Alphabet stock is entering a new phase of the artificial-intelligence trade after a Google DeepMind executive linked the technology industry’s record infrastructure spending to the pursuit of recursive self-improvement, or RSI. The comments, reported on August 3, put a dramatic new frame around Alphabet’s plan to spend between $195 billion and $205 billion in 2026: Google may be building not only for today’s cloud demand, but also for AI systems that can help develop increasingly capable successors.
For shareholders, the stakes are enormous. Success could give Alphabet a compounding advantage across AI models, custom chips, cloud computing and consumer products, while failure could leave the company carrying massive depreciation costs and weaker free cash flow from one of the most expensive investment cycles in corporate history.
Google DeepMind Connects AI Spending to RSI
Jasjeet Sekhon, Google DeepMind’s chief strategy officer, described recursive self-improvement as a key part of the investment thesis supporting the unprecedented capital expenditure now flowing into AI infrastructure. His remarks were made at an agentic-AI event at the University of California, Berkeley, according to The Information and Seeking Alpha.
RSI generally refers to an AI system playing an active role in improving the software, training methods, evaluation tools or research processes used to build future AI systems. In the most ambitious version of the idea, an improved model becomes better at helping create its next generation, potentially accelerating development beyond the normal pace of human-led research.
That does not mean fully autonomous self-improving AI has arrived. A recent review of more than 1,000 research papers found that bounded forms of self-refinement are already being used, but open-ended RSI remains constrained by computing requirements, unreliable self-evaluation and the continuing need for human direction.
The distinction matters for Alphabet stock. Investors are not being shown a proven new revenue stream; they are being given a possible explanation for why Alphabet and its rivals believe access to extraordinary amounts of computing power could become strategically decisive.
Alphabet AI Capex Has Surged Far Beyond Expectations
Alphabet began 2026 by forecasting capital expenditure of $175 billion to $185 billion, almost double the $91.45 billion it spent in 2025. Wall Street had expected roughly $115 billion, meaning the initial guidance was already far above analysts’ models.
The company subsequently raised the range to between $180 billion and $190 billion as demand for AI computing continued to exceed available capacity. After its second-quarter results, Alphabet lifted the forecast again to $195 billion to $205 billion.
At the top of that range, Alphabet would be investing more than $560 million per day on an annualized basis. Management has indicated that approximately 60% of the spending will go toward servers, with the remaining 40% directed to data centers and networking equipment.
That escalation is now central to the Alphabet stock debate. The company is no longer making a conventional technology upgrade; it is committing a level of capital more commonly associated with national infrastructure programs.
Sekhon’s RSI comments suggest Alphabet believes this compute capacity could eventually do more than serve current customers. It may also support increasingly automated research loops capable of improving models, software tools and perhaps even the hardware used to train them.
Why RSI Could Transform the Alphabet Stock Thesis
Alphabet already earns revenue from AI through Google Cloud, advertising tools, Gemini subscriptions and workplace products. A meaningful advance in recursive self-improvement could add a powerful second layer to that business model.
The same servers used to deliver AI services to customers could help Alphabet develop more capable models, improve training efficiency and automate parts of its own research process. Better models could attract more cloud customers and users, generating more revenue that could then fund the next infrastructure cycle.
That creates the possibility of a compounding economic advantage. Alphabet would not simply be selling access to computing power; its infrastructure could help improve the technology that makes the infrastructure more valuable.
Google also controls much of its own AI stack. It develops Gemini models, operates global data centers, owns major consumer-distribution platforms and designs tensor processing units, or TPUs, for AI workloads.
Control across those layers could allow improvements in one area to benefit several others. A more efficient model might lower computing costs, a stronger TPU could improve margins, and a better AI assistant could increase engagement across Search, YouTube, Android and Workspace.
This is the most bullish interpretation of Alphabet’s spending. The company may be attempting to secure enough computing capacity to create a research and commercialization loop that smaller competitors cannot afford to replicate.
Google Cloud Growth Supports the Bull Case
The AI-capex strategy is not based solely on a distant scientific possibility. Alphabet’s latest financial results show that demand for its existing cloud infrastructure is already accelerating.
Google Cloud revenue rose 82% to approximately $24.8 billion in the second quarter, according to Reuters. Alphabet’s total revenue reached about $119.8 billion, exceeding the roughly $116.9 billion expected by analysts.
That growth suggests Alphabet is gaining from the wave of companies training models, deploying AI agents and moving more workloads onto specialized infrastructure. It may also indicate that Google Cloud is taking market share from larger competitors in certain categories.
Management has said demand remains above available capacity, forcing Alphabet to accelerate infrastructure deliveries and rent additional data-center space from outside providers. Those arrangements can help meet customer demand, although they may produce lower margins than infrastructure owned directly by Alphabet.
Google’s custom TPUs offer another potential advantage. By designing its own accelerators, Alphabet may reduce its dependence on outside suppliers, optimize performance for Gemini and potentially turn its silicon capabilities into an additional source of external revenue.
The strength of Google Cloud gives investors evidence that the spending is already supporting growth. The unresolved question is whether revenue can expand quickly enough to offset the vast cost of building and operating the required infrastructure.
