Pulley, Linera, Switchboard: three shutdowns, three different failure signals (2026)

Between September 14 and September 20, 2026, three funded companies said they were closing. Pulley, a cap-table platform that had raised more than $50 million and spent seven years taking on Carta, put a notice on its own website and handed its customers to the rival it was built to beat. 1 2 Linera, a layer-1 blockchain with $12 million from a16z crypto and Borderless Capital, ceased operations before its mainnet ever shipped. 3 Switchboard, an oracle network serving 550 feeds across eleven chains, gave the four largest Solana DeFi protocols six days to find another data provider. 4

This week's three closures point at three different quantities to watch: what the customer would use instead, what money had actually closed, and what the buyer could build alone. Each of the three is measurable before a shutdown is announced.

Startup / product	Event and sector	Disclosed cause or evidence boundary	Stakeholder outcome	Counter-metric
Pulley	Notice on the company's own website, reported on September 15, 2026; all operations and services end December 8. Seven-year-old cap-table and equity-management software. 1 2	The notice states no cause, and the company gave none to reporters. TechCrunch reported a former employee's public speculation that Pulley competed with spreadsheets rather than with Carta, and that AI makes spreadsheets cheap to maintain. That reading is inference. 5	Existing customers can migrate to Carta with assisted transfer and keep Pulley pricing for the first year, with the unused portion of their subscription credited. Prospects are being sent to Carta directly. 2	Share of revenue whose honest replacement is a free default rather than a competing vendor
Linera	Founder Mathieu Baudet announced immediate cessation in the project's official Discord on September 18, 2026. Layer-1 blockchain; mainnet never launched. 3 6	Disclosed: funding. The LNRA community sale on Sonar raised 848,300 USDC against a 1.5 million USDC minimum, so it could not close and all commitments were refunded; emergency financing talks then failed to fund development to mainnet. 7	Applications and the Discord community close. Community point balances stay recorded, with no guaranteed token, value, or rights attached. All sale contributions were returned. 3 7	Twelve-month plan covered by money already closed, excluding any round that has not closed
Switchboard	Switchboard Technology Labs announced on September 19, 2026 that every implementation is deprecated immediately, with support ending September 25. Oracle and data-feed infrastructure. 4 8	Disclosed: AI tooling lowered the cost of building an oracle in-house; the bear market compressed the budgets and launch schedules of the chains that were its customer base; large consumers now arrange data directly, bypassing the middleware. 4	Kamino Finance, Jito, MarginFi and Drift Protocol were directed to migrate to Pyth Network or RedStone within six days. 4	Cost of the customer's own build, and the share of revenue held by customers who can bypass the vendor

1. Pulley: the substitute sitting on the customer's desktop

The event

Pulley's own website now opens with the notice: the company will shut down and cease all operations and services on December 8, 2026. 1 The notice describes an exclusive partnership with Carta, the competitor Pulley was founded to challenge. Existing customers are offered assisted migration; prospective customers are pointed at Carta. 1

Yin Wu began building Pulley in 2020, after Microsoft bought her earlier Android app, Echo. 9 Over seven years the company raised more than $50 million from General Catalyst, Stripe, Founders Fund and 8VC. 2 Business Insider published the wind-down on September 15, 2026; TechCrunch followed the next day. 2 5

Customers who move keep their current Pulley pricing for a year, and Carta credits the unused portion of any Pulley subscription. 2

What the record supports

The notice states no cause, and the company gave none when TechCrunch asked. 5 Wu's own public statement thanked her team and investors, and said she and many of her strongest colleagues would build again: "though this is the closing of one chapter, I, along with many of our strongest team members, have no intention of riding off quietly into the night." 5

What the reporting adds is one former employee's public argument that Pulley's real competitor was the spreadsheet. 5 A cap table is a list of who owns what, and a founder can keep one in a spreadsheet. The claim is that maintenance, not the logic of the table, was the paid work — and that AI has made spreadsheet upkeep cheap enough that the paid version loses its reason to exist. 5

That reading is an inference; the company has published no cause of its own. The documented facts point to a product mismatch against the customer's default option. A wedge built as "better than the incumbent, for founders" can still lose to the thing the founder already has open on another screen. When the substitute is free and the incumbent is bundled, the paid product has to clear a higher bar than the alternative vendor's feature list.

The wind-down route carries the same signal. A company that set out to outlast Carta is closing by routing its own customers into Carta and holding their old price for a year. 1 2

The counter-metric

For every revenue line, ask what the customer does the week after your product disappears. Then compute:

the share of revenue whose honest replacement is a manual artifact or a tool the customer already pays for elsewhere, counted apart from revenue you win against named vendors;
the price premium you charge over that default, expressed as the customer's cost of switching rather than as a feature comparison;
win and renewal rates measured specifically against the do-nothing option, kept separate from rates against named competitors; and
the migration path you would have to offer on the way out, and what it implies about who you were really competing with.

A company whose honest answer is "they would go back to a spreadsheet an assistant now maintains" is pricing against a rival whose price is falling every quarter.

