Capital Formation Monitor: Week Ended July 29, 2026
What changed
Capital is no longer merely funding AI compute capacity. It is increasingly assuming the construction, power-price, transmission, technology and credit risks required to make that capacity operational.
The week’s major developments were:
-
A proposed $100 billion data-center and energy campus moved the AI megaproject benchmark materially higher.
-
Venture capital concentrated further into semiconductor interconnects and automated chip design.
-
China’s state-backed venture model produced an extraordinary public-market exit through CXMT.
-
Insurance balance sheets emerged as the next major source of private-credit demand, even as retail investors remain concerned about liquidity.
-
Biotech’s IPO and crossover market continued reopening around clinically differentiated science.
1. The AI megaproject moved from multi-billion to $100 billion scale
NextEra Energy and Brookfield announced plans on July 29 for a data-center campus at the former Paducah uranium-enrichment site in Kentucky with an estimated total investment of
$100 billion. Brookfield would own and operate the planned 1.8 GW campus, while NextEra would provide approximately 2 GW of gas-fired generation and 2.6 GW of battery storage. Completion is targeted by 2032. (Reuters)
The structure is revealing. Power generation, battery capacity, data-center ownership and tenant pricing are being designed together rather than financed as separate assets. Customers are expected to pay premium electricity rates intended to insulate ordinary
ratepayers from project costs. (Reuters)
Digital Realty’s July 23 earnings provided a more immediate confirmation that contracted data-center real estate remains strong. The REIT raised its annual funds-from-operations forecast because of cloud and AI demand. (Reuters)
Why it matters: financeable electricity has displaced GPUs as the first-order constraint. The viable AI campus is now an integrated industrial system comprising power, storage, cooling, networking, land, tenant credit and construction financing.
2. Technology companies are absorbing energy-project risk directly
U.S. clean-power purchase agreements reached a record 10.4 GW in the first quarter of 2026, while technology companies increasingly accepted transmission, tariff and equipment risks previously borne by project developers. Google’s 1.9 GW agreement with
Xcel Energy is one example. (Reuters)
The shift is driven partly by sharply higher contract prices. Average solar PPA prices have almost doubled since 2020 to approximately $61.40 per MWh, while wind agreements reached approximately $83.79 per MWh. Developers are also pre-purchasing equipment to
preserve eligibility for federal tax credits. (Reuters)
A separate voluntary U.S. initiative would encourage data-center companies to fund or build their own power infrastructure, but consumer groups question whether non-binding commitments will prevent costs from migrating to utility customers. Reuters reported
that data-center demand represents a substantial portion of rising capacity costs within the PJM power market. (Reuters)
What changed: hyperscalers are evolving from electricity customers into project-finance counterparties and quasi-utilities. This accelerates deployment, but it exposes technology companies to permitting delays, construction overruns and commodity-price
risk.
3. Venture funding moved deeper into the semiconductor bottleneck
Eliyan raised $145 million in Series C financing at a $1 billion valuation on July 29. The company develops chiplets and licensable interconnect technology intended to reduce the data-transfer bottleneck between processors, a problem that can leave expensive
AI accelerators underutilized while waiting for data. Cisco, Lumentum and semiconductor industry investors participated. (Reuters)
ChipAgents separately added $60 million to its Series A, bringing total capital raised to $131 million. Its AI agents automate semiconductor design and verification tasks, including identifying functional errors before chips enter fabrication. The company
has expanded its collaboration with Nvidia to develop a specialized chip-design model. (Reuters)
By contrast, DeepSeek reportedly paused its second external fundraising process. The reason was not disclosed, but the decision demonstrates that even prominent foundation-model companies are reassessing capital needs and financing strategy. (Reuters)
Early versus late stage: the week did not produce another foundation-model mega-round. Instead, capital targeted Series A and Series C companies solving measurable engineering constraints. That is healthier than funding another generic application layer,
although it remains concentrated in companies adjacent to large compute budgets.
