Salesforce Paid ~$30 Million for Loops to Teach Tableau Why Your Metrics Changed. The Scout Agent Is the Signal That BI Is Dead and Agentic Analytics Has Arrived.

The Financial Times reported on October 16–17, 2026 that Anthropic is structuring a $60 billion debt package to fund the procurement of approximately 1.5 gigawatts of Google-designed TPU chips. The deal, known internally as Project North Star, would be the largest AI chip financing arrangement in history. Blackstone-managed funds are being offered $9 billion of the junior tranche at a yield above 10%. Broadcom, which co-develops the chips with Google, would guarantee the senior tranche.

Neither Anthropic nor Blackstone confirmed the terms as of the time of reporting. The structure described comes from people familiar with the talks, per the Financial Times. But the structure itself, regardless of whether final terms shift, tells a story that enterprise AI teams should understand: this is what competitive frontier AI costs in 2026, and the capital structure required to maintain it has implications that run from chip procurement all the way to the API pricing that developers and enterprise buyers pay.

$60 Billion of Debt to Fund Google Chips: What Project North Star Is

To understand Project North Star, it helps to start with the underlying asset: Google-designed TPU chips. Tensor Processing Units are Google's custom AI accelerators, designed specifically for the matrix multiplication operations that dominate deep learning workloads. Google has been developing TPUs since 2016; the current generation — deployed in Google's own data centers and made available to external customers through Google Cloud — is designed for both training and inference at scale.

Anthropic has been deepening its relationship with Google since Google's initial equity investment in 2023, followed by an expanded compute partnership that represented billions of dollars in committed cloud spend. Project North Star takes that relationship a step further: rather than simply paying Google Cloud rates for TPU access, Anthropic is financing the procurement of 1.5 gigawatts of dedicated TPU capacity through a structured debt arrangement that is collateralized by Anthropic's own chip lease payments.

1.5 gigawatts of compute capacity is a meaningful milestone to benchmark against. For reference, a single high-end AI training cluster might consume 10–50 megawatts of power. 1.5 gigawatts is therefore the equivalent of 30 to 150 large training clusters — enough to simultaneously serve large-scale inference workloads while continuously running the training and fine-tuning operations required to maintain and advance frontier model capabilities. The LLM CapEx bubble analysis published earlier this year documented the scale at which AI infrastructure investment was accelerating; Project North Star is a concrete data point in that acceleration.

The deal is called the largest AI chip financing in history because no comparable transaction has been publicly disclosed at this scale. OpenAI's Stargate partnership committed $500 billion in total AI infrastructure investment but structured it as a combination of equity, debt, and committed purchasing from multiple corporate partners rather than a single debt facility. Microsoft's compute for MAI-Base-2 — reported as 16,000 GB200 GPUs — was funded from Microsoft's balance sheet rather than debt financing. Project North Star, as a standalone debt facility for a single company's compute procurement, is without precedent.

The Architecture of $60 Billion: Senior, Junior, and What Each Tranche Does

The $60 billion is not monolithic. It is divided into two tranches with materially different risk profiles, yield expectations, and collateral structures. Understanding the distinction is essential for interpreting what the deal says about Anthropic's capital structure.

Senior tranche: $42 billion. The senior tranche is reportedly guaranteed by Broadcom, which co-develops the Google TPU chips in partnership with Google. Senior debt holders have first priority on repayment in any default scenario — they get paid before junior holders, which means they face lower risk of loss. The Broadcom guarantee adds a further credit enhancement: if Anthropic cannot service the debt, Broadcom's guarantee provides a backstop. This structure allows Anthropic to raise a very large amount of senior debt at rates closer to Broadcom's creditworthiness (investment grade) than to Anthropic's own credit profile (BB+, sub-investment grade). The mechanics here mirror infrastructure debt financing in other industries: the underlying asset (the chips and the lease payments) combined with a creditworthy guarantee creates a financeable structure even when the primary borrower's standalone credit is below investment grade.

Junior tranche: $18 billion. The junior tranche is subordinated — junior holders are repaid after senior holders, meaning they absorb losses first in any adverse scenario. This subordination is compensated by higher yield. Blackstone-managed funds are being offered $9 billion of this junior tranche at a yield above 10%. The remaining $9 billion in junior debt is expected to be offered to additional investors in subsequent months, possibly at lower yields as the market establishes familiarity with the asset class. FourWeekMBA's analysis of the Blackstone terms notes that the fee and discount structure would push the total return above the stated coupon rate, making the effective yield higher than the nominal 10%.

