Pulse (174e6669-1b82-4c41-a4d8-95cc1c6c7141)

Capital Formation Monitor: August 5, 2026

What changed this week

Capital formation remained exceptionally strong, but the composition changed in three important ways:

1. Venture funding broadened beyond frontier-model companies, with large rounds in space, defence, energy, robotics and semiconductors. The market is still highly concentrated, but no longer concentrated exclusively in model developers.

2. AI infrastructure financing became more visibly balance-sheet intensive. Hyperscalers are using long-term leases, project-level debt, private-credit vehicles and guarantees to keep infrastructure obligations away from their conventional corporate debt metrics.

3. Governments are moving from grants toward strategic equity ownership, particularly in semiconductors, memory, advanced packaging and low-power computing.

The buildout is not suffering from a shortage of capital. The emerging constraint is whether the resulting assets can generate enough revenue to service their financing commitments.

1. Venture deployment broadened, but mega-rounds still dominate

Global venture funding reached approximately $65 billion in July, up roughly 100% from July 2025 and 10% from June. Fourteen companies raised rounds of at least $1 billion, the highest monthly count recorded by Crunchbase. AI companies received approximately $35 billion, or 53% of the monthly total.

The largest round was Blue Origin’s reported $10 billion external financing. Safe Superintelligence raised $5 billion, Moonshot AI raised $3.5 billion following the release of its Kimi K3 model, and Kling AI raised $2.8 billion. Germany’s Helsing and Quantum Systems also completed billion-dollar defence and autonomous-systems rounds.

This represents modest diversification from the first half of 2026, when a few frontier-model companies absorbed an extraordinary proportion of global venture dollars. July’s billion-dollar rounds also included energy, industrial robotics, semiconductor and security businesses. However, the underlying regime remains late-stage and growth-heavy. A record number of billion-dollar rounds is not evidence of a broad startup recovery. It is evidence that capital remains readily available for a narrow group of companies perceived to control strategic technology.

Fundraising also showed signs of recovery among established firms. Index Ventures raised $2 billion across a $400 million seed fund, a $900 million venture fund and an additional $700 million for its growth strategy, bringing its available capital to approximately $3.5 billion.

The Financial Times separately reported that Chinese venture firms are seeking approximately $35 billion across at least 60 new U.S.-dollar funds, following several years of weak foreign fundraising. Interest is reportedly stronger among European and Middle Eastern institutions than among U.S. investors, while government-backed renminbi funds continue to play a larger role domestically.

My read: venture fundraising is reopening, but LPs are favouring established managers, national champions and sectors with strategic relevance. Emerging managers without differentiated access to semiconductors, defence, scientific computing or late-stage AI will remain under pressure.

2. Early-stage capital is present, but engineering proof is becoming mandatory

The new Index funds provide meaningful fresh seed and Series A capacity, but actual deployment continues to favour companies solving specific physical or computational constraints rather than generic application-layer startups.

The most investable early-stage themes currently involve:

chip design automation;

interconnects and advanced packaging;

memory technologies;

low-power computing;

robotics and autonomy;

materials and biological discovery;

energy systems supporting compute.

This is a healthier allocation pattern than funding hundreds of lightly differentiated AI applications. It also means early-stage companies increasingly need working prototypes, credible performance benchmarks, proprietary data or a direct strategic partner before institutional capital becomes available.

3. Crossover capital is returning selectively through biotech and pre-IPO science

Braveheart Bio launched an IPO seeking approximately $319 million at a valuation of up to $1.2 billion. Fidelity indicated interest in purchasing as much as $75 million of the offering. Braveheart is developing BHB-1893, a cardiac myosin inhibitor for obstructive and non-obstructive hypertrophic cardiomyopathy.

Latigo Biotherapeutics is seeking approximately $288 million at a valuation of up to $1.08 billion for non-opioid pain therapies, including an oral drug targeting acute and postoperative pain. BlossomHill Therapeutics is seeking roughly $133 million to fund small-molecule oncology and autoimmune programmes.

These deals show that crossover investors are returning, but primarily where there is differentiated biology, clinical progress or a credible mechanism of action. Public investors are not broadly financing speculative discovery platforms.

Why it matters: biotech can provide an alternative destination for institutional capital as investors begin asking whether AI spending will produce returns outside data-center construction. The strongest candidates are companies where AI or computational methods improve discovery, but where the investment case still rests on pharmacology and clinical evidence.

4. Private credit is simultaneously raising record capital and experiencing liquidity stress

Ares raised approximately $36 billion in the second quarter, including $23.7 billion for credit strategies. Assets under management reached $671.3 billion, with roughly $170 billion of available capital.

At the same time, Fitch estimated the U.S. private-credit default rate reached a record 6.0% during the 12 months through June. Defaults were particularly elevated in industrial and manufacturing borrowers at 10.4% and healthcare borrowers at 9.4%. Industry inflows were down approximately 25% year to date.

Semi-liquid funds continue to face large redemption requests. During the second quarter:

Blue Owl Technology Income received requests equal to 38.1% of NAV.

Blue Owl Credit Income received requests equal to 18.9%.

Apollo Debt Solutions received requests equal to 16.8%.

