Alphabet Hedges Both Sides of the AI Interconnect War as Cloud Backlog Reaches $514 Billion

By funding both Nvidia's NVLink Fusion and AMD's UALink simultaneously, Alphabet is extending the vendor-independence logic of its TPU program to the next critical layer of AI infrastructure—interconnect fabric—while operating at a scale that makes the bet structurally significant.

The strategic logic behind Alphabet's reported decision to simultaneously fund Nvidia's NVLink Fusion and AMD's UALink—two competing interconnect frameworks for next-generation AI clusters—reveals as much about the company's infrastructural ambitions as it does about the maturing economics of hyperscale AI. According to Tech Times, Google is hedging across both camps of what is becoming an AI chip interconnect standards war, a move designed to avoid the kind of supplier lock-in that has historically ceded pricing leverage to Nvidia. The posture is consistent with a longer pattern: Alphabet began developing its own tensor processing units years before custom silicon became fashionable across the industry, and it is now applying that same logic to the fabric that ties accelerators together at scale. Whether the funding commitments translate into architectural influence over one or both standards remains to be seen, but the intent signals a company that views the interconnect layer as the next contestable frontier.

The scale at which Alphabet is operating makes any single infrastructure decision more than a footnote. Sundar Pichai raised the company's full-year capital expenditure guidance to $205 billion in the most recent quarter—a figure corroborated across multiple earnings disclosures and analyst reports—while Google Cloud's contracted backlog reached $514 billion, the largest in the unit's history. To finance the buildout, Alphabet executed a $25 billion bond sale in early August, reportedly the largest single data-center financing raise on record, followed by an $80 billion equity raise that attracted Berkshire Hathaway as a marquee backer. Greg Abel, who is set to succeed Warren Buffett as Berkshire's chief executive, this week publicly described Google as a 'significant player' in AI—a phrase that carries weight given that Berkshire's $10 billion commitment was made after Buffett's retirement announcement and reflects the firm's own long-duration capital calculus, not a momentum trade.

Google's relationship with custom silicon is simultaneously entering a new commercial phase. In Q2 2026, for the first time, Alphabet recorded external revenue from its TPU chips; by late July, Google Cloud had begun shipping TPU systems directly to enterprise customers, effectively becoming a chipmaker in the commercial sense rather than merely an internal consumer. This transition matters because it converts a historically captive cost center into a revenue line and a competitive differentiator—and because it adds a second axis of pressure on Nvidia beyond Alphabet's chip procurement decisions. Alphabet has also backed 2.4 gigawatts of GPU and high-performance compute capacity across ten leasing projects, including facilities from crypto miners pivoting to AI infrastructure, a pragmatic acknowledgment that power and physical capacity now define the binding constraint in hyperscale AI buildout more than chip supply alone. Hyperscalers confirmed as much in their Q2 earnings calls: power, networking, and deployment speed have displaced GPU scarcity as the decisive bottleneck.

Not every signal from Alphabet's recent period is unambiguously constructive. Delays in the Gemini AI model family attracted investor scrutiny ahead of the Q2 print—Alphabet shares slid roughly 7% on the earnings day despite a revenue beat, with the market's reaction centered on the capex trajectory rather than the cloud performance numbers. Leadership continuity at Google DeepMind is also a variable: Demis Hassabis has stepped up to Alphabet Chief Scientist, and Jeff Dean, who spent 27 years at Google, departed in August to found an AI science startup called Discovery Loop. Whether these transitions affect research velocity at the group's foundational AI lab is, at this stage, genuinely uncertain. Separately, PPC Land reported that as much as 48% of Google Cloud's 2027 revenue could depend on compute consumed by OpenAI and Anthropic—a figure that has not been independently corroborated—but the concentration risk it implies is structurally real: both companies have publicly signaled ambitions to build and own their own AI infrastructure over time, which would erode that demand organically.

The competitive picture remains fluid. Morgan Stanley projected in August that global cloud capital expenditure will reach $1.2 trillion by 2027, a 30% upward revision from prior estimates, with Alphabet among the firms driving the acceleration. One analysis from 24/7 Wall St. characterized Alphabet as capturing the 'capex alpha' of 2026—benefiting from its early TPU silicon buildout—while projecting Amazon as the accelerating force in 2027 through its Trainium and Graviton programs. Microsoft, by most accounts, has maintained its spending pace without the sharp upward revisions Alphabet has made. The decision to hedge across Nvidia and AMD's interconnect roadmaps, rather than committing to either, fits this competitive posture: Alphabet is a company that has found it worth spending heavily to preserve optionality, because in a capital-intensive technology cycle, owning the stack—or at minimum avoiding dependency on any single vendor's definition of it—tends to be the only durable form of leverage.

Three signals deserve particular attention in the quarters ahead. First, whether TPU external revenue grows into a meaningful share of Google Cloud's income statement or remains a rounding error against the $25 billion quarterly run rate—the answer will indicate whether Google's chipmaker ambitions are commercially viable at scale or primarily strategic. Second, the Gemini model release cadence: sustained delays would compound competitive pressure from OpenAI and Anthropic at the precise moment Alphabet is also their largest compute supplier, a tension that has no clean resolution. Third, the specific terms and uptake of the interconnect funding commitments to NVLink Fusion and UALink—if Alphabet secures genuine architectural influence over either standard, the return on that investment extends well beyond avoiding a vendor surcharge and into the structural economics of every AI cluster the industry builds for the next decade.