Investment Case

The infrastructure honeymoon is over. The $1 trillion bill has arrived. For two years, the script was simple. Companies bought Nvidia H100s as fast as possible. Investors cheered every massive capital expenditure. Stock prices rose on the promise of potential. Even Jensen Huang suggested that ROI shouldn’t be forced yet. He was famously jeered for that comment.

The market is no longer in the mood for patience. That era ended in early 2026. Markets no longer reward “buying the future.” They now demand to see the receipts. The “infrastructure honeymoon” is officially over. Every H100 must now justify its existence. If a chip isn’t generating revenue, it’s a liability. This is the brutal math of Phase 2. And it defines the AI Phase 2 Reality Check now facing the entire industry.

Phase 1 was a frantic land grab. Tech giants and startups raced to build compute. They treated GPUs like digital real estate. But real estate is worthless without tenants. In Phase 2, those tenants must pay rent. OpenAI is reportedly under pressure to triple its revenue. Anthropic is chasing a $14 billion annual run rate. The goal is no longer just “smarter models.” The goal is “profitable products.”

Enterprise CIOs are the ones holding the chequebook. They are the “revenue base” underwriting the boom. For a while, they played along with the hype. They ran pilots and experimented with chatbots. Now, the “drug dealer” phase of cheap credit is ending. CIOs are facing aggressive sales targets from vendors. They are pushing back against “black box” pricing. The question shifted from “What can AI do?” to “How does this save us money?” Monetisation proof is the only currency that matters now.

The reality check isn’t just about balance sheets. It is also about the physical world. The “energy wall” has become a major bottleneck. The United States leads in chips and frontier models. However, it is struggling with power generation. Data centres are straining ageing power grids. Regulatory hurdles slow down new transmission lines. Investors are looking toward the Middle East for power. The race for compute is now a race for gigawatts. This is another core dimension of the AI Phase 2 Reality Check, because computing without electricity is just a stranded ambition.

Geopolitics adds another layer of complexity. The US-China AI race has entered a new chapter. Washington maintains a lead in advanced semiconductors. US chips are still five times more powerful than Chinese rivals. But Beijing is playing a different game. China is focusing on AI diffusion and massive scale. By 2030, China could have 400 gigawatts of spare energy capacity. That is four times the current US nuclear capacity. While the US builds the fastest cars, China is building the widest roads.

Supply chains remain dangerously interconnected. The world depends on a fragile triangle. Nvidia designs the chips in America. TSMC manufactures them in Taiwan. ASML provides the machines from the Netherlands. Any disruption in this chain collapses the AI economy. “De-risking” is the buzzword of the year. But moving a $500 billion ecosystem is slow. The threat of conflict remains the ultimate “black swan” event.

We are seeing the rise of “AI bipolarity.” The world is splitting into two technological spheres. One side uses closed-source American frontier models. The other uses open-source Chinese models for mass automation. Each side is building its own “sovereign AI” stacks. Global standards are fracturing along political lines. This makes ROI even harder to achieve for global firms.

Phase 2 is a filter. The “infrastructure honeymoon” allowed everyone to look like a winner. Now, the markets will separate the signal from the noise. Companies that cannot prove monetisation will fade. Venture capital is no longer a bottomless pit. Efficiency is the new growth. The AI Phase 2 Reality Check is painful but necessary. AI is moving from a miracle to a utility. Only the efficient will survive the transition. The bill is on the table. It is time to pay up.