Many of these vehicles typically repurchase only around 5% of shares per quarter.

Private-credit secondary transactions reached approximately $20.4 billion in the first half, up 122% from a year earlier and already exceeding the total recorded in all of 2025. GP-led transactions accounted for approximately 83% of activity.

Why it matters: private credit has plenty of committed capital, but liquidity and credit quality are diverging. Strong infrastructure projects with hyperscaler guarantees can still obtain financing. Traditional software, healthcare and leveraged-buyout borrowers are facing much tighter underwriting.

5. AI financing is moving into long-term leases and project guarantees

Microsoft, Meta, Oracle, Amazon and Alphabet have approximately $1.09 trillion of future payments associated with leases that have been signed but have not yet commenced, predominantly for data-center capacity. This compares with approximately $285 billion of recognised lease liabilities. Including additional Meta agreements disclosed in July, the known pipeline may be closer to $1.16 trillion.

Reported uncommenced commitments include roughly:

Microsoft: $329 billion

Meta: $279 billion, before additional July agreements

Oracle: $260 billion

Amazon: $137 billion

Alphabet: $85 billion

Many agreements extend for 15 years or longer.

The structure allows technology companies to secure capacity without directly owning every data center or recognising all associated obligations as conventional corporate borrowing. Infrastructure funds, banks, private-credit managers and project developers finance construction based on long-duration tenant commitments.

A proposed $15 billion bank financing for a Texas data-center project associated with Anthropic reportedly involves Google technology and contractual support. This is increasingly typical of the new financing stack: a developer owns the facility, banks provide project debt, and a hyperscaler or major technology company supports the project through chips, capacity contracts or guarantees.

My read: the financing architecture is becoming more sophisticated, but not necessarily less risky. Moving obligations into leases and joint ventures changes where leverage appears. It does not make the economic obligation disappear.

6. Credit markets are beginning to differentiate between AI borrowers

S&P lowered Oracle’s credit rating to BBB-, one notch above speculative grade. Oracle had approximately $129.5 billion of debt and debt-to-EBITDA of roughly 4.3 times, while several hyperscaler peers remain below one times. Oracle also has around $260 billion of uncommenced data-center leases and expects fiscal 2027 capital expenditure of up to $95 billion.

Oracle bonds are reportedly yielding approximately 7% to 8%, demonstrating that investors are demanding considerably more compensation for AI-related borrowers without hyperscaler-level balance sheets.

SoftBank also faces substantial funding requirements. Its announced AI and OpenAI-related commitments exceed $60 billion, including approximately $30 billion due during the second half of 2026. It also has a $40 billion bridge facility maturing in March 2027 and a $20 billion loan backed by Arm shares. A planned OpenAI-backed financing has reportedly been delayed as lenders assess the structure.

Why it matters: capital availability remains excellent for strategically important AI assets, but the cost of capital increasingly depends on balance-sheet quality, collateral and contractual revenue. The market is no longer treating every AI obligation as equivalent to Microsoft or Alphabet credit.

7. Infrastructure managers are raising record funds for digital and energy deployment

KKR closed its fifth global infrastructure fund at $19.2 billion, its largest infrastructure vehicle to date. The fund will concentrate on North America and Western Europe, with digital infrastructure and energy transition assets among its principal targets. KKR’s infrastructure platform manages approximately $120 billion.

The fund had already committed more than $9 billion by closing. KKR reported approximately $34 billion of infrastructure-related inflows during the second quarter, showing that institutional capital continues to view digital infrastructure, power and energy systems as long-duration portfolio allocations rather than temporary technology trades.

Infrastructure M&A is also expanding around power demand. Williams agreed to acquire Momentum Midstream for up to $5.5 billion, including cash, assumed debt and stock. Momentum owns more than 4,000 miles of pipeline and approximately 6 billion cubic feet per day of gathering capacity. Williams is also proceeding with its $1.5 billion Delta Access project, which is intended to serve electricity-generation and LNG demand.

Why it matters: AI-related power demand is affecting valuations far beyond data-center REITs. Pipelines, generation, transmission, cooling, construction and electrical equipment are all becoming components of the compute-capacity supply chain.

8. Government programmes are becoming strategic equity portfolios

On July 30, the U.S. Commerce Department announced preliminary agreements worth up to $874 million with seven semiconductor companies. Unlike conventional grants, the proposed transactions include minority government equity stakes.

The targeted technologies are closely linked to AI engineering constraints:

GlobalFoundries: up to $300 million for co-packaged optics, combining optical data transmission with AI processors.

Kepler Computing: up to $245 million for new memory technology.

Multibeam: up to $140 million for advanced packaging equipment.

Extropic, Thintronics, OBSIDIA and Aeluma: funding for low-power computing, advanced materials, semiconductor authentication and optical components.

This represents an important change in industrial policy. The government is not merely subsidising a factory. It is seeking equity participation in companies developing technologies considered strategically important.

China is using a more established version of this model. CXMT, following its approximately $8.6 billion IPO, is discussing support for a second Beijing DRAM facility that could cost more than $10 billion. Financing talks reportedly involve Beijing government-backed entities and state-owned technology companies. The expansion could more than double CXMT’s overall wafer capacity.