Free Cash Flow Is the Warning Sign for Alphabet Investors
Alphabet’s second-quarter results also revealed the clearest danger. Despite its revenue growth, the company reported negative free cash flow of approximately $5.9 billion as capital spending accelerated.
That was a sharp reminder that revenue growth and shareholder returns are not the same thing. Data centers, processors and networking equipment require substantial upfront cash, while the related depreciation expense continues to affect earnings for years.
The risk is that Alphabet’s current earnings make the stock appear less expensive than it will look once the full depreciation burden of the 2026 and 2027 investments reaches the income statement.
Investors must also consider what happens if AI demand slows. Infrastructure built for extreme growth could become less profitable if pricing falls, customers develop more efficient models or competitors add too much capacity at the same time.
The broader technology industry is confronting the same pressure. The four largest U.S. hyperscalers are expected to spend hundreds of billions of dollars on AI infrastructure in 2026, with their combined investment since the beginning of the generative-AI boom already exceeding $1 trillion.
Heavy spending does not automatically signal a bubble. Demand can remain strong for years, and capacity shortages may justify aggressive construction. But investors are increasingly demanding evidence that each additional dollar of capex will generate an acceptable return.
RSI Remains a Thesis, Not a Proven Business Model
Sekhon’s remarks are important because they reveal how Google DeepMind may be thinking about the next stage of AI development. They should not be interpreted as confirmation that Alphabet has achieved open-ended recursive self-improvement.
Existing AI systems can write code, create synthetic training data, evaluate outputs and assist researchers. Those capabilities can improve productivity and reduce the amount of human work required in parts of the development process.
However, fully closing the research loop remains difficult. An AI system must be able to propose useful improvements, test them reliably, recognize when results are misleading and avoid reinforcing its own errors.
The research literature identifies self-evaluation as a major weakness. Formal verification works well in areas where answers can be objectively checked, but broader research decisions often require judgment, context and long-term planning that current systems do not consistently demonstrate.
Compute itself is another limitation. Even an AI system capable of discovering a better training approach may still require vast amounts of energy, data and advanced hardware to implement it.
Alphabet therefore faces a timing problem. It is spending at extraordinary levels now, while the most transformative returns from RSI may remain uncertain or years away.
Alphabet Stock Could Benefit Even Without Full RSI
Alphabet does not necessarily need to create a runaway self-improvement cycle for the investment to pay off. More limited forms of AI-assisted development could still generate meaningful financial returns.
AI coding agents could help engineers release products faster. Automated testing could reduce development costs, while AI-assisted chip design could improve the efficiency of future TPUs.
Models could also help optimize data-center power usage, identify hardware failures and allocate computing capacity more effectively. Even modest efficiency gains can become financially significant when applied across a $200 billion investment program.
The consumer businesses offer further monetization opportunities. Alphabet can deploy improved AI in Search, advertising, YouTube recommendations, Workspace and Android without needing to acquire a new audience from scratch.
This distribution advantage differentiates Alphabet from AI laboratories that must depend on subscriptions or outside partnerships. Google can use AI to defend existing revenue, improve conversion rates and create new premium features across products already used at enormous scale.
For Alphabet stock, the most realistic bull case may therefore be gradual rather than explosive. The company does not need to prove science-fiction-style RSI immediately; it needs to show that AI is improving growth, margins and product development faster than infrastructure costs are rising.
Wall Street Will Demand Clearer Returns
Alphabet’s second-quarter reaction demonstrated that investors remain sensitive to capital intensity even when revenue exceeds expectations. The shares fell after the report as higher spending and negative free cash flow overshadowed Google Cloud’s exceptional growth.
Future earnings calls will likely focus less on whether AI demand exists and more on whether Alphabet can convert that demand into durable cash generation.
Investors will watch cloud margins, backlog growth, TPU sales, Gemini adoption and the percentage of capex devoted to capacity that is already supported by customer commitments.
They will also want evidence that AI is strengthening Alphabet’s advertising and productivity businesses. Revenue tied directly to Gemini, AI agents or specialized infrastructure would make the spending easier to value.
The company may face pressure to disclose more about the economics of its AI products. Without clearer numbers, shareholders must estimate how much of the cloud growth is producing attractive margins and how much is being purchased through exceptionally heavy investment.
Outlook: What Alphabet Stock Investors Should Watch Next
Alphabet’s RSI thesis has given investors a new way to interpret its $195 billion to $205 billion spending plan. The company appears to be preparing for a future in which AI systems help improve the software, research processes and infrastructure behind the next generation of AI.
The immediate financial test will be more conventional. Alphabet must sustain rapid Google Cloud growth, restore positive free cash flow and demonstrate that Gemini and its AI infrastructure are producing measurable returns.
Investors should watch for changes to 2027 capex guidance, evidence of improved cloud capacity, new TPU revenue and specific examples of AI accelerating Alphabet’s internal research or product-development cycle.
The opportunity is historic, but so is the spending. Alphabet may be constructing the engine of a self-improving AI economy—or building ahead of a breakthrough that has not yet arrived.
