BlockBeats News, July 8th. Amazon once again raised debt for AI infrastructure, Meta rebounded in July due to a potential cloud computing rental plan, while Wall Street was both chasing AI concept stocks and reassessing the ROI of this infrastructure construction frenzy.

After more than two years of a market driven by Nvidia chips, cloud capital expenditure, and large model financing, investors' questions are becoming more specific: will the hundreds of billions of dollars in data centers, GPUs, and power input eventually translate into profit, or will it only become depreciation and cash flow pressure?

This week, Bank of America Securities and SemiAnalysis provided two interrelated clues. The former raised the AI capital expenditure of Amazon, Alphabet, and Meta to new highs, attempting to measure the future revenue potential of cloud giants using GW data center capacity; while the latter depicted Anthropic as one of the few frontier model companies that has already approached "profit validation".

In an internet industry report released on July 7th, Bank of America Securities raised the capital expenditure expectations of large cloud companies, believing that Amazon, Alphabet, and Meta will continue to prioritize data center capacity over short-term free cash flow. A day later, a financial analysis report on Anthropic by SemiAnalysis provided another aspect of the answer: if model companies can transform computing power into high-margin software revenue, AI infrastructure investment may be entering a commercially viable stage.

Bank of America expects that the data center capacity of the three internet giants will increase from about 27GW at the end of 2025 to 39GW in 2026, reaching 57GW in 2027. Amazon is considered the fastest-expanding company, with an estimated additional capacity of about 15GW from 2026 to 2027, Google around 9GW, and Meta around 6GW.

This means that AI competition is no longer just about model parameters or application deployment, but about competition in power, chips, servers, memory, and data center construction capabilities. BofA estimates that AWS's 2026 capital expenditure will be $159 billion, Alphabet's $195 billion, and Meta's $145 billion; by 2027, these figures will rise to $230 billion, $290 billion, and $185 billion, respectively.

Investors' concerns are also direct: will these expenditures turn into long-term depreciation pressure and declining profit margins, rather than future revenue.

BofA's response is somewhat optimistic. The bank believes that each GW of data center capacity is becoming a valuable asset in itself. Its calculations show that AWS's annual revenue per GW of cloud capacity is around $10.6 billion in 2026, while Google Cloud is around $15.7 billion. By 2030, AWS's cloud revenue could reach $409 billion, and Google Cloud could reach $387 billion. If Meta can externalize some AI capacity, the potential revenue opportunity could also reach $110 billion.

SemiAnalysis's analysis of Anthropic takes this logic to the level of a modeling company. Screenshots show that the report states Anthropic's profit in the third quarter of 2026 will exceed $1 billion and describes it as the "Anthropic IPO Financial Peak." The report states that Anthropic secretly filed for an IPO on June 1 and believes it could become the first cutting-edge AI lab to enter the public market with a clear path to profitability.

The core conclusion of this report is that Anthropic has a stronger monetization capability in the B2B market, Claude Code, and enterprise software development scenarios. Unlike AI products that rely on consumer traffic or high subsidies, enterprise development tools, once integrated into workflows, are more likely to form high-frequency usage, paid subscriptions, and higher gross margins.

Looking at both reports together, the narrative of the AI industry is changing. Cloud providers are building underlying capacity, while modeling companies are proving that this capacity can be repackaged into software, APIs, enterprise agents, and code tool revenue. The former determines the supply, and the latter determines the return on investment.

However, risks still loom large. BofA points out that power supply may be a key bottleneck for data center capacity expansion, and high capital expenditure will continue to suppress free cash flow. For model companies like Anthropic, the sustainability of profitability depends on inference costs, customer retention, price discipline, and whether competitors launch price wars.

But the market is seeking a replicable formula: if each GW of computing power can consistently generate billions of dollars in revenue, if model applications can cover computing and infrastructure costs and generate operating profits, then the AI infrastructure cycle is no longer just a capital expenditure cycle but a new round of cloud revenue cycle.

Anthropic thus becomes a crucial test case. It does not have the global cloud infrastructure of AWS or Google, yet it may convert external compute power into high-value revenue through models, code tools, and enterprise customers. For investors, this is the dividing line between AI bubble theory and platform revaluation theory.