Answers from the documentation; checking which model is on duty
Hi - I answer from the OpenSmartRoute documentation: routing, the API, plans and quotas, self-hosting. Ask away, or open a support ticket if you need a person.
Grounded in the docs - follow a source before acting on it.
Blackstone says AI debt fears are overblown - OpenSmartRoute
Blackstone is lending billions to AI data centers and neoclouds. Jas Khaira says current debt worries are exaggerated because major tech companies back the loans.
Key points
Blackstone holds the largest AI compute stake outside hyperscalers.
CoreWeave debt reached about $50 billion recently.
Blackstone lent $2.3 billion to CoreWeave in 2023.
Blackstone invested $5 billion in Crux AI with Google.
Why it matters: Runners need to know if AI compute costs are rising or if debt risks will stop supply.
By OpenSmartRoute editorial · written through the router by writer-small
From TechCrunch AI - “Blackstone’s Jas Khaira says fears over AI-buildout debt are overblown”
Private equity giant Blackstone is putting itself at the center of the AI boom. Firm veteran Jas Khaira discussed this on stage at TechCrunch Disrupt last week. The firm is writing checks via various strategies. These strategies range from offering massive loans to data centers. They also include investing in frontier AI models and applications.
Khaira emphasized that Blackstone sets itself apart from traditional venture capitalists. Traditional investors usually focus on equity. Blackstone focuses on compute financing and flexible capital structures. These structures range from debt to equity. This approach allows them to fund projects that need significant capital.
Unlike earlier tech cycles, the transition to AI requires staggering amounts of investment. Earlier cycles, like mobile or SaaS, allowed companies to reach scale with moderate capital. The need for capital is precisely why Blackstone could play such a critical role. Blackstone has $1.3 trillion in assets under management. This massive pool of money supports the tech shift.
Khaira said the firm is the largest owner of compute outside of hyperscalers today. Hyperscalers are huge cloud providers like Google or Amazon. Blackstone invests in neoclouds and data centers. They use this compute strategically to support companies. This support helps businesses build their AI capabilities.
How Blackstone built its AI lending model
Blackstone pioneered a financing structure that allowed it to lend massive sums to venture-backed cloud neoclouds. Neoclouds are cloud providers that rent out AI GPU infrastructure. Blackstone analyzes both committed customer contracts and using GPUs pledged as collateral.
The firm looks at the nature of the customer before lending money. If the customer is an investment-grade company, the risk gets priced relative to that customer. Examples of such customers include Microsoft or Meta. These companies have massive balance sheets and solid earnings.
Khaira explained that lenders are charging higher premiums on top of already higher base interest rates. This happens to finance the AI buildout. The cost of capital to price in debt for a Meta or a Microsoft has widened out. Even tech giants with massive balance sheets are paying more to borrow now.
Google released TorchTPU to make AI models run on its chips instead of Nvidia hardware. This change aims to reduce reliance on Nvidia's dominant position in the industry.
Inception AI launched Mercury 3, which is up to eight times faster than GPT-6 Luna and Claude Haiku 5.5. The model uses a new diffusion approach to generate text at 1,068 tokens per second.
This model relies on the value of the hardware pledged. Blackstone grew more comfortable with the loan because CoreWeave had contracts with blue-chip customers. These contracts provide a safety net for the lender. The customer base ensures the debt is likely to be repaid.
The CoreWeave loan that started the trend
Blackstone’s first such deal came in 2023 with CoreWeave. CoreWeave is a neocloud that saw demand explode following the launch of ChatGPT. Despite having only “30 million of revenue and minimal profitability” at that time, CoreWeave approached Blackstone for a $2.3 billion loan.
CoreWeave was borrowing against the considerable value of its large collection of Nvidia GPUs. Nvidia also acted as an investor in this transaction. Blackstone used the GPUs as collateral for the loan. This collateral provided security for the lender.
Khaira noted that CoreWeave had contracts with blue-chip customers like Microsoft. These contracts made Blackstone more comfortable with the loan. That transaction quickly became the financing model for how Blackstone and other credit institutions provided billions in follow-on debt. They provided debt to CoreWeave and rival neoclouds.
The model focused on the strength of the customer base. If the customer is an investment-grade customer, the risk gets priced relative to that customer. This approach allowed Blackstone to lend to companies that were not yet profitable. The future revenue from these customers justified the loan.
New risks from rising interest rates
Khaira acknowledged that in recent weeks, the large debt burdens accumulated by these companies have begun to raise red flags. Coreweave’s debt alone has since ballooned to about $50 billion. This increase has caused concern among investors and analysts.
Altimeter Capital founder Brad Gerstner noted in a recent CNBC interview that the collateral is not good enough anymore. He stated that GPUs are not enough to secure cheap debt anymore. The market is thinking about what is the right price for neoclouds.
Interest rates ticked up in September. Hotter-than-expected inflation signals that borrowing costs may keep climbing. The flood of AI infrastructure debt offerings isn't helping either. These factors contribute to market jitters on the credit side.