The junior tranche is specifically collateralized by Anthropic's five-year chip lease payments. This means the cash flow structure is as follows: Anthropic leases the TPU capacity at an agreed rate over five years; those lease payments are the revenue stream that services the junior debt; if Anthropic's business generates the lease payment obligations and makes them on schedule, junior holders receive their yield; if Anthropic cannot make the lease payments — because the business shrinks, runs out of cash, or faces some other disruption — junior holders have recourse to the lease arrangement but not to Anthropic's general assets ahead of other creditors.

Structure based on FT reporting from October 16–17, 2026; terms unconfirmed by Anthropic or Blackstone.

Why the Yield Is Over 10% When Similarly-Rated Borrowers Pay 7%

Per reporting from Newsbytes citing the Financial Times, similarly rated public bond market borrowers at Anthropic's credit tier pay approximately 7% on comparable debt. The 3-plus-point premium Blackstone is being offered represents the combined effect of several distinct risk factors.

Private credit versus public bond markets. Public bonds trade on exchanges, providing liquidity: a bond holder who needs cash can sell the bond. Private credit does not. Blackstone's $9 billion position cannot be easily sold — Blackstone is committed to holding it until maturity or some negotiated exit. Investors demand a liquidity premium to compensate for this lock-up. Depending on the structure and duration, private credit liquidity premiums typically run 100–200 basis points above equivalent public bond rates for investment-grade borrowers.

BB+ credit and novelty premium. BigGo Finance's reporting on Anthropic's credit profile indicates a BB+ rating from credit analysts — one notch below investment grade, at the top of the sub-investment-grade tier. This rating reflects Anthropic's revenue trajectory (strongly positive) against its cost base (extremely high) and balance sheet dynamics. Critically, AI chip lease-backed debt is a new asset class: there is no historical default rate data, no recovery rate analysis, no bankruptcy court precedent for how these obligations are treated in an insolvency. Institutional lenders price novelty through yield because they cannot actuarially quantify the risk the way they can for asset classes with 20-year track records.

Junior subordination. As a subordinated debt holder, Blackstone absorbs losses before the senior tranche. In a severe adverse scenario — say, Anthropic's revenue contracts sharply and the company cannot service its obligations — Blackstone's junior position could face partial or total impairment while senior holders recover in full. The subordination premium typically adds 200–400 basis points on top of the senior rate for this type of structure.

Adding these components: public comparable rate (~7%) + private credit premium (~1.5–2%) + junior subordination premium (~2–3%) = approximately 10.5–12%, which is consistent with the "above 10%" yield being offered plus fees and discounts that push the total return higher.

The market implication is that Blackstone, one of the most sophisticated institutional investors in the world, has looked at Anthropic's credit profile, revenue trajectory, the collateral structure, and the novelty risk premium required, and concluded that greater than 10% yield is sufficient compensation. That is not a statement of distress — it is a market clearing price for a specific risk profile at a specific moment in the development of AI chip financing as an asset class.

The Google-Broadcom-Anthropic Triangle

The AI infrastructure arms race coverage from earlier this year documented how frontier AI companies are pursuing compute strategies that bypass commodity hardware markets. Project North Star is the most sophisticated expression of that strategy to date.

Google designs the TPU chips. Broadcom co-develops them, handling the custom ASIC engineering that turns Google's architecture into manufacturable silicon. Anthropic uses them for training and inference. This three-party arrangement has implications for compute strategy, vendor relationships, and competitive positioning that go beyond the financing mechanics.

For Anthropic, TPUs instead of NVIDIA GPUs is a significant commitment. NVIDIA's H100 and H200 GPUs have been the dominant AI training hardware, and CUDA — NVIDIA's programming environment — is the most widely used framework for AI workloads. Anthropic's bet on Google TPUs at $60 billion scale is a bet that Google's custom silicon is competitive with NVIDIA hardware at training and inference scale, and that the compute relationship with Google is more strategically valuable than hardware vendor neutrality. The practical implication is that Anthropic's training infrastructure becomes deeply entangled with Google's chip development roadmap. If Google's next-generation TPUs underperform expectations, Anthropic's frontier capabilities are constrained in ways that a hardware-diverse strategy would not produce.