Scientific link: AI is now being used to design the chips that run AI. The potential feedback loop is material: faster design verification can shorten development cycles, while interoperable chiplets can let customers assemble custom accelerators without
relying entirely on one vendor’s proprietary architecture.
4. China’s state venture model achieved a spectacular public exit
Chinese memory producer CXMT raised 57.92 billion yuan, approximately $8.6 billion, in Asia’s largest IPO of 2026 and began trading in Shanghai on July 27. Proceeds are intended for production expansion, semiconductor R&D and working capital. (Reuters)
CXMT’s shares rose nearly 500% during the debut, although its limited public float contributed to the extreme move. (Reuters)
The more important capital-formation story is the state’s role. Hefei government-linked investors initially helped establish CXMT in 2016 and now hold approximately 36.8% of the company, a stake valued near $31.5 billion at the IPO price. Provincial and national
semiconductor funds increase effective state-linked ownership further. (Reuters)
Why it matters: CXMT is a working example of government functioning as founder, venture investor, industrial planner and long-duration shareholder. The listing recycles capital-market valuation into China’s semiconductor self-sufficiency strategy, particularly
in DRAM and advanced memory required by AI systems.
The danger is obvious: spectacular state-backed winners can encourage overinvestment and copycat projects, particularly when local governments expect industrial policy to generate venture-style returns.
5. IPO markets are opening through multiple channels
Hong Kong reduced listing thresholds on July 24, including cutting the minimum market capitalization for weighted-voting-rights companies from HK$40 billion to HK$20 billion. Confidential filing was also extended to all prospective issuers. Hong Kong
IPOs had raised $33.8 billion through July 21, more than twice the comparable 2025 level. (Reuters)
In U.S. biotech, Scribe Therapeutics raised approximately $128.7 million, and its shares rose 44.3% on their first trading day. Scribe’s lead programme uses epigenetic editing to suppress PCSK9 expression and lower LDL cholesterol without directly cutting
DNA, potentially reducing some of the risks associated with permanent gene editing. (The
Wall Street Journal)
Apnimed also launched an offering seeking up to $160 million at a valuation of up to $608 million. Its late-stage programme is developing an oral drug for obstructive sleep apnea, a condition generally treated using mechanical devices rather than pharmaceuticals.
(Reuters)
Why it matters: the IPO window is widening, but it is not indiscriminate. The strongest offerings combine strategic infrastructure, manufacturing capacity or credible clinical differentiation. Public investors are still imposing meaningful valuation
discipline on leveraged or weakly differentiated issuers.
6. Private credit is shifting from wealthy individuals to insurers
A Marsh survey found that 57% of insurers expect to increase private-credit allocations during the next 12 to 24 months. Interest is strongest among large and life insurers and is broadening beyond conventional middle-market direct lending into private
placements, asset-based finance, structured credit and investment-grade lending. (Reuters)
This institutional demand contrasts with continuing caution among wealthy investors in semi-liquid credit vehicles. Secondary markets are expanding to provide exits, while regulators are scrutinizing the connections among private equity ownership, insurers,
affiliated assets and reinsurance arrangements. (Reuters)
Blackstone’s second-quarter results illustrate both sides. The firm raised almost
$70 billion of new capital, total assets reached $1.35 trillion and investor withdrawal requests slowed. It also generated $31.8 billion from asset monetizations, including data-center and infrastructure holdings. Nine of its ten largest appreciating
investments were reportedly AI-related. (Reuters)
Why it matters: private credit is not contracting uniformly. Capital is migrating toward investors with long-duration liabilities, especially insurers, while retail-style liquidity promises are being tested.
The risk is that insurers may treat illiquid private assets as natural liability matches while underestimating correlated exposure to data centers, software disruption and technology-linked structured credit.
7. Corporate R&D is moving toward fabrication, interconnects and life-science workflows
Intel raised its 2026 capital-expenditure forecast from $18 billion to $20 billion, citing demand for data-center CPUs used alongside AI accelerators. Earlier CPU orders exceeded available production, supporting management’s increased spending plans.
