Drive from San Francisco�s airport today and you�ll pass a wall of AI billboards, cranes over new towers, and founders pitching a single story: this time is different? Is it? Under the hood, the AI boom rhymes with the 1990s�yet with three crucial twists that matter for investors and operators: unprecedented capital intensity, a looming power constraint, and market breadth buoyed by companies with no profits?
The new capex cycle � and the power problem
Unlike the web era, today�s AI buildout is an industrial project? The biggest platform companies are set to spend about $342 billion this year in the U?S? alone on data centers and compute, and�at current consumption trajectories�industrywide investment could near $7 trillion by 2030, according to the Financial Times? That dwarfs annual telecom outlays during the dot-com boom even if fiber took years to pay off?
There�s a new chokepoint too: electricity? AI�s economics now hinge on grid upgrades, permitting timelines, and the pace of nuclear and renewables deployment? That puts utilities, energy suppliers, and power markets squarely in AI�s critical path? For CFOs and CIOs, capacity reservations and power-aware workload planning are no longer back-office details; they�re strategy?
Profits up top, risk in the tail
Here�s the paradox? Today�s leaders are cash-rich, equity-funded, and unlikely to implode like WorldCom? Yet the broader market is flashing dot-com-era signals? Since April, negative-earnings companies in the Russell 2000 have outperformed profitable peers, Apollo�s chief economist Torsten Slok has noted�a pattern last seen when investors chased unproven web plays? If AI enthusiasm cools, the wealth effect could be larger than in 2000 because more Americans now hold equities?
Inside the enterprise: adoption is real�and uneven
On the ground, the ROI story is becoming concrete, but not universal? JPMorgan has rolled out an in-house large language model (LLM) assistant to help staff draft year-end performance reviews, advising that outputs be a starting point, not a final decision? It�s part of the bank�s broader LLM Suite deployment, which reached 200,000 users within eight months and now touches code review, pitchbooks, and contract analysis? CEO Jamie Dimon says the bank spends roughly $2 billion annually on AI, within an $18 billion tech budget for 2025?
That�s the upside? The constraint: capability gaps? A Kyndryl survey of 3,700 executives finds 87% expect AI to transform their organizations in 12 months, but only 29% think their workforce has the skills to capitalize? Some 57% cite foundation tech issues that delay deployment; 62% of efforts are still pilots, and just 13% qualify as �pacesetters� moving from vision to scaled execution?
Jobs: task displacement, not a tidal wave (yet)
Labor data across the U?S?, U?K?, and Western Europe show no broad, AI-driven employment collapse? The Financial Times� analysis finds the clearest early impact among freelancers�graphic designers and copywriters�who report lower volumes and pay since generative AI�s breakout? In tech, junior developer roles in the U?S? are down nearly 20% versus late 2022? Bill Gates frames the nuance: AI can replace simple coding tasks today, but not the most complex ones yet? In Sweden, where two-thirds of companies have implemented AI (87% in ICT), 80% say staffing levels haven�t changed�though a fifth of firms planning further deployments expect reductions?
For leaders, the near-term pattern is consistent: AI is reshaping tasks inside jobs faster than it is eliminating whole roles? That favors organizations that redesign workflows and upskill teams quickly?
Pricing and policy friction arrive
Regulators are already probing how AI gets bundled and sold? Australia�s consumer watchdog has sued Microsoft, alleging it hiked Microsoft 365 prices by roughly 30% when adding Copilot while burying a same-price �Classic� option behind a cancellation flow�affecting an estimated 2?7 million people? The case spotlights a coming fault line: providers may need AI-driven price increases to fund capex and power, but customers and regulators may resist opaque upsells?
Follow the money � beyond chatbots
The frenzy isn�t limited to software interfaces? Periodic Labs, founded by former OpenAI and Google Brain researchers, raised a striking $300 million seed round to combine AI, robotics, and simulation for materials discovery�starting with superconductors? The bet: bring experiments into the training loop and compress lab timescales? If successful, this kind of AI-for-science could diversify the ROI base beyond text and image generation?
The playbook for 2025
- Secure power and capacity early; model energy costs in AI unit economics?
- Move pilots to production with explicit ROI guardrails; avoid tool sprawl?
- Re-architect workflows around AI copilot patterns; invest in targeted upskilling?
- Demand pricing transparency from vendors; avoid lock-in via hidden AI �taxes?�
- Stress-test exposure to negative-earnings beta if your equity strategy leans on AI proxies?
Yes, the vibe feels 1999? The difference is that this wave runs on steel, silicon, and megawatts�and its winners will be those who can finance, power, and operationalize it at scale?