Why it matters: advanced memory has become a national-security and industrial-policy asset because AI accelerators are ineffective without sufficient memory bandwidth. Governments are investing not only in leading-edge logic chips, but across the entire compute stack.

9. Corporate R&D is focusing on bottlenecks rather than general-purpose expansion

MediaTek’s board approved a discretionary financing budget of up to $5 billion for long-term growth initiatives, including AI data-center chips. Its first custom AI accelerator is expected to enter production in the fourth quarter of 2026, with another generation planned for 2028. Management increased its estimate of the 2027 custom AI-chip market to approximately $80 billion and is targeting a 15% to 20% share.

Samsung and Netlist entered a five-year patent licence and supply agreement covering server memory and high-bandwidth-memory technologies. Samsung will supply DRAM and NAND, purchase 10 million Netlist shares and settle existing patent litigation.

IonQ’s proposed $1.8 billion acquisition of SkyWater Technology received U.S. antitrust clearance. The transaction would combine a quantum-computing company with a domestic semiconductor foundry, giving IonQ greater control over fabrication and supply-chain execution.

Steris also announced a $600 million manufacturing, distribution and R&D complex in North Carolina focused on chemicals used for infection prevention and contamination control. The facility is supported by state and local economic-development programmes.

There was no major new aggregate R&D-intensity dataset released this week. The more useful signal is compositional: corporate and public investment is increasingly directed toward optics, memory, packaging, fabrication, power efficiency and specialised scientific manufacturing.

10. Corporate venture and cross-border strategic investment are becoming operational

Aramco Ventures led a $9.5 million Series A in India-based Mitti Labs, joined by Lightspeed India, Godrej, Cisco and other investors. Mitti combines satellite synthetic-aperture-radar data, AI, field measurements and digital twins to improve rice cultivation. The company claims its methods can reduce water use by approximately 40% and methane emissions by more than 50% without reducing yields.

The company is expanding from India into the Philippines and Indonesia. This is a good example of corporate venture becoming strategic deployment capital: the investor is funding a technology that could contribute to emissions reduction and carbon-market development across multiple operating regions.

Indian electric-scooter manufacturer River raised $120 million in Series C financing, led by Indian investors with participation from existing backers including Yamaha, Mitsui and Al-Futtaim. The financing is predominantly primary equity, with debt representing less than approximately 10% to 12%. River plans a second plant with annual capacity of roughly 700,000 to 800,000 vehicles.

Why it matters: strategic investors are increasingly providing market access, manufacturing relationships and regional distribution, not just minority capital.

11. IPO and M&A exits are beginning to recycle venture capital

Twelve venture-backed companies completed public offerings at valuations above $1 billion in July. CXMT was the largest, while other notable listings included Bending Spoons and Lime.

Venture-backed M&A exceeded $9 billion during July, with five acquisitions valued above $1 billion. Transactions included Nscale’s acquisition of Anyscale for approximately $1.65 billion and Cyera’s purchase of Oasis Security for $1 billion.

This is important for fund formation. Venture funds do not need every portfolio company to achieve a successful IPO, but they do need distributions. A sustained increase in strategic acquisitions and public listings would allow LPs to recycle capital into new funds rather than continually adding exposure to illiquid portfolios.

Sector concentration snapshot

Sector Current capital signal Scientific or engineering driver

AI models and infrastructure 53% of July VC; enormous lease and project-finance commitments Inference scaling, model training, agentic workloads

Semiconductors State equity programmes, MediaTek capex, Samsung-Netlist agreement HBM, co-packaged optics, memory bandwidth, packaging

Digital infrastructure and real estate KKR $19.2B fund; trillion-dollar lease pipeline Power-ready sites, cooling, fibre, tenant guarantees

Energy Pipeline M&A and infrastructure deployment Gas supply, generation, grid congestion, storage

Quantum IonQ-SkyWater vertical integration Domestic fabrication and specialised quantum hardware

Biotech Braveheart, Latigo and BlossomHill IPOs Cardiac myosin inhibition, non-opioid pain, targeted therapeutics

Transport River Series C and factory expansion EV manufacturing, battery systems, distribution scale

Climate and agriculture Aramco-Mitti strategic investment Satellite radar, AI agronomy, methane measurement

Robotics Unitree IPO process and large industrial rounds Physical AI, actuators, control systems, manufacturing scale

Bottom line

The capital cycle is moving from funding technological possibility to financing industrial execution.

The winning companies increasingly need several capabilities at once:

1. Proprietary science or engineering

2. Access to manufacturing and power

3. Long-duration financing

4. Strategic customers or government support

5. A credible route to utilisation and cash flow

The positive signal is that capital remains abundant across venture, infrastructure, private credit and sovereign programmes.

The risk is that long-dated leases, project guarantees and off-balance-sheet commitments are accumulating faster than proven AI revenue. That does not imply an imminent collapse. It does mean the next market distinction will be less about who can announce the largest project and more about who can operate it at high utilisation while earning a return above its cost of capital.

Pulse

Snapshot

Sentiment Read-Through

Run history