Nomura Calls Tech Selloff a Mistake, Arguing Negative Free Cash Flow Reflects AI Build-Out Timing

On November 2, Nomura Securities put out a research report with a blunt verdict on the recent slide in tech stocks: the market got it wrong. The firm argues that investors saw large technology companies swing to negative free cash flow and read it as proof that AI does not make money. In Nomura's view, that conflates profit with cash, and it underrates how profitable cloud and data center investment actually is. The drop in semiconductor shares, the report says, is an unjustified selloff.

Nomura traces the pressure to two narratives. One is the worry that AI will cannibalize traditional businesses. The other is alarm over weakening free cash flow. On the first, the firm says the fear actually shows that big tech cannot cut AI spending, which is not bad news for chipmakers. On the second, it says the market is mistaking a gap between when capital is spent and when revenue is recognized for a profitability problem.

The core of the argument is a J-curve. Capital spending produces cash outflows in the year it happens, while depreciation and revenue are recognized gradually over the 5 to 20 years after assets enter service. Nomura points to third-quarter results as evidence. AWS revenue reached $46 billion, up 39% from a year earlier and its fastest growth since 2021, with operating margin at 38.7% versus 34.6% a year before. Azure grew 48% year over year, in Microsoft's fiscal 2027 first quarter. Google Cloud brought in $30.7 billion, more than double the prior year, and its operating margin rose to 36.2% from 23.7%. Together the three providers booked roughly $110 billion in cloud revenue, up about 56%. Operating income is still climbing and margins are still rising even as capital spending surges, which Nomura says is far from losing money.

The report then gives three reasons the spending cannot stop. First, performance scales with inference compute: reasoning models and AI agents use far more compute per response than training did, and post-training reinforcement learning is similarly demanding, so too little compute caps commercial scale. Second, revenue is limited by capacity rather than demand. Google Cloud's backlog stood at $560.8 billion at the end of the quarter, AWS's at about $496 billion, and Microsoft's commercial remaining performance obligations at $677 billion, up 73%. Microsoft also disclosed that its actual third-quarter capital expenditure fell short of a $50 billion plan because of third-party capacity delays, which Nomura reads as a supply-side bottleneck. Third, customers are the data: once enterprises embed workflows, data and agents in one model, switching gets costly. Nomura believes the next three years of compute buildout will largely set market structure, so slowing down risks long-term position more than near-term cash flow.

Nomura sketches four paths: accelerate spending, freeze capex, use joint ventures with operating leases, or monetize partially by preselling future cash flows. It judges the likeliest outcome to be continued spending paired with financing tools, not a freeze. Meta's Hyperion project is the precedent for the lease model, with Meta holding 20% and a Blue Owl-managed fund 80%, a $27 billion SPV bond rated A+ by S&P, and a four-year renewable operating lease. Oracle combines structures: leases and SPVs, customer prepayments ($11.4 billion received in the third quarter), and customers buying hardware themselves.

The biggest downside risk, Nomura says, is compute oversupply that leaves revenue per gigawatt below expectations, though it rates the near-term odds as low. In partial oversupply, big tech could lock in prices with long-term contracts, use idle compute for training, and trim capex modestly, with limited effect on chip demand. Even in a severe case, memory makers have about 50% to 70% of shipments locked in under long-term agreements with guaranteed margins, and operating margins estimated above 60%. Current data show little sign of oversupply: cloud growth accelerated across the board, and Microsoft's shortfall reflects too little capacity.

Nomura's conclusion is that negative free cash flow changes neither semiconductor demand nor the valuation logic, and that price falls on that basis are a buying chance for long-term investors. The source notes, though, that the debate is not settled. In the same week, OpenAI reportedly cut its funding target for this round from $30 billion to $20 billion, with about $11 billion in confirmed commitments; S&P Global moved its outlook on Meta from stable to negative; and Anthropic, heading toward a public listing, was flagged as possibly receiving a BB speculative-grade rating. Nomura says to watch cloud growth deceleration, GPU rental prices and backlog growth.