Understanding corporate cost structures and strategic trade-offs will help you more clearly assess the industry's true direction.

Conclusion

SAP's decision to freeze travel and hiring due to soaring AI costs is a loud wake-up call for the industry. It reminds us that behind the glamour of the AI wave lie real and heavy bills. In an era where everyone is talking about AI transformation, whoever first solves the equation between cost and value will be the one laughing last in this marathon. Until then, an increasing number of enterprises will likely face similarly difficult trade-offs.