Is Google Cloud Subsidizing One Of Its Biggest AI Rivals?

The global playing field for artificial intelligence infrastructure just experienced an unexpected shakeup. SpaceX has suddenly emerged as a major player in the cloud computing market. Wall Street observers are already calling this new infrastructure wing ‘Elon Web Services’ (EWS). By securing back-to-back mega deals with both Anthropic and Google, SpaceX has fundamentally altered the financial expectations surrounding its upcoming public market debut. But while this arrangement provides an incredible financial cushion for SpaceX, it exposes some incredibly worrisome trends for Google.

This looks less like a triumphant partnership and more like an emergency rescue mission for a tech giant caught flat-footed by its own planning failures. To understand how we arrived here, we must look at the mechanical issues that plagued xAI. xAI’s Colossus 1 facility in Memphis was assembled using an eclectic, highly mismatched collection of NVIDIA hardware architectures. It was a mess. This configuration severely throttled model training efficiency. Internal communications revealed a disappointing 11 percent Model FLOPs Utilization rate. Because the engineering team could not seamlessly parallelize training workloads for their Grok model, xAI chose to abandon Colossus 1 for primary training purposes. They shifted those intensive operations over to their newer Colossus 2 facility. Rather than allowing billions of dollars of high-grade silicon to sit completely idle, SpaceX stepped in to monetize the infrastructure. Musk moved fast. It transformed an operational headache into a multi-billion-dollar revenue stream.

For SpaceX, the timing is immaculate. The company recently filed its initial documentation to list its shares on the Nasdaq exchange under the ticker symbol SPCX. The company is chasing a massive valuation between 1.75 and 1.8 trillion dollars, aiming to raise up to 75 billion dollars in the process. Before these cloud leasing deals came to light, public market investors were highly focused on the company losses. SpaceX lost a staggering 4.9 billion dollars in 2025 because of an aggressive 20.7 billion dollar capital expenditure cycle. Even with healthy revenue from Starlink, the high costs of rocket development and satellite manufacturing weighed on the balance sheet.

The creation of what many are calling ‘Elon Web Services’ completely changes that narrative. SpaceX can now claim a high margin, highly predictable recurring cloud infrastructure revenue stream. This allows them to seek a software-style valuation multiple rather than being judged purely as a capital-intensive aerospace manufacturer. Locking in over two billion dollars a month in combined revenue from Anthropic and Google provides immediate financial insulation. It gives public investors clear sight lines to massive cash injections that offset heavy manufacturing costs. Even with 90 day cancellation clauses hanging over the contracts, it creates a comforting financial buffer for the public offering.

Now consider the Google side of this equation, where the outlook is far less rosy. The optics are terrible. Google has spent the last decade positioning itself as a pioneer in custom silicon design. Yet they are now cutting a near billion dollar monthly check to their direct competitor for access to standard NVIDIA chips. We view this as a clear admission of strategic failure. Google claims this is a calculated move to ease an internal capacity crunch. The reality is that their planning failed. Demand for their Gemini Enterprise platform apparently outpaced internal expectations. To make matters worse, Google recently allocated roughly 40 percent of its newest internal hardware capacity directly to Anthropic. We propose that this situation has caused massive allocation, triggered an immediate, chaotic internal scramble for compute power.

Google is essentially paying an exorbitant premium to mask its own supply chain mismanagement. Building physical data centers and securing the necessary electrical grid power is a slow process. Google is currently looking at an 18 to 24 month queue to bring new raw data center capacity online. Leasing 110,000 NVIDIA GPUs from SpaceX is a desperate stopgap measure designed to bypass that timeline. The EWS cloud environment aims to deliver immediate computational capacity to external clients who refuse to adopt Google internal silicon. It is an incredibly expensive way to buy time.

This arrangement highlights a fundamental flaw in Google custom chip strategy. Google has invested heavily in its proprietary Tensor Processing Unit (TPU) architecture. These custom chips are highly cost-efficient for running Google internal workloads and training Gemini. However, the enterprise software ecosystem overwhelmingly prefers the NVIDIA CUDA platform. External developers, corporate researchers, and enterprise clients are natively built into the NVIDIA software environment. They simply do not want to rewrite their code for Google TPUs.

Unsurprisingly, Google is forced to lease 110,000 NVIDIA GPUs at bulk rates of roughly 11.45 dollars per hour per GPU. Google must do this simply to retain its enterprise customers. They plan is to layer their own management software and service guarantees onto this rented hardware. They will then resell those NVIDIA instances to enterprise clients at steep markups ranging from 18 to over 70 dollars per hour. Google is acting as a high priced broker. Google are using Musk hardware to host clients like Apple, which utilizes these fleets for its Gemini upgraded Siri infrastructure.

This dependency introduces massive operational complications. Because xAI is a direct rival to Gemini, Google has been forced to mandate strict software-level firewalls. Google must run its own isolated environment on top of the physical SpaceX hardware to prevent data bleeding. They have to ensure that zero user data or proprietary model weights leak back into the Musk ecosystem. It is a highly complex, precarious setup born entirely of necessity. Put simply, Google used its massive cash reserves to buy immediate market share because its internal infrastructure strategy broke down.