Accel backed CrowdStrike and SentinelOne, and 68 more pairs of competitors

We embedded 13,823 portfolio companies belonging to the 40 venture firms with the largest portfolios in the FundingWatcher startup file, and counted how often one firm holds two companies whose descriptions land on top of each other. Accel holds 69 such pairs, against 15.76 expected from a random portfolio of the same size. Source: FundingWatcher, September 2026.

Across all 40 firms there are 942 pairs against 318.7 expected, a ratio of 2.96x. We expected the big multi-stage funds to sit at the top. They do not. The top of the ranking is Salesforce Ventures at 12.67x, a corporate arm, and the firm furthest below random is Founders Fund at 0.36x, which holds fewer than five pairs where a random portfolio of 220 companies holds 2.8.

Key findings

1. The 40 firms hold 942 pairs of portfolio companies scoring 0.85 or above against each other, where 318.7 pairs are expected from size-matched random portfolios. That is 2.96x.
2. Accel holds 69 pairs against 15.76 expected, 4.38x. 17.5% of its 515 vectored portfolio companies sit in at least one pair, against 5.4% in the random control.
3. Salesforce Ventures tops the ranking at 12.67x, with 38 pairs against 3.0 expected. Global Founders Capital is second at 6.52x and Tiger Global Management third at 5.16x.
4. 19 of the 40 firms sit above 3x. 32 of 40 are at or above 2x.
5. Three firms sit at or below random: Founders Fund at 0.36x, Global Brain Corporation at 0.82x and Felicis at 0.99x.
6. Antler holds the most pairs of any firm, 119, but on a portfolio of 838 vectored companies that is 2.88x.
7. Named pairs at the top of the file include CrowdStrike and SentinelOne at Accel (0.927), AnyDesk and TeamViewer at Insight Partners (0.932), Amplitude and Heap at New Enterprise Associates (0.936), and Amplitude and Mixpanel at Sequoia Capital (0.914).
8. In 19 pairs at 0.85 both companies carry the unicorn flag, among them Clari and Gong at Sequoia Capital (0.927) and Glean and Moveworks at Lightspeed Venture Partners (0.887).
9. Cleaning removed 42 pairs at 0.80 before any count: 17 where one company bought the other, 12 where the same acquirer bought both, and 13 where one company appears twice under two names or two domains.
10. 533 of the 3,687 pairs at 0.80 have founding years more than 8 years apart, so some of what we count is two companies converging rather than two companies that competed from the start.

Accel holds 69 competitor pairs where a random portfolio of its size holds 15.76

Accel has 701 portfolio companies with a usable domain in the startup file, and 515 of them have a vector. Those 515 companies make 132,355 possible pairs, and 69 of those pairs score 0.85 or above. We drew 50 random portfolios of 515 companies each from the 87,210 funded startups in the file that have a vector, cleaned them by the same rules, and counted the same way. The mean was 15.76 pairs. So Accel holds 4.38 times the pairs a random portfolio of its size holds.

The share measure says the same thing without depending on how many pairs one crowded corner produces. 17.5% of Accel's 515 vectored companies sit in at least one pair at 0.85. In the random draws the figure is 5.4%.

The ten highest-scoring Accel pairs, after cleaning:

CrowdStrike and SentinelOne both sell endpoint security, and Accel is on both cap tables. SentinelOne and Vectra AI score 0.918, CrowdStrike and Vectra AI 0.903, so Accel's security holdings form a small cluster rather than one coincidence. InVision and Miro both describe themselves as visual collaboration platforms, at 0.926. Further down the file Accel holds Flink and Trade Republic at 0.903 and Public and Trade Republic at 0.853, three retail brokerage and trading apps.

Salesforce Ventures holds 12.67 times the pairs of a random portfolio, more than any other firm

The ranking is by ratio at 0.85. Portfolio is portfolio companies with a usable domain, with vector is how many of those have an embedding, expected is the mean over 50 size-matched random draws, and share in a pair is the share of the vectored portfolio sitting in at least one pair. Counts under 5 print as <5.

