Artificial intelligence infrastructure rather than telecoms or venture capital now defines SoftBank’s market identity
SoftBank’s trading screens in Tokyo lit up after reports that SB Energy and OpenAI were moving toward initial public offerings. Within days, SoftBank shares surged nearly 30% in two straight trading sessions, then jumped almost 13% intraday to an all-time peak of 8,000 yen. The rally carried a strange arithmetic beneath it: investors were no longer valuing the Japanese conglomerate primarily as a telecom operator or venture investor, but as a liquid wrapper around artificial intelligence infrastructure, semiconductor architecture, and private-market equity scarcity. By then, SoftBank’s OpenAI stake had ballooned to roughly US$80 billion in value, generating around US$45 billion in unrealized profits, while its roughly 90% ownership of Arm gave it control over one of the foundational layers of global AI computing infrastructure.
That re-rating did not happen in isolation. The United States has already recorded more than 424 IPOs year to date, while IPO activity accelerated through early 2026 after several years of subdued issuance. Banks that spent two years waiting for deal flow suddenly found themselves operating in the richest underwriting environment since the post-pandemic boom. Goldman Sachs called its 2025 investment-banking results its second highest ever. Citigroup posted its best-ever M&A quarter. Morgan Stanley reported record total revenues. JPMorgan broke all-time revenue records for a US bank. The firms collecting fees from the AI boom were becoming among its largest balance-sheet beneficiaries.
SoftBank’s decision to hire JPMorgan, Goldman Sachs, Morgan Stanley, Citi and Mizuho to take SB Energy public in a US IPO, while simultaneously preparing a US listing for robotics spinoff Roze, concentrated that power even further inside Wall Street’s underwriting cartel. Lead managers remain entitled to the full underwriting spread, while syndicate members split the concessions according to distribution strength. The mechanics matter because the scale of capital now moving into AI has begun to overwhelm the traditional distinction between venture finance and public markets. Companies once expected to remain private for years are racing toward liquidity events precisely because private valuations have detached from conventional operating metrics.
Scarcity premiums begin to erode once every AI company seeks liquidity at the same time
The numbers underpinning those valuations have become extreme even by technology-bubble standards. The median revenue multiple for AI companies stood at 24.2x, driven largely by larger capital raises rather than outright acquisitions. Founders chasing those prices confront an uncomfortable asymmetry: valuation multiples in fundraising rounds are far higher than in 100% sales, and sale offers typically come in significantly lower. That gap has turned pricing itself into a form of leverage. Revenue multiples now translate directly into negotiating leverage and dilution, especially for firms attempting to secure an “AI premium” before public markets reset expectations.
Wall Street has begun warning that the machine financing the boom is approaching saturation. Goldman Sachs warned that a larger wave of IPO equity supply could temporarily overwhelm Wall Street liquidity as crowding in AI-related trading themes intensifies. Tony Pasquariello, Goldman’s global head of hedge fund coverage, noted that US IPO activity has fully rebounded while AI-related technology companies continue raising capital at scale, forcing investors to ask whether US equity markets can absorb the surge of new listings. The risk is not an outright collapse in the bull market; the more immediate effect is that valuation expansion becomes harder as capital reallocates across positions. When every AI-linked company comes public at once, scarcity disappears first.
That pressure lands hardest on software companies that dominated the last IPO cycle. The IPO market has changed considerably since 2020 and 2021, when most listings came from the software sector. By 2025, software companies faced a much higher bar to going public because investors increasingly viewed them as structurally vulnerable in an AI-driven economy. Yet the irony inside the spending data is striking: enterprises do not appear to be reallocating IT budgets away from traditional software toward AI. Budgets are expanding instead, which means incumbents must finance AI integration while defending slowing legacy growth. Investors once rewarded software firms for recurring revenue and scale; now they reward compute access, proprietary models, and infrastructure control.
