When Google Also “Prints Stocks” to Build AI, Who Broke the Narrative of Highly Valued Neoclouds?
TechFlow Selected TechFlow Selected
When Google Also “Prints Stocks” to Build AI, Who Broke the Narrative of Highly Valued Neoclouds?
Google’s $80 billion is a golden箍 (tightening ring) placed on NeoCloud’s “Three Swordsmen”—CoreWeave, Nebius, and IREN.
Navigating Web3 tides with focused insightsAuthor: Ada, TechFlow
Recently, Google announced its first equity financing since 2005. When viewed collectively, Google’s three major moves over the past 90 days signal that its $80 billion capital raise targets more than just capacity constraints—it directly challenges Nvidia’s dominance over the entire AI compute market. The most immediate impact falls on the “NeoCloud Three Musketeers”—CoreWeave, Nebius, and IREN—whose valuations are predicated on Nvidia’s irreplaceability.
The Complete Picture Emerges from Three Moves
On April 22, at the Google Cloud Next ’26 conference, Google unveiled its eighth-generation Tensor Processing Unit (TPU), splitting it into two dedicated chips: the TPU 8t for training and the TPU 8i for inference. In the same product announcement, Google explicitly stated—for the first time—that it would sell TPUs to selected third-party data center operators. This marks the first official commercialization of TPUs outside Google Cloud since their mass production began in 2015—a decade-long milestone.
On May 24, Google and Blackstone announced the formation of a joint venture. Blackstone initially contributed $5 billion in equity, with leverage potentially scaling total capital to $25 billion; Blackstone serves as the majority shareholder, while Google contributes TPUs and software. The new company is positioned as a compute-as-a-service provider—the standard business model of NeoClouds—and aims to deploy 500 megawatts of capacity by 2027, led by former Google executive Benjamin Treynor Sloss. On the day of the announcement, CoreWeave’s stock fell 3.8%, and Nebius dropped 1%.
On June 1, Google announced an $80 billion equity financing round—the first time since 2005 that Google has fully deployed its equity issuance tools. The package includes $15 billion in convertible preferred shares, $15 billion in underwritten Class A and C common shares, a $40 billion at-the-market (ATM) equity offering program, and a $10 billion private placement with Berkshire Hathaway.
Together, these three moves reveal Google simultaneously pursuing three parallel strategies: building its own data centers, selling chips, and entering the NeoCloud market. Framing this merely as infrastructure expansion vastly understates Google’s ambition—it is attempting to rebuild the entire Nvidia GPU–dominated compute market around TPUs.
The Real Reason Behind the $80 Billion Equity Raise
Media coverage attributing the entire financing to AI infrastructure investment is misleading. Google’s own SEC filing clarifies that approximately $30 billion of the $40 billion ATM program is earmarked to cover tax obligations arising from employee stock-based compensation in 2026—a purely administrative arrangement, not new capital expenditure.
Excluding this amount, roughly $50 billion qualifies as “new money” for AI infrastructure: $30 billion from the underwritten offering, $10 billion from the Berkshire private placement, and $10 billion from the remaining ATM allocation.
Put in context, Google’s full-year 2026 capital expenditure guidance stands at $180–190 billion—with further “significant increases” expected in 2027. Thus, the $50 billion equity raise covers only slightly more than one-quarter of annual capex; the remainder must be funded via operating cash flow, debt, and future financings.
This explains why Google had no choice but to tap equity. Google Cloud’s Q1 2026 revenue surged 63% year-on-year, and backlog ballooned from $230 billion last quarter to over $460 billion. Customer contractual demand alone already far outpaces Google’s internal capacity build-out pace. In other words, even a cash-generating powerhouse like Google now faces AI capital intensity so extreme that equity dilution has become unavoidable.
Berkshire Hathaway’s $10 billion private placement warrants separate attention. Over the past 60 years, Buffett has almost never participated in primary-market financings—especially not in capital-intensive “new economy” deals. His decision to acquire shares at fixed prices ($351.81 per Class A share, $348.20 per Class C share) functions less as an investment and more as a formal endorsement—essentially stamping “AI compute as an infrastructure asset class” with institutional legitimacy.
Microsoft’s Path vs. Google’s Path: A Strategic Fork
To grasp the true significance of this financing, compare the two largest compute buyers.
Microsoft pursues a “build-plus-NeoCloud outsourcing” strategy. Its in-house chip Maia has fallen behind schedule, while OpenAI’s training and inference compute demands continue growing exponentially. Since late 2025, Microsoft’s committed contracts with NeoCloud providers have exceeded $60 billion: $23 billion to Nscale (for deploying 200,000 GB300 GPUs), with the remainder distributed among CoreWeave, Nebius, IREN, and Lambda Labs—all exclusively using Nvidia GPUs. Microsoft remains deeply dependent on NeoClouds because its internal capacity cannot keep up with demand, and its custom silicon cannot yet match Nvidia.
Google takes the opposite path: developing TPUs in-house, building its own data centers (without relying on NeoClouds), selling TPUs to third parties—and now competing directly in the NeoCloud space via its Blackstone JV. Google doesn’t need NeoClouds; it intends to become their competitor.
This fork represents the strategic pivot point of the entire financing. The deeper Microsoft entrenches itself in NeoClouds, the more aggressively Google must dismantle them. Their divergent choices stem from fundamentally different underlying assets: Microsoft lacks high-end AI chips of its own; Google has TPUs.