Salesforce Ventures holds 38 pairs where 3.0 are expected, 12.67x. 20.5% of its vectored portfolio sits in at least one pair, against 2.5% in the random control. That is the highest share of any firm in the set. Reading its pairs explains most of it: BigMachines and SteelBrick (0.947) both sold quote-to-cash software, Jitterbit and Workato both sell integration and automation (0.900), and three of its top ten pairs are Salesforce implementation consultancies scoring each other, Beryl8 against Traction on Demand at 0.894 and Beryl8 against OSF Digital at 0.894.

We expected the household-name multi-stage firms to run high, because they write hundreds of cheques a year across every category. Sequoia Capital is 2.82x, Andreessen Horowitz 2.16x and Kleiner Perkins 2.0x, all in the bottom half of the ranking. The firms above 4x are a corporate arm, two firms known for funding a category broadly and fast, Accel, and Intel Capital.

Founders Fund is the one firm well below random. It has 220 vectored portfolio companies and fewer than five pairs at 0.85, where a random 220 companies produce 2.8. Its share in a pair is 0.9% against 2.4% for the control. At the looser 0.80 threshold it holds 13 pairs against 12.1 expected, 1.07x, so the effect is not an artefact of a single threshold. Global Brain Corporation at 0.82x and Felicis at 0.99x are the two other firms at or under random.

The median firm sits at 2.94x. 19 of the 40 firms are above 3x. Another 13 are at or above 2x and under 3x, and 5 are between 1x and 2x. Summed over all 40 firms the count is 942 pairs against 318.7 expected, 2.96x. At the looser 0.80 threshold it is 3,687 pairs against 1,438.9 expected, 2.56x.

19 pairs at 0.85 are two unicorns in the same portfolio

The highest-scoring pairs in the file are Gullak.Money and Jar at 0.963, held by Edward Lando and by Pareto Holdings, and AeonCharge and Emobi at 0.959 at Goodwater Capital. The better-known names sit lower in the list. The table below lists the pairs where our file flags both companies as unicorns.

Sequoia Capital holds Clari and Gong at 0.927, two revenue intelligence platforms sold to the same sales teams, and Cockroach Labs and PingCAP at 0.882, two distributed SQL databases. Lightspeed Venture Partners holds Glean and Moveworks at 0.887, both enterprise search and AI assistants for internal support. Insight Partners holds AnyDesk and TeamViewer at 0.932, two remote desktop products, the highest-scoring named pair in the set. New Enterprise Associates holds Amplitude and Heap at 0.936, and Sequoia Capital holds Amplitude and Mixpanel at 0.914, so Amplitude appears against a different product analytics rival in two different portfolios. Another 183 pairs at 0.85 have a unicorn on one side only.

Cleaning removed 42 pairs, 17 of them one company that bought the other

Some pairs of look-alike companies in the same portfolio are one company, or two companies that ended up under one roof. We dropped those before counting, and we dropped them from the random control too, so neither side gets an advantage.

• 17 pairs where one company's Crunchbase acquirer is the other company. Hearsay Systems and Yext scored 0.966, and Yext bought Hearsay Systems. Apttus and Conga scored 0.953, and Conga and Apttus merged.
• 12 pairs where the same acquirer bought both. Cumulus Networks and Mellanox Technologies scored 0.883 and NVIDIA bought both. Some of these were real competitors before the buyer arrived; we dropped them anyway.
• 13 pairs that are one company listed twice. Ten of them matched on the name after we stripped legal and filler tokens, including Rossum and Rossum Ltd and FinetuneDB against itself. Three matched on the domain label under two top-level domains, including Keyzy and Keyzy UK.

That is 42 pairs at 0.80 and 31 at 0.85. All 42 are in excluded_pairs.csv with the reason, so anyone who disagrees with a rule can put those pairs back. Rows sharing a domain never reach the pairing step at all: 290,757 rows with a usable domain collapse to 288,215 company records first.

The name rule is blunt on purpose and it costs us two pairs that may be different companies: NOVO (buildingnovo.com) against Novo (novo.eco) at 0.888, and Apollo.io against Apollo (withapollo.com) at 0.884.

