Anthropic Hits $65 Billion in Annualized Revenue and Reopens the Bill for European Banks Treating AI as an Efficiency Project
August 17 announcement puts Anthropic on a vertical growth trajectory while Deutsche Bank, HSBC, and Santander try to convert hiring into cuts. The question is no longer when to adopt, but who pays the bill.
Reports from TechCrunch and Axios on August 17 revealed that Anthropic's annualized revenue reached approximately $65 billion in July 2026, up from $47 billion in May. About 80% of the revenue comes from API calls by companies and startups, according to data cited by investors. This curve validates, through demand, the spending of European banks on generative AI that Deutsche Bank, HSBC, and Santander have been promoting as an efficiency initiative.
The number matters to the financial C-level for two practical reasons. The first is that the actual bill is shifting from capex on training to opex on inference, and this transition changes the return calculation: spending $20 million on internal modeling makes different sense than spending $20 million a month on Claude Opus 5 calls in production. The second is that demand is surging just as frontier prices are dropping, which can only be sustained if the volume of use cases validates the traffic curve. European banks are part of this equation.
Deutsche Bank and Santander Lead the Talent Race, HSBC Trails on the Payroll
The Evident AI Index placed Deutsche Bank and Santander among the European leaders in AI talent hiring in 2026, with HSBC, Barclays, and BNP Paribas in the next group. This is not the same movement that Wall Street executed in 2024, when American banks focused their efforts on quantitative research teams. The European cycle is operational: KYC, compliance remediation, credit reviews, reconciliation automation. It is the mid-office layer that employs volume, not the glamour of the derivatives desk.
On the other side, HSBC announced plans to reduce approximately 10% of its global workforce over a two-year window, with the official narrative tied to modernizing operations and adopting AI. This cut is the most explicit move by a major European global bank within the AI justification basket. Morgan Stanley analyzed 35 European banks and projected 200,000 jobs at risk by 2030, approximately 10% of the workforce of the largest incumbents. The picture is consistent between what the market projects and what the big banks deliver.
There is a significant dissonance worthy of note. A survey by EY with 240 financial sector CEOs in 2026 found that 60% expect AI adoption to maintain or increase the total workforce in the next twelve months. This data needs to be interpreted as a signal, not a repudiation. Increasing the total workforce is compatible with cutting senior mid-office staff while hiring AI engineers and quantitative risk analysts. It is recomposition, not neutrality.
The Reading for Germany, the UK, Poland, and the Operational Line in Asia
The mechanical reading is misleading. Deutsche Bank has around 90,000 employees and operates with a heavy headquarters presence in Frankfurt but maintains technology centers in Berlin, Bonn, and Bucharest, with about 8,000 professionals in Bangalore and Pune. KYC automation does not impact the three locations the same way. Bangalore focuses on manual second-line reviews that the RPA architecture has been compressing since 2022, and Claude is starting to replace RPA itself instead of the senior analyst. Bucharest, with average costs below Frankfurt and above Bangalore, is in the middle of the pinch.
In the UK, the slack is less. HSBC operates much of its technological backbone in Kraków and Manila, and the cuts announced so far affect both locations. Santander maintains its centers in Boadilla and Porto. All these capabilities are subject to simultaneous rearrangements, with timelines that unions can stretch in Frankfurt and Madrid, but not in Kraków or Manila.
In Brazil, Itaú, Bradesco, and Santander Brasil operate their own centers, which are simultaneously clients and suppliers to their parent companies. Santander Brasil, in particular, has treated the development of internal agents as a pillar of its efficiency narrative, with consolidated figures only available in the next conference call. The question that Brazilian presidents still avoid answering on stage is the same: if Sol and Opus 5 continue to fall in price, how many mid-office analysts will still be needed three years from now?
Where the Hyperscaler Analogy Fails
The counterargument is worth mentioning. Anthropic is burning capital to sustain its curve, and it is not the same case as a profitable hyperscaler financing inference with other margins. If the price curve stabilizes or reverses in 2027, many of the accelerated cuts in 2026 will appear premature. Banks that are buying token volume without equivalent exit metrics risk discovering they have reduced human costs while importing supplier costs with less predictable margins. No one on the credit desk's risk committee wants to see this balance sheet in 2028.