Borrowing to buy GPUs, issuing shares to survive: AI enters the capital race

Just saw some news. ByteDance expanded its syndicated loan from $20 billion to $29.6 billion, and Alibaba launched a $10 billion stock offering. Put these two actions together, and the vibe becomes clear: AI is no longer just a model race; it's entering a heavy-asset cycle.

The evidence is direct too. ByteDance's loan reportedly had an order book exceeding $30 billion, coordinated by Citi, Goldman Sachs, and JPMorgan, with about 20 banks participating. The term is three years, extendable to five. Banks lining up with real money indicates the market is willing to view compute, data centers, power, and depreciation as priceable assets. On Alibaba's side, orders approached 3x the fundraising amount after the placement started, with sovereign funds and long-term capital participating, and key figures buying in.

From an industry cycle perspective, this looks a lot like the early stages of semiconductor expansion. The demand narrative is hot, but ultimately, it comes down to who can turn capex into cash flow. Segments like NVIDIA and TSMC were revalued because capacity, delivery, and ecosystem were continuously bought into.

So when looking at AI companies going forward, I'll pay less attention to demos and more to financing costs, compute procurement, depreciation schedules, and free cash flow. Only those who can convert loans and stocks into stable compute assets have a logic for valuation recovery.