Alphabet’s accelerating investment in AI infrastructure is placing new pressure on the decentralized computing sector to prove it can serve customers that large cloud providers cannot easily reach. Google’s growing TPU supply, data-center buildout and AI product distribution give it more control over the hardware, power and software layers needed for advanced models, narrowing the room for token-based networks built largely around speculative demand.
The company has not announced Gemini 4.0, despite unverified screenshots and blind-test claims circulating online. Google said during its July earnings call that it had begun its “most ambitious pre-training” effort to date, but it did not attach that work to a Gemini 4.0 release schedule or identify the model involved.
Google’s latest disclosed benchmark results instead concern Gemini 3.8 Flash, introduced in early September. In results published by Google, the model scored 73.7% on DeepSWE v1.1, a software-engineering benchmark. That was close to the 74.0% score reported for Anthropic’s Claude Opus 5 on the same test. The narrow gap illustrates why infrastructure capacity has become as strategically important as model releases: leading developers are competing to train, deploy and serve increasingly capable systems at scale.
Google expands distribution beyond the Gemini app
Google reported that the Gemini app exceeded 1 billion monthly active users in August. The company also reported more than 100 million monthly active users on iOS, while Gemini’s Android system-level functions were operating across more than 40 commonly used apps.
Those figures give Google a consumer distribution channel that many AI developers lack. A model does not need to win every benchmark to become commercially consequential if it is embedded in search, mobile operating systems and productivity software used by hundreds of millions of people.
Google has also expanded AI features in its advertising and search products. Alphabet reported that Google Search & Other revenue rose 17% year over year in the second quarter, while YouTube advertising revenue increased 13%. The company has expanded AI Overviews and continued rolling out AI Mode, where it has begun testing native advertising formats.
The commercial value of those tools will depend on whether Google can preserve search quality while finding ways to monetize AI-generated answers. Native ad testing in AI Mode points toward a familiar Google strategy: use a high-volume consumer product to support the capital costs of the underlying computing infrastructure.
Apple’s updated Siri architecture also lists Apple Foundation Models developed with Google based on Gemini, according to the supplied description. If that integration expands, Gemini would gain another route into consumer devices without Google needing to own the surrounding hardware ecosystem.
Cloud contracts support large infrastructure commitments
Alphabet said Google Cloud revenue rose 82% to $24.8 billion in the second quarter. Its remaining performance obligations, a measure of contracted revenue yet to be recognized, exceeded $510 billion, with more than half expected to be recognized during the following two years.
That backlog provides a more concrete basis for infrastructure spending than the market excitement surrounding new model names. Enterprise demand for cloud capacity, data tools and AI services can support multi-year contracts, allowing Google to plan data-center and chip procurement on a longer horizon.
Google is also positioning its TPU, or Tensor Processing Unit, infrastructure beyond internal model development. The supplied account describes TPUs increasingly being offered as an external compute product, bringing Google into more direct competition with providers that rent Nvidia-based graphics processing units and with decentralized networks seeking to aggregate independently owned hardware.
Anthropic’s reported TPU commitments underline the scale of demand. Prior disclosures cited in the supplied material say Anthropic secured access to as many as 1 million TPUs, involving more than 1 gigawatt of computing capacity and spending described in the hundreds of billions of dollars. Broadcom has said that Anthropic is expected to obtain roughly 3.5 gigawatts of next-generation TPU AI compute through Broadcom beginning in 2027.
Broadcom’s relationship with Google is also described as extending to 2031. Marvell has entered Google’s in-house AI chip efforts involving inference chips, interconnects and networking, according to the supplied account. These arrangements show that AI infrastructure is no longer only a contest over individual accelerators. It increasingly depends on networking, chip design, electricity contracts and the ability to operate large clusters reliably.
Power and data centers become limiting factors
Google is building a data center in Finland that it describes as Europe’s largest, backed by a €13 billion investment, according to the supplied material. The project is paired with efforts to secure long-term power resources, reflecting an industry constraint that is becoming as consequential as chip availability.
Large AI clusters require extensive electricity, cooling and grid connections. A company that can reserve power and complete data-center construction gains an advantage that cannot be replicated quickly by purchasing additional processors. Delays in permits, transmission upgrades or generation capacity can leave expensive hardware underused.
Google Home’s decision to open its MCP Server to third-party agents adds another dimension to the infrastructure strategy. Model Context Protocol, or MCP, enables AI agents to connect with external tools and systems. In Google Home’s case, that can allow third-party agents to interact with devices including cameras, lights and temperature controls. The value lies less in the protocol itself than in the possibility of turning AI assistants into systems that can take actions across physical environments.
Waymo offers a separate potential demand channel for Alphabet’s AI stack. The supplied account says the autonomous-driving business could expand beyond several US cities into markets including Tokyo and Singapore, though Alphabet has not provided financial projections for those markets.
A harder market for decentralized hardware tokens
For decentralized physical infrastructure projects, the expansion of corporate AI capacity creates a difficult dividing line. Networks that rent GPUs, coordinate storage or reward node operators may benefit when enterprise demand exceeds centralized cloud supply in particular regions or for particular workloads. They face a tougher challenge when hyperscalers can offer integrated hardware, networking, enterprise support and multi-year contracts.
The supplied material says the combined market value of decentralized hardware networks fell sharply earlier in the year before stabilizing near $7 billion in late summer. It also puts network revenue across distributed hardware systems at $72 million last year. The contrast between those revenue figures and earlier token valuations leaves little room for projects that cannot demonstrate recurring usage.
Corporate demand could reduce the availability of new processors for smaller node operators, especially where major providers secure supply years ahead. Yet scarcity alone does not guarantee demand for decentralized alternatives. A network must show that it can deliver dependable uptime, transparent pricing, suitable hardware and sufficient geographic coverage for a paying customer.
The next earnings reports from Alphabet, chip suppliers and cloud providers will therefore matter beyond traditional technology equities. They will help indicate whether AI spending remains concentrated in massive centralized campuses or whether capacity bottlenecks create durable openings for decentralized compute and storage networks with measurable revenue.
Explore how AI reshapes finance—learn crypto basics and market structure in our guide, start your crypto journey today.
Disclaimer: The content on this page is provided for general informational purposes only and does not represent the views or financial advice of Toobit. We make no guarantees regarding the accuracy or completeness of this information and shall not be held liable for any errors, omissions, or outcomes resulting from its use. Investing in digital assets involves risk; users should independently evaluate their financial situation and the risks involved. For further details, please consult our Terms of Service and Risk Disclosure.
