AI Spending Under Scrutiny: From Hype to Hard Questions
AI investment skepticism is the growing view that current enterprise AI spending, from cloud compute to tokenmaxxing costs, may exceed the real business value created, forcing leaders to question whether experimental deployments and headline valuations are justified by measurable productivity gains or lasting revenue growth. This concern is moving from fringe commentary to boardroom agenda. Zoho founder Sridhar Vembu calls the current surge “clearly an investment bubble”, arguing that enormous AI valuations and cloud backlogs may reflect financial engineering more than sustainable demand. Meanwhile, corporate AI usage is exploding across functions, from coding assistants to internal copilots. Yet many organizations admit they lack rigorous AI ROI measurement frameworks and often treat spending as a strategic tax rather than a line item that must pay for itself. The result is a widening gap between AI ambition and proven outcomes.
Bubble Fears and the Circular Revenue Loop
Sridhar Vembu’s warning focuses on how AI revenue can be propped up by circular money flows between cloud giants and the startups they fund. Investments in firms like OpenAI and Anthropic often arrive as cloud credits, which are then spent back on the investor’s own infrastructure and booked as fresh revenue. According to OfficeChai, corporate filings suggest OpenAI and Anthropic account for more than half of a combined USD 2 trillion (approx. RM9.2 trillion) future cloud backlog at several major providers. Tech giants also report large paper gains by marking up stakes as startup valuations climb, even as they pour real capital into data centers. Critics see echoes of the dot-com era, where revenue swaps disguised weak demand. The concern is that enterprise AI spending may rest on accounting loops and speculative valuations rather than durable customer need.
Tokenmaxxing Costs and Uber’s Productivity Puzzle
While investors debate an AI investment bubble, operators are wrestling with tokenmaxxing costs. Tokenmaxxing describes pushing AI systems to process huge numbers of tokens to drive productivity, even when the payoff is uncertain. Uber COO Andrew Macdonald says he has not seen a clear link between higher token usage and concrete gains, noting that it is hard to claim “25% more useful consumer features” from internal stats alone. Meanwhile, reports that Uber burned through its annual AI budget in four months have amplified concerns that enterprise AI spending is outpacing discipline. Engineers and founders complain that “millions of dollars” worth of tokens have been consumed with little proven ROI. CIOs, Google’s Sundar Pichai says, are increasingly worried about how quickly budgets are being depleted, suggesting a looming backlash against unbounded experimentation.
Salesforce, Uber and the Search for AI ROI Measurement
Companies like Salesforce and Uber illustrate the tension between aggressive AI adoption and immature AI ROI measurement. Salesforce has rolled out agentic coding tools across engineering, only to discover that its initial token budget was wildly optimistic as usage exploded. Uber’s experience with rapid budget overruns mirrors this pattern: once AI tooling becomes available, teams default to tokenmaxxing without a clear economic model. Many enterprises are still unsure when to treat AI as a cost of innovation and when to demand profitability. They track adoption metrics, token counts and feature output, but struggle to isolate how much new business value AI systems generate. As budgets swell, finance leaders are pushing for cost-per-feature benchmarks and payback periods. The debate is narrowing from “Should we invest in AI?” to “Which AI workflows deserve sustained spending, and which should be cut?”.
From Experimental AI to Proven Enterprise Value
The broader industry now faces pressure to prove that enterprise AI spending can stand on solid economic ground. Early adoption focused on experimental AI implementations, internal pilots and headline partnerships that signaled technological leadership more than financial discipline. Now, skepticism from voices like Vembu and Macdonald is forcing leaders to sort signal from noise. Investors want assurance that cloud backlogs are backed by real customer demand, not recycled credits and paper gains. Operators want clarity on which AI agents, copilots and coding tools meaningfully boost productivity once tokenmaxxing costs are accounted for. The next phase of AI will likely reward companies that build granular ROI models, set firm budget guardrails and shut down underperforming use cases. Without that shift, doubts about an AI investment bubble will grow, and the current spending surge could quickly lose its momentum.