What makes Google’s approach viable is real TPU progress. Anthropic migrated large-scale training workloads to TPUs in 2025; Meta, SSI, and xAI have reportedly entered TPU procurement talks. Internally, Google claims TPUs deliver 3–5x better cost-performance than Nvidia GPUs for specific inference workloads—a figure corroborated by multiple independent analysts.
The Asymmetric Fates of the Three Musketeers
Revisiting the NeoCloud trio—CoreWeave, Nebius, and IREN:
In the short term, Google poses no threat to their cash flows. CoreWeave’s Q1 backlog reached nearly $100 billion, including a newly signed $21 billion contract with Meta in March and a multi-year deal with Anthropic. Nebius posted $390 million in Q1 revenue, up 841% year-on-year, with full-year 2026 guidance of $3–3.4 billion in revenue and an annualized operating run rate of $7–9 billion; it has already secured a $27 billion five-year contract with Meta. IREN holds a $9.7 billion contract with Microsoft plus a $5.5 billion agreement with Nvidia. All these contracts are locked-in Nvidia GPU commitments—unsubstitutable by TPUs today.
What’s being disrupted is the valuation narrative. Their lofty valuations rest on three assumptions: extreme AI compute shortage, Nvidia GPUs as the sole viable option, and hyperscalers’ inability to scale internal capacity fast enough. Google’s multi-pronged campaign systematically undermines each assumption: TPUs represent a credible alternative; new capacity is ramping; and where internal build-out lags, the JV accelerates deployment.
Yet the three companies face starkly different realities.
CoreWeave has partially de-risked its high valuation—but its debt leverage remains unresolved. Positioned as “AWS for the GPU era,” its greatest ambition commands its highest valuation premium. Nvidia already owns ~11% of CoreWeave—valued near $4.9 billion—and in January 2026 doubled down at $87.20 per share. This deep entanglement leaves CoreWeave no room to pivot to TPUs: to customers, it *is* the Nvidia GPU proxy. Google’s strategy succeeds if it convinces markets that TPUs are truly a first-tier option—immediately compressing CoreWeave’s valuation premium.
Nebius occupies the middle ground. Its tech stack is relatively open (its Soperator platform is open-sourced, mirroring CoreWeave’s SUNK approach); although its customer base leans toward Nvidia GPUs, its flexibility is higher. Nebius’s debt and cash positions nearly offset each other. Leopold Aschenbrenner’s hedge fund Situational Awareness—founded by a former OpenAI researcher—began accumulating Nebius shares at the end of May, precisely after Google’s entry became clear. His bet is on whether growth or valuation compression will dominate.
IREN is the outlier. Originally a Bitcoin miner, it’s the heaviest-asset, lowest-valuation-premium player among the three. Its $9.7 billion Microsoft and $5.5 billion Nvidia contracts provide ample cash flow to sustain fundamentals. With no “high-valuation narrative” to shatter, IREN shifts from “weakest” to “most resilient” in the new landscape—though it’s hardly cheap anymore.
From Supply Shortage to Client Segmentation in the Compute Market
The second-order implication lies in structural transformation across the compute market.
For the past 18 months, the AI compute market has been a classic seller’s market: Nvidia dictates supply cadence, and all buyers queue up. Now three simultaneous layers of segmentation are emerging.
First, frontier model labs are embracing multi-stack adoption. Anthropic publicly uses Google TPUs, AWS Trainium, and Nvidia GPUs; OpenAI is reportedly evaluating TPUs. Once multi-stack becomes standard among top labs, the “Nvidia-exclusive” NeoCloud label transforms from a strength into a client-perceived limitation.
Second, hyperscaler strategies are diverging. Microsoft (deeply tied to NeoClouds), Google (building + selling chips + becoming a NeoCloud), and Amazon (primarily self-reliant via Trainium) pursue entirely distinct paths. This divergence directly shapes NeoCloud client composition. Today, NeoClouds depend heavily on Microsoft and Meta—while Google remains completely absent. If Microsoft reduces outsourcing due to Maia improvements or shifting OpenAI dynamics, NeoClouds face structural revenue risk.
Third, funding-cost segmentation is intensifying. Google finances via equity + Berkshire’s endorsement + operating cash flow—effectively zero-cost capital. CoreWeave’s latest loan pricing stands at SOFR (Secured Overnight Financing Rate) + 4.5%. In a capital-intensive business where GPU depreciation cycles span only 5–7 years, this funding-cost gap compounds into a fatal disadvantage. NeoClouds exist today because Nvidia GPUs remain scarce; once GPUs shift from scarcity to relative abundance, the lowest-cost-capital players will reassert market dominance—precisely Google’s wager.
Three Metrics to Watch Next
Returning to the $80 billion equity raise: its true signal is that Google now treats AI compute as a market ripe for redistribution. CoreWeave, Nebius, and IREN still have two-to-three years of contracted revenue ahead—but the “Nvidia-only” thesis underpinning their high valuations has been cracked open by Google’s coordinated offensive.
Just three developments matter going forward: whether the Google-Blackstone JV delivers its promised 500 MW of capacity on schedule by 2027; whether TPU customer adoption expands beyond Anthropic to include Meta and xAI; and whether Microsoft—amid heightened tensions with OpenAI—reopens TPU discussions. If any two of these materialize, the NeoCloud Three Musketeers’ story will need rewriting.
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
深潮TechFlow
