Artificial intelligence is producing two seemingly contradictory signals. Trillions of dollars are flowing into data centers, GPUs and power infrastructure, while some of Wall Street’s most sophisticated investors are increasing bets against companies building that capacity. CoreWeave captures the tension: its revenue backlog reached $104.2 billion in the second quarter, yet the Nvidia-backed AI cloud provider was also among the most-shorted stocks tracked by Hazeltree in July.
The divergence does not necessarily mean Wall Street is turning against AI. It suggests investors are beginning to distinguish between companies that can convert unprecedented infrastructure spending into durable cash flow and those whose capital requirements may grow faster than their returns.
Morgan Stanley estimates that roughly $2.9 trillion will be spent on global data center construction between 2025 and 2028, while broader estimates including chips, power and related infrastructure approach $3.5 trillion. AI is no longer behaving like a conventional software investment cycle. It is becoming a capital-intensive industrial buildout measured in trillions of dollars.
$104 Billion in Orders — and Still a Short Target
CoreWeave’s second-quarter revenue jumped 112% from a year earlier to about $2.58 billion, while its backlog rose to $104.2 billion. The company says near-term computing capacity is effectively sold out, and more than $25 billion in additional commitments were signed early in the current quarter.
But expansion comes at a high cost. CoreWeave’s quarterly interest expense climbed to roughly $640 million, while the company raised its 2026 capital expenditure outlook to as much as $39 billion. Strong demand is therefore only half of the equation; the company must generate enough cash from its GPU infrastructure to cover enormous capital, depreciation and financing costs.
That tension helps explain why hedge funds increased short positions in CoreWeave, Nebius and Super Micro Computer in July, according to Hazeltree. They also reduced long exposure and increased shorts in Nvidia, although their overall Nvidia position remained net long.
The signal is more nuanced than a broad rejection of AI. Investors are taking profits in some infrastructure winners and becoming more selective about where AI spending can produce sustainable returns. The question is shifting from how much computing capacity a company can build to how much profit that capacity can generate.
From AI Capex to AI Capital Efficiency
The financing challenge will grow with the industry. Morgan Stanley estimates that around $1.4 trillion of projected data center construction could be funded from hyperscaler cash flows, with another $200 billion from corporate debt and $150 billion from securitized credit. That could leave an $800 billion opportunity for private credit and asset-based financing.
Power presents another constraint. Morgan Stanley estimates U.S. data center demand could reach 74 gigawatts by 2028, with available power potentially falling short by about 49 GW. The economics of AI infrastructure therefore depend not only on access to capital and GPUs, but also on whether companies can secure electricity at a cost that preserves returns.
This changes how the AI boom will be judged. Companies that own the most GPUs or build the largest data centers may not necessarily create the most value. Utilization rates, power costs, financing expenses and the ability to convert long-term contracts into free cash flow are becoming increasingly important.
For South Korea, continued AI infrastructure investment supports demand for high-bandwidth memory from SK hynix and Samsung Electronics as well as opportunities for power equipment and semiconductor suppliers. But if global investors shift from rewarding capacity expansion to demanding returns, customers will also put greater pressure on the cost and efficiency of every layer of the AI supply chain.
CoreWeave’s $104.2 billion backlog and rising hedge-fund short positions capture that transition. Wall Street is not betting that AI will disappear. It is betting that not every company benefiting from the AI investment boom will earn an adequate return on its capital. After years of rewarding more GPUs, larger data centers and higher spending, the next phase of AI will be measured by something simpler: how much cash those trillions of dollars can actually produce.
