Goldman Sachs: Global AI Server Market to Reach $1.24 Trillion by 2030

The demand signals were even stronger in the forward-looking metrics. AI server orders during the quarter reached $24.4 billion, and the company's AI backlog swelled to more than $51.3 billion . This backlog is a critical indicator: it represents signed contracts for servers that cannot yet be shipped, a reflection of the supply constraints in the semiconductor and advanced packaging chain. Dell's executives, citing "no signs of slowing" demand, projected that revenue from AI-optimized servers could hit $60 billion for the fiscal year ending January 2027, an increase from their previous $50 billion estimate .

Cloud and Hyperscale Operators: China's $100+ Billion Blitz

While U.S. cloud providers have been vocal about their AI investments, the scale and velocity of China's capital expenditure surge have become one of the most potent forces in global AI infrastructure demand.

The parent company of TikTok, ByteDance, has committed to spending between $59 and $74 billion on AI capital expenditure in 2026, with internal discussions targeting a figure that could reach $100 billion in 2027 . This is a near-tripling of its 2025 spend, funded in large part by an estimated $50 billion in profit from the previous year . ByteDance's plans alone put it on par with the capital outlays of the largest U.S. technology companies.

Alibaba and Tencent have not been idle. Alibaba signaled it would likely overshoot its original three-year capex target of 380 billion yuan (US$56 billion) to fund the buildout of AI data centers . Combined with Tencent's commitments, their 2026 AI capex alone has been reported at $52 billion and rising . TrendForce projects that the top eight global cloud service providers—which include Alibaba, Tencent, and Baidu alongside Google, AWS, Meta, Microsoft, and Oracle—will invest a combined $710 billion in capex during 2026, a 61% year-over-year increase .

Inference Demand: China’s 140 Trillion Token Economy

The single most vivid illustration of how AI infrastructure demand is compounding comes from China's token consumption. In March 2026, the head of China's National Data Administration, Liu Liehong, told the China Development Forum that the country's daily AI token calls had exceeded 140 trillion, up from just 100 billion in early 2024—a 1,400-fold increase in just over two years . By the end of 2025, the figure had already reached 100 trillion daily, and it grew by another 40% in the first quarter of 2026 alone .

JPMorgan’s long-term forecast puts this trajectory into stark relief: the bank projects China’s AI inference token consumption will grow to approximately 390,000 trillion by 2030, a 370-fold increase from 2025 levels . While a specific "350 trillion by December 2026" figure was not confirmed in available reports, the current official rate of 140 trillion makes that target appear plausible if the 40% quarterly growth rate persists.

This matters for infrastructure because training a model consumes a large, one-time burst of compute, but running that model for hundreds of millions of users creates a permanent, growing base of server demand. Every 140 trillion daily tokens requires a physical fleet of inference servers, and that fleet must scale with usage.

Supply Constraints and the Semiconductor Cascade

The critical insight from the current data is that demand is outstripping supply at every node. Dell's $51.3 billion backlog is not a sales problem; it is a supply problem. Goldman Sachs' Asia team has repeatedly adjusted its AI server shipment forecasts, and the persistent shortage of high-end GPUs and ASICs remains a binding constraint that is expected to keep the supply/demand imbalance extending into 2027 .

Goldman Sachs forecasts that AI chip demand will reach 10 million, 14 million, and 17 million units in 2025, 2026, and 2027, respectively, with custom ASIC contributions rising from 38% to 45% over that period . The shift toward ASICs is a direct response to GPU supply tightness, as companies like Google and the Chinese hyperscalers design their own chips to secure compute capacity.

The effects cascade downstream in ways that are easy to miss but impossible to overstate. Every AI server requires significantly more multi-layer ceramic capacitors (MLCCs), power management ICs, high-bandwidth memory (HBM) modules, and advanced packaging capacity than a traditional server. Goldman Sachs has noted that AI server volumes will grow approximately 4.3 times between 2025 and 2030, which implies a proportional or greater uplift in demand for these electronic component suppliers . The global server buildout is simultaneously a global MLCC, HBM, and power-chip buildout.

The Full-Stack Picture

The data points no longer exist in isolation. The $1.24 trillion server market forecast is the top-level aggregation of a buildout that is visible at every altitude:

• At the hardware layer, Dell's results and backlog show that even the largest server integrators cannot keep up with orders.
• At the cloud layer, the world's eight largest operators investing over $710 billion in a single year demonstrates that this is not a speculative bubble—it is a capacity arms race.
• At the silicon layer, the growing share of ASICs and the upward revisions to chip demand forecasts show that the industry is diversifying its supply chain under constraint.
• At the component layer, the exponential growth in server volumes guarantees sustained demand for the unglamorous but essential parts—capacitors, memory, and power chips—that make a server work.

Goldman Sachs' upward revision of its 2026–2030 forecasts, triggered in part by Dell's single-quarter blowout, reflects a market reality that is now visible in earnings reports, government statistics, and supply-chain backlogs. The bottlenecks are not a sign of weakness. They are a sign that demand is structural, global, and—for the foreseeable future—ahead of anything the supply chain can deliver.