(Reuters)
Dassault Systèmes agreed to acquire life-sciences software company ArisGlobal for
$1.8 billion in cash, with another $200 million contingent on future AI-linked revenue. The acquisition combines clinical, regulatory and pharmacovigilance workflows with Dassault’s life-sciences software platform. (Reuters)
Corporate venture activity also appeared directly in semiconductor rounds. Cisco and optical-networking supplier Lumentum backed Eliyan, while Nvidia is collaborating technically with ChipAgents. (Reuters)
What changed: strategic investment is becoming less about minority exposure to interesting startups and more about securing technologies that improve the investor’s own product architecture, supply chain or customer economics.
8. Biotech crossover capital is returning before broad generalist capital
Goldman Sachs data showed hedge-fund positioning in U.S. healthcare stocks near a five-year high. Investors increased exposure to life-science tools, healthcare equipment and pharmaceutical companies amid stronger financing conditions, AI-assisted research
and rising M&A expectations. Healthcare deal volume could reach approximately $173 billion in 2026. (Reuters)
Approximately 24% of hedge-fund launches in 2026 have reportedly focused exclusively on healthcare, the largest share since 2009. (Reuters)
This is a form of crossover capital returning through public markets rather than traditional late-stage private rounds. Investors can buy listed clinical and research-tool companies while retaining daily liquidity, rather than accepting long private holding
periods.
Why it matters: capital may rotate toward biotech as investors demand evidence that AI investment can produce economic value outside data-center construction. However, the investable winners are likely to be companies with clinical data, manufacturing
tools or genuine improvements in discovery productivity, not firms that merely append “AI” to conventional drug-development processes.
9. Sovereign and cross-border deployment became more diversified
Singapore’s GIC announced plans to allocate another $30 billion to hedge funds over three years, emphasizing global macro, quantitative and multi-strategy managers that can adjust exposures rapidly. GIC also said it is investing across AI infrastructure,
model and product developers, and companies adopting AI operationally. (Reuters)
GIC’s 20-year annualized real return declined to 3.4%, its weakest since 2020. The allocation change therefore appears partly defensive: maintain AI exposure while using more flexible managers to manage valuation, geopolitical and interest-rate risk. (Reuters)
Saudi Arabia’s PIF-led consortium also received European Commission merger approval for its
$55 billion acquisition of Electronic Arts, although a separate foreign-subsidy review remained pending. The deal is a reminder that sovereign capital is expanding beyond infrastructure into intellectual property, digital platforms and entertainment
ecosystems. (Reuters)
In transport, Ford and Geely formed a 66%-34% joint venture to manufacture Geely electric SUVs at Ford’s Valencia plant beginning in 2028 and jointly develop a European model. The structure pairs Chinese EV engineering with underutilized European manufacturing
capacity and local-content compliance. (Reuters)
Investment conclusion
The central development this week was the industrialization of risk allocation.
Capital providers are no longer satisfied with buying exposure to AI growth. They are deciding explicitly who bears:
-
power-price and transmission risk;
-
construction and permitting risk;
-
semiconductor-design risk;
-
private-credit liquidity risk;
-
clinical-development risk;
-
and geopolitical supply-chain risk.
The strongest opportunities remain in genuinely scarce bottlenecks:
-
Power generation and interconnection
-
Data-center platforms with credible tenants
-
Memory, packaging and chip interconnects
-
AI-assisted semiconductor design
-
Clinically differentiated biotech
-
Research and manufacturing tools
-
Sovereign-backed industrial platforms with export potential
My read: the probability that the AI buildout stalls because capital is unavailable declined again this week. The probability that capital is misallocated rose.
A $100 billion campus, state-financed chip champions, insurer-funded private credit and corporate-backed chip startups can create extraordinary productive capacity. They can also create extraordinary excess capacity if inference revenue, electricity economics
or customer demand fail to justify the installed asset base.
The market’s next phase will be less about announcing capex and more about demonstrating
utilization, cash yield and scientific productivity per dollar invested.