The bank is 200 to 325 basis points cheaper. It is also the lender that cannot hold the paper.

Which lender is actually available to you?

For a leveraged transaction below $500m of enterprise value, mostly not a bank. Direct lenders finance roughly 90% of buyout transactions in that range, individual bank hold sizes in middle-market leveraged lending have contracted from $75m to $100m down to $30m to $50m, and several large regional banks have exited the segment entirely since 2024 (ABF Journal, 19 March 2026). That is a structural retreat from a product, and it has not reversed.

It is important to separate that from a claim about current bank appetite, because the two are frequently confused and the second one is not true. In the July 2026 senior loan officer survey the net percentage of domestic banks tightening standards for commercial and industrial loans was 0.0% for large and middle-market firms and 1.8% for small firms, the Federal Reserve described standards as basically unchanged for firms of all sizes, and a net 26.8% of banks narrowed spreads over their cost of funds to large and middle-market borrowers (Federal Reserve, July 2026 senior loan officer survey, published 3 August 2026). Commercial and industrial standards are easier than their post-2005 midpoint, the only loan category of which that is true.

Banks are also winning share back where they want it. Bank commercial and industrial loans stood at $2,894.2 billion in June 2026, up 8.0% year on year after being flat through 2025 (Federal Reserve H.8), and across eleven super-regional banks commercial and industrial lending was the largest source of commercial loan growth in the second quarter of 2026, with two lenders explicitly citing share taken back from private credit and one adding $2.3 billion driven by middle-market lending (Trepp via GlobeSt, 3 August 2026). SPP describes bank lenders competing more aggressively for quality transactions than at any point in the recent cycle (SPP Capital Partners, July 2026). The bank bid is real. It is selective about what it bids on.

What is the price difference?

Between 150 and 325 basis points for the same seniority, widening as the borrower gets smaller. A bank cash flow facility for a business below $10m of EBITDA prices at S+350 to S+425, against S+550 to S+750 for a non-bank unitranche. Above $25m of EBITDA the bank prices at S+275 to S+350 against S+425 to S+575 (SPP Capital Partners, Market At A Glance, July 2026).

In all-in terms at a 1-month Term SOFR of 3.65% (CME, 4 August 2026), that is 7.15% to 7.90% from a bank against 9.15% to 11.15% from a direct lender for a sub-$10m borrower. On a $20m facility, 200 basis points is $400,000 a year. That is a real number and it should be treated as one, but it is not the whole comparison, because the two facilities are not the same instrument.

There is a separate cost that has nothing to do with lender type and is larger than most borrowers assume. Research covering bank loan pricing finds that the dispersion in rates paid does not appear to be due to risk, that over a third of firms behave as if they do not comparison shop, and that half of all firms appear to obtain only two quotes before picking a lender (Amiti, Kashyap, Kovner and Weinstein, Why Do Firms Pay Different Interest Rates on Their Bank Loans?, NBER Working Paper 34870, February 2026). The spread between running a process and not running one is plausibly wider than the spread between the two lender types.

What does the extra cost buy?

Leverage, speed, certainty and a single decision-maker. A bank cash flow facility for a business below $10m of EBITDA sizes at 2.00x to 2.50x senior. Unitranche structures in the core middle market run 4.00x to 6.50x depending on size (SPP Capital Partners, July 2026, and Lincoln International Private Credit Snapshot, as of 1 May 2026). If the transaction requires more than roughly two and a half turns at the small end, the bank is not a cheaper version of the same answer. It is a different and smaller answer.

It also buys structure. A single lender can hold the whole facility, which means one credit committee, one set of definitions, one amendment process and no syndication risk between signing and funding. For an acquisition on a deadline that is worth paying for, and it is why direct lenders hold the buyout market at the sizes they do.

