Vlad Saigau posts 1GW $41.6B estimate

Vlad Saigau said on May 14 that 1 gigawatt of AI compute could translate into $41.6 billion in annual revenue globally, according to a post on X that circulated among AI infrastructure and semiconductor investors. Saigau broke that figure into a stack-level split that assigned 48% of value to models and 16% to chips, leaving 36% for the rest of the system, including areas such as storage and networking. The post framed AI economics around power and throughput rather than only around chip unit counts or data-center capital spending. Saigau’s thread was still visible on May 15. ### Where does the $41.6 billion figure sit in the broader AI buildout debate? Goldman Sachs said on May 1 that its baseline model implies $765 billion in annual AI capital expenditure in 2026, rising to $1.6 trillion in annual capex in 2031. The firm said estimates for the total AI buildout over five years have ranged from $4 trillion to $8 trillion, and that the outcome depends heavily on assumptions including silicon replacement cycles, data-center complexity and bottlenecks in power, labor and equipment. McKinsey said in an April 2025 report that data centers would require $6.7 trillion worldwide by 2030 to keep pace with compute demand, including $5.2 trillion tied to AI processing loads and $1.5 trillion for traditional IT. That framing places Saigau’s post inside a wider market argument over how much economic output can be supported by each increment of deployed AI infrastructure. (goldmansachs.com) ### Why did Saigau split the value so heavily toward models? Saigau’s post assigned 48% of stack value to models, more than the 16% he allocated to chips. That implies the largest economic share accrues above the hardware layer, with the remainder spread across the physical and systems infrastructure needed to deliver AI services at scale. NVIDIA Chief Executive Jensen Huang used similar language on the company’s February 25 earnings call, saying “compute equals revenues” and linking architecture and capacity decisions directly to customer earnings. (mckinsey.com) Huang made that remark as NVIDIA reported fiscal 2026 revenue of $215.9 billion and fourth-quarter data-center revenue of $62.3 billion. ### What is included in the 36% outside models and chips? Saigau’s remaining 36% points to the parts of the stack that sit between silicon and end-user revenue: networking, storage, power, cooling, buildings and other data-center systems. Those categories have become more prominent as AI clusters grow in size and power density. McKinsey said the compute-power value chain runs from real-estate developers and utilities to semiconductor firms and cloud providers. (s201.q4cdn.com) Goldman Sachs said the scale of AI investment is highly sensitive to the cost and complexity of next-generation data centers as workloads push power density higher and system integration deeper. ### Does the post line up with other measures of AI economics? Stanford HAI said in its 2026 AI Index that global corporate AI investment more than doubled in 2025, while AI company revenue rose rapidly and compute costs and infrastructure spending also reached record levels. The report said leading frontier companies were reaching meaningful revenue scale quickly even as major cloud providers accelerated capital expenditure. (mckinsey.com) That does not verify Saigau’s exact math, but it does place his estimate alongside a set of widely cited indicators showing that revenue growth, compute deployment and infrastructure spending are now being discussed together rather than as separate markets. NVIDIA’s earnings commentary on February 25 used the same linkage explicitly. (hai.stanford.edu) ### What can readers check next? Saigau’s original X thread is the next place to check for any follow-up assumptions, revisions or replies from other analysts. NVIDIA is scheduled to hold its next quarterly earnings release for fiscal 2027 after the period referenced in its February 25 call materials, and large-cap AI infrastructure spending figures will also be updated through company filings and investor presentations in coming weeks. (hai.stanford.edu) (s201.q4cdn.com)