Edward ZitronFounder & CEO

The industry has been sold as transformative software that can cure diseases and replace all labor, but in reality, it is expensive, unprofitable cloud software that remains unreliable. Major technology companies overstate its capabilities and underlying economics while exploiting structural weaknesses in financial journalism, government regulation, and analyst evaluations.

Most reported AI revenue across the largest cloud providers comes directly from funding two unprofitable companies, OpenAI and Anthropic. Cloud hyperscalers inject billions of dollars into these labs, which the labs immediately return by purchasing cloud compute credits. Outside of these subsidized entities, actual global enterprise spend on generative software remains very modest compared to the trillions invested in capital expenditure.

Standard twenty-dollar monthly subscriptions allow users to burn hundreds of dollars worth of compute tokens. On higher tiers, users can consume thousands of dollars in underlying inference costs. Consumer and business adoption numbers appear massive only because providers are selling computing resources at steep financial losses to create the illusion of universal demand.

When providers attempted to shift enterprise clients onto usage-based token pricing rather than flat subscriptions, companies experienced immediate budget shock. Organizations burned through annual software allowances in a matter of months, forcing leadership to reconsider whether the marginal productivity gains justified the actual ongoing infrastructure costs.

Hyperscalers and asset managers are pouring hundreds of billions of dollars into giant GPU facilities based on speculative demand projections. These data centers require immense power and capital, but the underlying compute hardware depreciates rapidly and cannot easily be repurposed for traditional web hosting, leaving operators vulnerable if demand slows down.

Pushing unvetted, auto-generated code into corporate codebases has introduced widespread platform instability and security vulnerabilities across the software industry. Major developer repositories and cloud platforms have experienced increased downtime because engineering teams are reviewing less code and relying on probabilistic models that introduce subtle errors.

Historical technological breakthroughs like the automobile, mobile phones, or early internet protocols offered obvious utility and scalable unit economics that steadily decreased in cost per unit. In contrast, generative language models require exponentially more data, electricity, and expensive training runs to achieve marginal accuracy improvements without lowering operational inference costs.

Benchmark leaderboards evaluate models on narrow, synthetic tests that developers actively train their systems to pass. Performing well on a standardized multiple-choice exam does not translate into autonomous operational capability in complex, real-world business environments with context-dependent edge cases.

The tech industry groups distinct technologies under the single marketing umbrella of artificial intelligence to capture credit for unrelated breakthroughs. Specialized high-performance computing systems like AlphaFold solve concrete biology problems, whereas generative language models simply predict text tokens and require vastly larger amounts of compute with lower determinism.

Corporate leaders promote narratives of imminent labor replacement to boost stock multiples, yet economic data shows zero measurable correlation between token expenditure and revenue per employee. While certain low-cost freelance tasks have been replaced, core white-collar professions require nuanced human judgment, context, and legal accountability that statistical models cannot replace.

Early narratives focused on existential risk to cultivate mysticism, position founders as essential guardians, and deter competitors. As public sentiment turned negative and regulatory scrutiny increased, leadership pivoted toward promising unlimited productivity to justify escalating infrastructure budgets to public market investors.

OpenAI runs at tens of billions of dollars in annual losses and holds massive long-term cloud compute commitments. If the company cannot maintain an escalating private capital cycle or execute a successful public listing, it will struggle to secure the hundreds of billions required to finance ongoing model training runs.

A sharp contraction in AI infrastructure spending would lead to major revenue drops for semiconductor suppliers, downward guidance revisions from cloud hyperscalers, and significant valuation markdowns across institutional portfolios that have concentrated heavily in mega-cap technology equities.

Venture capital returns have struggled significantly because firms celebrate paper valuation markups rather than realized liquidity events. Pouring over half of venture capital into unprofitable wrapper applications built on top of third-party models creates fragile businesses that cannot go public or find strategic acquirers at current valuations.

I would need to see fundamental hardware breakthroughs that drastically reduce compute costs by orders of magnitude, eliminate destructive environmental power drains, and demonstrate software architectures that reliably execute end-to-end tasks without requiring human error mitigation.

True fulfillment comes from supporting and uplifting the real human beings around you rather than chasing speculative trends. Taking time to appreciate collaborators, maintaining authentic relationships, and building honest communities provides enduring stability when speculative market cycles fluctuate.