What it does not automatically buy is looser documentation, and borrowers frequently assume it does. Covenant-lite structures account for 21% of direct lending deals, up from 4% in 2023, but 91% of those sit above $50m of EBITDA (Proskauer via ABF Journal, June 2026). Below that, maintenance covenants remain the norm (First Eagle Investments, March 2026, citing PitchBook LCD as of 28 February 2026). And the direction of travel is toward the lender: 98% of surveyed private credit lenders report notably stricter underwriting since the start of 2026, the share expecting looser documentation fell from 33% to 4% over the year, and 55% would not provide payment-in-kind flexibility on a new buyout (Houlihan Lokey, Q2 2026 Private Credit Survey).

Which one behaves better in a downside?

The honest answer is that it depends on how many people have to agree, not on what kind of institution they work for. A bilateral direct lender can waive on its own authority, and lower middle market lenders typically hold unassailable senior positions with clearer enforcement rights and more direct engagement with borrowers (First Eagle Investments, March 2026). A syndicated bank facility requires consents. A widely held loan can produce lenders with irreconcilable views: two funds holding the same stressed credit have been observed marking it nearly 40 points apart on a fair value to par basis (Octus, 11 May 2026).

Direct lenders do accommodate, and the current data shows how much. Amendment activity in direct lending rose 13% quarter on quarter, with covenant holidays up 14%, maturity extensions up 14% and sponsor equity infusions up 31% (Lincoln International, 11 February 2026, as of the fourth quarter of 2025), while headline covenant defaults stayed flat at 3.1% (Lincoln International, as of 31 March 2026). The flat default rate is the product of the accommodation, not evidence that it was unnecessary.

They also enforce, and out of court. Direct lenders foreclosed on $24.2 billion of principal in 2025 and $15.2 billion in the year to date, against $13.6 billion across the preceding three years combined, with close to 75% relating to 2021 and 2022 vintage deals (Lincoln International, as of 31 March 2026). What that recovers is genuinely contested and should be read as two numbers rather than one: Octus puts average recoveries on private credit restructurings near 50 cents on the dollar (Octus, 11 May 2026), while Fitch reports 70% to 90% with minimal realized losses across its rated private credit cohort (Fitch Ratings, 6 March 2026). Different populations, both current, and they should not be averaged.

One more input belongs in the comparison and rarely appears in a term sheet: whether the lender can fund. Direct lending volume fell 55% quarter on quarter to $33.6 billion in the second quarter of 2026 across 154 deals, the weakest since the second quarter of 2023, even as fundraising rose to $16.25 billion from $1.3 billion (Preqin and PitchBook LCD via Reuters, 10 July 2026), and non-traded business development companies fielded redemption requests well above their 5% quarterly caps, with one manager's two funds honouring roughly 27% and 13% of requests (AltsWire, 6 July 2026). A fund managing a gate has less appetite for a delayed draw or an incremental tranche than its term sheet suggests.

How should the choice be run?

As a limited process rather than a relationship conversation. Approach a handful of direct lenders and a bank for the revolver at the same time, on the same information, with the same requested structure, and compare the term sheets on all four axes rather than on the headline spread. A second quote is the only thing that reliably holds pricing and leverage, and the research is unambiguous that most borrowers do not obtain one (NBER Working Paper 34870, February 2026).

Compare the total cost rather than the coupon. Original issue discount is currently doing work that the spread does not show: the share of lenders quoting at 99 or tighter fell from 75% to 56% over the year while quotes at 98.5 more than doubled from 18% to 42% (Houlihan Lokey, Q2 2026 Private Credit Survey), and lower middle market unitranche original issue discount runs 2% to 3% at funding with call protection at 102, 101 and par (ABF Journal, 1 June 2026). Add the fee, the discount, the undrawn cost on the revolver and the call protection before the coupons are compared.

And consider a split rather than a single answer. A bank revolver alongside a direct lender term facility is common for a reason: the revolver is the cheapest money in the structure and the bank is the natural holder of it, while the term debt goes where the leverage is available. That structure also gives the borrower two relationships rather than one at the point where it matters, which is worth more than the arithmetic suggests. The one thing to check before splitting is where the springing covenant sits, since a covenant that comes into existence once revolver usage passes roughly 40% is created by the cheap part of the structure (Sidley Austin, 24 March 2026).