AI Demand Is Still Booming

Key Takeaways

• SemiAnalysis AI spend hits $7 M annualized run‑rate
• Anthropic ARR climbs to $35‑45 B with 72%+ margins
• Memory shortages could double DRAM prices within a year
• H100 GPU clusters last 7–8 years; resale prices surge
• Model hoarding creates power concentration among high‑token spenders

Pulse Analysis

The AI economy is entering a hyper‑growth phase, with corporate budgets soaring and token usage expanding faster than any prior technology wave. SemiAnalysis’s jump to a $7 million annualized spend illustrates how even mid‑size firms are allocating multi‑million dollars to large‑language models. Meanwhile, Anthropic’s revenue leap to $35‑45 billion and margins north of 72% signal that providers can monetize scarcity by raising prices and limiting token caps. This surge is fueling a new economic reality where ideas are cheap but execution—measured in tokens—becomes the primary cost driver.

Supply‑side constraints are quickly becoming the bottleneck. DRAM capacity is growing only 20‑30% annually, prompting analysts to predict memory prices could double or even triple within a year. TSMC’s fab capacity is fully booked, with capital expenditures projected to exceed $100 billion by 2028, while GPUs like Nvidia’s H100 are staying in service for 7‑8 years, inflating secondary‑market values. These pressures translate into higher compute costs for enterprises, reinforcing the advantage of those who can lock in premium model access and amortize hardware spend over longer lifecycles.

For businesses, the strategic imperative is clear: secure the best models, pair them with high‑impact ideas, and out‑spend competitors on token consumption. Model hoarding is already concentrating power among a handful of deep‑pocketed players, a trend that could deepen as robotics breakthroughs—potentially delivering few‑shot learning within the next 18 months—drive a second wave of token demand. At the same time, societal backlash looms, with predictions of AI‑focused protests by late 2026. Companies that demonstrate tangible, productivity‑boosting use cases will not only safeguard market share but also help mitigate the emerging risk of an AI‑driven underclass.