For Google, Project North Star deepens an already significant relationship. Google is simultaneously Anthropic's equity investor, its cloud provider, and now a key participant in the collateral structure of a $60 billion debt deal (through Broadcom's guarantee). The financial interdependence creates alignment but also raises questions about independence: Anthropic's ability to negotiate model pricing with enterprise customers, explore alternative chip suppliers, or manage cloud costs is constrained by the depth of its Google dependency. Equity investment and supply chain entanglement are different governance structures, and Anthropic is now subject to both.

For Broadcom, providing guarantees on $42 billion of senior debt is a meaningful commitment. Broadcom's guarantee is what makes the senior tranche financeable at investment-grade-adjacent rates; without it, Anthropic's BB+ credit profile alone would push the senior tranche toward higher yields. In exchange, Broadcom deepens its relationship with one of the most important buyers of custom AI silicon in the market.

Project North Star Builds on the June Apollo Deal

Project North Star does not emerge from a standing start. It builds on a financing precedent established in June 2026, when Apollo Global Management led a $35 billion debt package that funded more than 1 gigawatt of Anthropic compute capacity. That deal — structured similarly, with compute lease payments as collateral — demonstrated that institutional debt markets would accept AI chip financing as an asset class. It established the template that Project North Star follows at larger scale.

The two deals together paint a clear picture of Anthropic's compute financing strategy. Rather than purchasing compute infrastructure outright (which would require equity capital at a scale that would be dilutive), or renting compute on spot or reserved cloud markets (which is more expensive per unit of sustained capacity), Anthropic is using debt financing to lock in large volumes of dedicated compute at a cost structure that is more attractive than cloud spot markets while preserving equity capital for model development, safety research, and business operations.

Combined across both deals, Anthropic has financed approximately $95 billion in compute infrastructure during 2026 alone. To put this in context: $95 billion is larger than the annual revenue of most major technology companies. It is larger than Apple's capital expenditure program. It represents a bet that the compute Anthropic is locking in will be productive — generating inference revenue and training capabilities — over the multi-year horizon of these debt facilities.

The systematic nature of the two deals — similar structures, different lead investors, consistent collateral approach — suggests that Anthropic has a finance team executing a deliberate capital strategy rather than making opportunistic one-off decisions. That strategy involves building a large fixed-cost compute base funded by debt, betting that revenue growth will outpace debt service obligations, and using the resulting compute capacity to maintain frontier model performance. It is a high-conviction, high-leverage bet.

The True Cost of Frontier AI: What the Capital Stack Actually Looks Like

Project North Star, combined with Anthropic's equity financing history, reveals the full capital stack required to compete at the frontier. The picture is materially more complex than the headline equity valuations suggest.

Equity financing. Anthropic has raised approximately $7–9 billion in equity across multiple funding rounds from investors including Google, Amazon, Spark Capital, General Catalyst, and others. This equity funds operations, research, and the portions of compute not covered by debt financing.

Strategic equity from hyperscalers. Google and Amazon have made large equity investments in Anthropic that come with attached cloud computing commitments. Amazon's $4 billion investment included a significant AWS compute commitment. Google's investment included access to Google Cloud infrastructure. These are not pure equity investments — they bundle capital with guaranteed customer relationships.

Chip financing debt. The June 2026 Apollo deal ($35 billion) and Project North Star ($60 billion, pending) represent $95 billion in chip lease-backed debt. This is the largest component of Anthropic's total capital picture by a significant margin.

Estimates from public reporting; Anthropic does not disclose financials.

The comparison to OpenAI's Stargate is instructive. Stargate committed $500 billion in AI infrastructure investment but distributed that commitment across Microsoft, SoftBank, Oracle, and other partners — Anthropic's individual exposure is not comparable to a $500 billion corporate guarantee. Microsoft, by contrast, funded the MAI-Base-2 training run on its own balance sheet using NVIDIA GB200 hardware it owns or controls — a different capital structure that avoids the debt service obligations but requires larger upfront equity capital.

Modal's $750M Series D and $1 billion ARR represent a fourth model: serverless compute that matches capital expenditure to usage, eliminating the fixed cost structure entirely. Modal's model is financially more conservative but requires customers to tolerate the variable cost model; Anthropic's model locks in large compute capacity but requires the revenue growth to service fixed debt obligations.