Khaira said the market is thinking about the right price for these businesses in this moment. Interest rates are moving up, which affects valuations. Shares of CoreWeave dropped over 20% in recent days. This drop reflects the rising cost of debt concerns.
Firmus scrapped its planned IPO on the Australian Securities Exchange at a target valuation of roughly $30 billion. Khaira said the market wasn't there at that valuation. Consequently, that deal didn't get done. The market is recalibrating based on current economic conditions.
Why this matters for AI compute buyers
AI companies simply can't get enough compute to power their models. Their appetite continues to drive data center buildout. This demand creates a need for massive amounts of capital. Buyers must consider the cost of this capital when planning their projects.
Khaira downplayed worries of doom-mongers that the massive accumulation of AI infrastructure debt could lead to a widespread credit crisis. He stated that this doesn't feel like something where you've got 2008 banks that are shaky credits. The risk profile is different from the 2008 financial crisis.
Instead, he pointed out that the vast majority of data centers are built for hyperscalers. The debt financing them is ultimately backed “by the most creditworthy companies in the world.” This backing reduces the risk for buyers of compute services. Buyers can rely on stable infrastructure even if interest rates rise.
Beyond funding data centers, Blackstone is helping companies adopt AI through a $1.5 billion joint venture called “Ode with Anthropic.” This venture supplies forward-deployed engineers to help companies adopt AI. It helps integrate AI into operations, starting with Blackstone's 280 portfolio companies.
When asked if Blackstone-owned companies are seeing tangible gains from AI, Khaira conceded that there is definitely ROI. However, he noted that it is uneven. Technology adoption could move in a “sawtooth-type pattern.” This pattern can look like an oversupply of compute before adoption catches up.
How OpenSmartRoute helps manage these risks
Teams that route their requests through OpenSmartRoute gain control over compute costs during volatile market periods. The router scores every candidate on quality, cost, speed and safety. The team sets the weights per request to prioritize what matters most.
Hard rules pin a request to prevent unexpected expenses. For example, text with personal data stays on an on-premises model. A region or a cost cap is never crossed. This prevents budget overruns when interest rates rise and prices fluctuate.
A savings ledger shows what each routed request cost next to what the most expensive model would have cost. This ledger provides transparency into spending. Teams can see exactly how much they save by choosing the right model.
The hosted platform keeps a models catalogue with prices and public rankings built from real traffic. This catalogue helps teams make informed decisions about which models to use. Teams can update the catalogue with new models easily.
OpenSmartRoute works with any OpenAI-compatible provider, with open-weight models served locally. It supports MCP tools and A2A agents. This flexibility ensures teams are not locked into a single vendor. They can switch providers if costs become too high.
The input guard spots prompt injection and personal data before a request leaves. This security feature protects sensitive information from exposure. It ensures that data remains secure regardless of the model chosen.
How it compares
The AI buildout needs more money than previous tech cycles did. Mobile and SaaS companies grew with less capital. AI firms need huge amounts of investment to train models and build data centers. Blackstone has $1.3 trillion in assets under management. This size lets it offer loans and equity where others cannot.
Blackstone focuses on compute financing. It lends money to companies that own GPUs and data centers. It also invests in the models and applications that run on that compute. This mix of debt and equity is different from traditional venture capital. VCs usually take equity in startups. Blackstone uses debt structures backed by assets like customer contracts or pledged GPUs.
Questions this leaves open
The source does not say how Blackstone plans to use its $1.5 billion joint venture with Anthropic. The Ode joint venture supplies engineers to help companies adopt AI. It is unclear if this will scale beyond Blackstone’s 280 portfolio companies.
Blackstone’s financing model relies on customer contracts with blue-chip names. The source does not explain how it handles companies with smaller or unknown customers. The risk profile of these loans could change if those customers leave.
The article mentions that interest rates and inflation are pushing up borrowing costs. It does not say if Blackstone will raise loan rates for new deals. It also does not address how rising rates affect the value of GPU collateral.
What to do when running AI agents
Start by defining your catalogue of models, agents, tools or skills. Each entry should represent a specific capability you need. Add prices and performance metrics to each entry for comparison.
Set hard rules for your requests to enforce cost caps and safety boundaries. Ensure that personal data never leaves your secure environment. Define the maximum cost per request to avoid budget surprises.
Use the 'osr eval' tool to measure routing accuracy on your own prompts. Run this evaluation regularly to ensure your routing decisions remain effective. If the evaluation drops, the build will fail and you will need to fix it.
Monitor the savings ledger to track costs over time. Compare the actual cost against the most expensive model option. Adjust your weights in the scoring system if certain models consistently underperform.
Keep your models catalogue updated with new models from OpenAI-compatible providers. Test new models before adding them to your active catalogue. Ensure they meet your quality and safety standards before deployment.
Check the public rankings in the hosted platform to see how your models perform. Use real traffic data to validate the rankings against your own experience. This helps you make better decisions when selecting models for your agents.
IRONSCALES finds that 77% of organizations report AI tools acted on malicious content. The report links this to faster adoption than security controls.