Capital markets increasingly reward ownership of power, chips and industrial capacity
That shift has redirected capital toward the physical layer of artificial intelligence. Analysts increasingly anchor baseline AI infrastructure models to NVIDIA’s forward data-center revenue estimates in order to infer future requirements for power, data centers, and supporting systems. With hyperscaler capex surging 36%, the bottlenecks no longer sit inside applications but inside semiconductors, electricity, and advanced manufacturing capacity. TSMC manufactures chips for Nvidia, AMD, Broadcom, Qualcomm and Apple, controlling roughly 68% of the global foundry market by revenue. It does not matter which chip architecture wins — TSMC makes them all.
The scramble for electricity has become equally consequential. Microsoft’s electricity demand for AI data centers is projected to surge more than 600% by 2030. Google spent US$4.75 billion acquiring power company Intersect Power. Meta signed a massive power purchase agreement tied to the Comanche Peak nuclear facility. Investors who initially treated AI as a software revolution now confront an industrial buildout closer to a utility expansion cycle, one requiring transmission capacity, generation assets, cooling systems, and financing structures capable of sustaining years of fixed investment.
That industrialization explains why sovereign wealth funds have accelerated into the sector. Global sovereign wealth funds amassed a record US$15 trillion in assets under management while state-owned investors deployed US$66 billion into AI and digitalisation investments during 2025. Mubadala alone invested US$12.9 billion in AI and digitalisation, helping the Gulf funds account for 43% of all capital invested by state-owned investors globally. The beneficiaries are not only technology firms but the banks orchestrating cross-border financing, listings, and secondary offerings. In a market where IPO supply is accelerating and infrastructure spending compounds across sectors, underwriting capacity itself becomes strategic infrastructure.
American finance consolidates power as AI compresses labour and liquidity into fewer institutions
The imbalance between American and European finance has widened accordingly. The top five US banks now capture roughly 55–60% of global investment-banking fees, while all European banks combined account for only 15–20%. US equity-market culture, deeper capital pools and a larger M&A market continue widening that structural advantage. Europe still retains specialized strengths — Société Générale remains known for derivatives and structured products and has invested heavily in renewable-energy project financing — but the gravitational center of AI finance now sits inside the American equity machine. Even institutions once synonymous with European banking power look diminished beside it. Credit Suisse was absorbed by UBS in 2023, and its market capitalization shrank to roughly US$3.56 billion.
Inside the banks themselves, artificial intelligence has already begun changing labor economics. JPMorgan’s LLM Suite can generate a presentation deck in roughly 30 seconds that previously took junior analysts hours to complete. Derek Waldron demonstrated the platform by generating a five-page presentation for a meeting with a major tech company almost instantly. Productivity gains that once justified armies of junior bankers increasingly accrue to the institutions controlling proprietary data, distribution relationships, and computing infrastructure. The fee pools remain concentrated at the top while the labor pyramid underneath them compresses.
The same concentration now defines market psychology. J.P. Morgan Private Bank strategist Kriti Gupta observed that when investors became excited about AI, worried about inflation, searching for growth, sustainability, or capex exposure, they bought tech anyway. Capital that once rotated between sectors has instead collapsed into a narrow set of AI-linked equities, ETFs, and infrastructure names. Funds tied to innovation, cloud computing, software and IPO exposure increasingly move together, reinforcing the correlations that dominate passive flows. The beneficiaries gain scale faster precisely because capital markets now treat AI exposure itself as an asset class.
SoftBank’s resurgence leaves it more exposed to the assumptions supporting the entire AI capital stack
That dynamic creates an unexpected vulnerability for the firms currently winning the underwriting race. Goldman Sachs warned that IPO supply itself can overwhelm liquidity, yet the banks underwriting the boom depend on the continuation of elevated valuations to sustain fee generation and secondary issuance. If public markets begin pricing AI businesses less on scarcity and more on operating discipline, the institutions that profited most from valuation expansion absorb the sharpest compression in deal economics. A financing regime built on ever-larger offerings becomes exposed when too many issuers arrive simultaneously seeking the same capital pool.