What This Means for Claude Pricing and the API Economy

The capital structure of Project North Star has direct implications for how Anthropic must approach model pricing — and for how enterprise buyers should think about the sustainability of current Claude API pricing.

The mechanics are straightforward. Debt generates interest obligations that are fixed regardless of revenue. If Anthropic's blended interest rate across the Apollo deal and Project North Star averages 8%, the annual interest burden on $95 billion in total chip financing runs to approximately $7.6 billion. That is a minimum annual cash outflow that must be covered by revenue before any other expense — before model development salaries, before safety research, before commercial operations.

Claude API pricing in 2026 has come down significantly as the model performance frontier advanced and competition from open-weight models pressured margins. The question the Project North Star financing raises is how much further pricing can compress before debt service constraints become binding. A company that can grow revenue at 3–4x per year while managing a $7–10 billion annual interest burden can afford aggressive pricing competition. A company whose revenue growth moderates — as markets mature and competition from lower-cost providers intensifies — faces a conflict between competitive pricing and debt service coverage.

Enterprise buyers should treat this as a procurement risk factor rather than a reason to avoid Anthropic. The frontier model market's competitive dynamics are producing pricing pressure that benefits buyers; the capital structure behind the models is what sustains the frontier. The risk to price stability runs in both directions: competitive pressure pushing prices down, debt service obligations creating a floor below which prices cannot fall without impairing the company's ability to service its capital structure.

Enterprise AI Strategy: What Teams Building on Claude Need to Know

Understanding the capital structure behind your model vendor is not standard enterprise procurement practice. It should be. The anthropic IPO watch coverage documented the governance and financial transparency questions that Anthropic's private company status creates for enterprise buyers. Project North Star adds a new dimension: a vendor carrying $95 billion in debt has a different risk profile than a vendor carrying none.

What to watch: Anthropic's debt service coverage — whether the company's operating cash flow is growing faster than its debt obligations. A company that is building a debt-financed compute base while growing revenue faster than interest expense is on a sustainable trajectory. A company where revenue growth is decelerating while interest obligations are locked in is in a more difficult position. Neither situation is fully visible from outside Anthropic's private financials, but the scale of the Project North Star deal is itself a signal: $60 billion in additional financing implies a commitment to maintaining a level of inference capacity and training throughput that requires a corresponding level of revenue to be sustainable.

What to do: 1. Build for model flexibility. An AI application architecture that can swap models with minimal code changes is insulated from vendor capital structure risk. If Anthropic's pricing must increase to manage debt service, a flexible architecture allows migration; a tightly coupled implementation does not.

2. Negotiate multi-year commitments carefully. Multi-year API contracts at fixed rates are attractive when prices are falling; they become constraints if your vendor's capital structure requires price increases. Prefer shorter-term or volume-based commitments that preserve flexibility.

3. Monitor Anthropic's revenue disclosure signals. As a private company, Anthropic does not publish quarterly financials. But it makes revenue claims in press contexts (crossing specific ARR thresholds, growth rates) and its credit profile is being established through structured debt deals. These signals give partial visibility into whether the company's revenue trajectory is consistent with servicing its capital obligations.

4. Maintain active vendor relationship with alternatives. Google Gemini, OpenAI GPT, and open-weight models such as Meta Llama provide hedging options. The existence of alternatives is not an argument against using Claude; it is an argument for keeping the capability to switch developed and current.

Takeaway

Takeaway: Project North Star is not simply a financing story. It is the most precise public window available into what frontier AI actually costs in 2026. The $60 billion in debt, on top of the $35 billion June Apollo deal, maps the capital requirements of maintaining frontier model capabilities at the scale Anthropic has chosen. Blackstone's willingness to provide $9 billion at greater than 10% yield on a BB+ credit is evidence that institutional debt markets believe Anthropic's revenue is real, growing, and sufficient to service these obligations over time — a significant validation of the underlying business. For the broader enterprise AI ecosystem, the implications are structural: the frontier model market requires capital investment at a scale that only a small number of companies can access, and the companies that can access it are building compute moats measured in billions of dollars that cannot be replicated by smaller entrants. Enterprise teams building on Claude should understand that the pricing stability of their most capable models depends on Anthropic's ability to grow revenue faster than interest expense — and should build their AI infrastructure with the flexibility to adapt if that calculus shifts.