Bank of America is telling clients the AI spending wave is still far from done, even as the bill keeps getting bigger. The firm says combined capital expenditures from hyperscalers will reach about $860 billion in 2025, up 80% from 2024, and it sees the market for AI-related semiconductors expanding to about $1.7 trillion by 2030. The message is clear: the infrastructure buildout is still in the early innings, and the biggest cloud players are not yet close to easing off the accelerator.
For investors, that matters because the AI trade has increasingly shifted from hype to balance sheets. The latest call from BofA, led by Vivek Arya, frames hyperscaler computing power as “mostly supply constrained,” not demand constrained. That helps explain why the market keeps treating AI capex as a durable theme rather than a one-quarter burst. It also keeps the focus on the same beneficiaries: chipmakers, networking gear suppliers and the data center ecosystem tied to the cloud giants.
BofA’s latest math points to an enormous spending base already in place and still rising. The firm’s forecast for about $860 billion in combined hyperscaler capital expenditures in 2025 implies a dramatic jump from 2024. A separate BofA note cited by Marketscreener says AI hyperscaler capex is expected to grow 35% next year. Put together, those estimates suggest the spending cycle is not peaking soon, even after a string of heavy investment plans from the largest technology companies.
That matters because the hyperscaler group has become the market’s clearest proxy for AI demand. These are the companies funding the chips, servers, power systems and networking equipment required to run large-scale AI workloads. When BofA says the market is still supply constrained, it is signaling that the bottleneck remains access to computing power, not a lack of interest from enterprise customers or consumers. In practical terms, that supports more spending to build capacity.
The report also keeps alive a question Wall Street has wrestled with for months: how much of this capex wave can be monetized quickly enough to justify it? BofA’s answer is not that the spending is painless. It says declining free-cash-flow remains a concern. But the firm’s view is that the pressure should be temporary. The analysts wrote that they see negative free-cash-flow margin peaking at around negative 5%-6% in 2027-2028, before likely returning to healthy profitability as AI investments proliferate.
The clearest stock-market implication is for Nvidia, the company most closely tied to the AI infrastructure cycle. BofA raised its Nvidia price objective to $350 from $320 and reiterated a buy rating, underscoring confidence that the chipmaker remains a central winner as hyperscalers keep building. The call fits the broader logic of the note: if the bottleneck is compute, and if hyperscalers are still spending aggressively, then the semiconductor layer remains the most direct lever on that demand.
That is also why BofA’s forecast for the AI-related semiconductor market matters beyond a single stock call. The bank expects that total addressable market to grow roughly threefold to about $1.7 trillion by 2030. A market that size would imply a much deeper and broader hardware ecosystem than the one investors were pricing in only a few years ago. It also suggests the opportunity is not confined to one or two chip names, even if Nvidia remains the clearest headline beneficiary.
Still, the risk for investors is obvious: bigger capex forecasts do not automatically translate into smoother returns. The hyperscaler buildout can strengthen revenue visibility for suppliers while also delaying the moment when customers see a full payback. That is why BofA’s free-cash-flow warning is important. The bank is not arguing that the investment cycle is cheap. It is arguing that the spending remains necessary, and that the payoff may arrive later than the market’s short-term critics expect.
BofA’s projection that AI-related semiconductors could reach about $1.7 trillion by 2030 is the kind of number that keeps market bulls engaged. But the real significance is not the headline figure alone. It is the assumption behind it: that AI workloads will continue to expand across training, inference and broader data-center use cases, forcing cloud operators to keep layering in compute. In other words, the spending cycle is being framed less as a one-time race and more as a structural upgrade of the internet’s infrastructure.
That framing also helps explain why hyperscaler capex forecasts have become so influential. If the largest buyers keep absorbing more hardware, the supply chain gets tighter, not looser. BofA’s view that computing power is still mostly supply constrained implies the industry has not reached a balance point where capacity catches up cleanly with demand. That leaves room for continued investment by the big cloud names and for ongoing pricing power across key semiconductor categories.
Investors should also note the tone of the earnings and valuation debate. AI trade skeptics have argued that capex intensity will eventually punish margins and cash flow. BofA does not dismiss that risk. Instead, it places the pain within a window, saying the negative free-cash-flow margin should max out around negative 5%-6% in 2027-2028. The implication is that investors may need to tolerate a period of heavy spending before the economics look more attractive again.
The next catalyst on BofA’s radar is Computex in Taipei next month, where the firm expects CPU-related announcements, including details on the data center CPU total addressable market. That makes the event important not just for chip headlines, but for the broader AI infrastructure narrative. If vendors use the stage to talk up data center demand, it could reinforce the view that compute needs are still expanding across multiple layers of the stack.
For now, the market takeaway from BofA is straightforward: the hyperscaler buildout is still in overdrive, the semiconductor opportunity remains large, and the profitability debate has been pushed further out. The bank’s revised Nvidia target and its bigger AI market forecast both point in the same direction. The AI spending cycle is not winding down. It is getting more expensive, and Wall Street still thinks the suppliers are worth backing.