533 of the 3,687 pairs at 0.80 were founded more than 8 years apart

Each vector was built from the company's homepage this month, so it carries the company as it trades now and not as it looked at the time of the investment. Two companies that drifted into the same market years apart score the same as two that were competitors from the first week. We cannot separate those two cases, so we flag the pairs where the second reading is most likely: 533 of the 3,687 pairs at 0.80 have founding years more than 8 apart, and flag_year_gap in pairs.csv marks them.

Accel's KDS and WegoPro are the clearest example. KDS was founded in 1994 and WegoPro in 2018, and both now describe travel and expense management for businesses, at 0.928. We read that match as two companies converging on one market rather than one firm funding two companies that competed from the start. The flag does not catch everything: a pair founded a year apart where one company pivoted later looks identical to a pair that competed from the start.

What this data cannot answer

The investors column is a flat list of names with no round, no date and no lead role. Three things follow.

We cannot say which of the two deals came first, so nothing here is evidence that a firm backed a challenger to a company it already held. A $200,000 angel cheque and a lead Series B count the same in this measure, so a firm that writes many small cheques is treated like one that leads.

Four permalinks with enough portfolio companies to make the top 40 have no row at all in the investor file, so their types could not be tested against the accelerator rule and they were dropped: sequoia-capital-china, tencent, google-for-entrepreneurs and easme. sequoia-capital-china, now HongShan, has 346 portfolio companies here and would almost certainly have qualified as venture capital, but it has no row in the investor file, so we could not test its type and left it out.

Vector coverage runs from 61.0% at Right Side Capital Management to 93.5% at HTGF. A firm at 61% is measured on two thirds of its book. Coverage cannot inflate a ratio, because each random draw is the same size as the vectored portfolio, but a firm with low coverage has more room for pairs we never saw.

Accel's own site lists 776 companies, and we hold 701

We checked one firm against a public number. Accel's relationships page at accel.com/relationships headed its list Companies (776) when we fetched it on 15 September 2026. Our file holds 701 Accel portfolio companies with a usable domain, 90.3% of that. We did not check the other 39 firms against their own sites.

Data funnel

The random frame is 87,210 domains out of the 107,436 funded domains in the file, 81.2%. It is not a sample: every one of the 107,436 was requested, and that is what came back with a vector.

Method

The 40 firms are the investors with the most portfolio companies in the startup file whose Crunchbase types contain none of accelerator, incubator, government_office, entrepreneurship_program or university_program. Corporate arms (Salesforce Ventures, Intel Capital, Google Ventures), two banks (Goldman Sachs, Silicon Valley Bank), an angel group (Alumni Ventures), a syndicate (Gaingels) and one individual (Edward Lando) all clear it and are in. The smallest firm in the set has 191 portfolio companies with a vector, Menlo Ventures.

Each company is one 1,024-dimension embedding. The text embedded answers the same questions about every company in the same order, taken from its current homepage: the product, the buyer, the delivery model, the market and the pricing model. We score a pair by the dot product of the two vectors after scaling each to length one. We reuse two of the three thresholds from our YC batch post: 0.80 is roughly "same product category" and 0.85 is "a reader would call these competitors".

A firm with 500 portfolio companies has 124,750 possible pairs and will hold some close ones by arithmetic alone, so every firm is compared against itself at random. For each firm we draw 50 random portfolios of the same size as its vectored portfolio from the 87,210 funded startups with a vector, clean them by the same rules, and count pairs at the same thresholds. expected is the mean of those 50 draws. pairs.csv stores scores to six decimal places and we apply the threshold at full precision. The random number generator is numpy default_rng, seeded with 17 so the draws repeat.

Nothing here is computed from a sample. The firm ranking, the pair counts and the random frame all run on the full file.

Data

Everything in this report is in the files below, under CC BY 4.0.

• firms.csv, one row per firm: portfolio size, vector coverage, pairs, expected pairs, excess, ratio and share in a pair, at both thresholds
• pairs.csv, 3,687 rows, one per cleaned pair at 0.80 or above: firm, both names, both domains, both one-liners, the score, both founded years, flag_year_gap and unicorn_a/unicorn_b. The 19 both-unicorn pairs above are the rows where unicorn_a and unicorn_b are both 1
• excluded_pairs.csv, the 42 pairs dropped before counting, with the reason
• README.md and METHOD.md

No vectors are published.