2. Linera: a milestone gated on money that had not closed

The event

Linera's founder, Mathieu Baudet, posted in the project's official Discord on September 18, 2026 that the company would "cease operations immediately" and had no plans to resume in the near term. 3 Baudet previously worked on Meta's Novi digital-wallet project, and Linera's technology ran many small blockchains, called microchains, in parallel to speed up transactions. 3

Linera raised $6 million in June 2022 in a seed round led by a16z crypto, then another $6 million in August 2023 led by Borderless Capital, for $12 million in total. 3 The second round was meant to fund headcount, protocol development, a testnet and expansion in Asia-Pacific. 3 Its mainnet never shipped. 7

Applications and the community server are closing. Point balances stay on the record, and the team said it cannot guarantee they will ever receive consideration. 10

What the record supports

Linera named its cause, and named it with numbers. Its community sale on the Sonar platform drew 848,300 USDC in commitments against a 1.5 million USDC minimum, roughly 651,700 USDC short. 7 The team had said before the sale opened that funds would be returned if the threshold was missed, so every commitment went back. With the sale closed, Linera went looking for emergency financing to carry development to mainnet and could not raise enough. 7

The team framed the decision around the money: "The decision does not reflect on the technology, the team, or this community," the announcement read, adding that it still hoped to finish the protocol later. 10

The diagnosis is burn rate against a financing gate. The last milestone before revenue — a working mainnet — was priced in money that had not closed. The public sale was the funding plan. It carried a published minimum, below which the company had no path. When commitments came in at 56.6 percent of that minimum, the plan had already failed, and the emergency round was a second attempt at the same thing.

The sale's threshold was also a public test of demand. It measured what strangers would pay for the network's token before the network existed. 7

The counter-metric

Take the next twelve months of your plan and split the spend by funding source. Then compute:

committed coverage: the share of that spend already in the bank or under contract, excluding any round that has not closed, any grant not countersigned, and any sale with a minimum threshold;
the threshold arithmetic for any conditional raise: at what percentage of target does the plan stop working, and what is the fallback at that number;
the date your next externally visible milestone ships on money you already hold; and
how much of the plan assumes the raise, restated as a percentage you would defend in a board meeting.

Linera's sale closed at 56.6 percent of its minimum. A company that knows before it opens a raise what it will do at 60 percent of target is running a plan; one that finds out afterwards is running the raise.

3. Switchboard: the layer customers learned to build themselves

The event

Switchboard Technology Labs announced on September 19, 2026 that all of its implementations are deprecated effective immediately, with remaining support ending September 25. 8 The statement opened "in light of recent exploits and after having exhausted all possible alternatives." 10

Oracle networks feed outside data, such as prices, into on-chain contracts, and DeFi protocols cannot liquidate or price positions without them. At the announcement Switchboard was serving more than 550 unique feeds across 53 programs on Solana and ten other chains, handling hundreds of millions of data requests a week. 4

Four Solana protocols — Kamino Finance, Jito, MarginFi and Drift Protocol — were told to migrate to Pyth Network or RedStone within six days. 4 For protocols holding open interest, an oracle migration means contract changes and testing on that clock. 4

What the record supports

Switchboard named three reasons, and all three describe what its buyers could do instead. 4 First, AI tooling has lowered the cost of building oracle infrastructure in-house — in the team's words, "AI has lowered the cost of building one's own oracle" — which erodes the pricing power a standalone network needs. Second, a prolonged bear market cut the budgets and launch schedules of the new chains and protocols that were its growth surface. Third, large data consumers now arrange supply directly, with Hyperliquid's data partnership with S&P cited as the model, removing the middleware layer entirely. 4

The company had raised $11.2 million over its life, including a $7.5 million Series A in May 2024 led by Tribe Capital and RockawayX with participation from the Solana Foundation, Aptos and StarkWare. 4 A governance token launched in September 2025 gave the business no protection when the customers went away. 4

One boundary matters here. Switchboard suspended feeds on Aptos, Sui, IOTA and Movement on August 29, 2026 after a suspected key compromise, and Solana feeds were unaffected at the time. The wind-down is a separate decision taken three weeks later, and the reporting separates the two. 4

The diagnosis is mistimed market entry into a layer that was being squeezed from both ends. The number of buyers shrank as chain launches slowed, while the two cheapest responses available to a buyer — build it yourself, or buy direct from the source — both skip the vendor. A middleware vendor can absorb one of those pressures. Both at once leave it with nothing to charge for.

The counter-metric

Price your product against the customer's next-best option rather than against your named competitors:

substitution cost: what it would take for your customer to build the function in-house this year, including the AI-assisted version of that build;
bypass share: the share of revenue held by customers large enough to buy the input directly from your supplier;
maintenance drag: the fixed cost of keeping the service safe and running, which does not fall when demand does; and
migration obligation: what you owe customers if you stop, and how long they would need.

Switchboard's three stated reasons map onto the first two lines exactly: the build got cheaper, and the biggest buyers stopped needing an intermediary. Anyone selling a layer between a customer and a supplier should be able to state both numbers without looking them up.

Three counter-metrics to carry into the next review

Default-option replacement share: the portion of revenue whose honest replacement is a free or already-owned tool, measured separately from revenue you win against named competitors.
Committed financing coverage: the share of the next twelve months of spend that is already closed, with every conditional raise and unpriced round excluded from the numerator.
Substitution cost and bypass share: the customer's cost to build your function themselves this year, and the share of revenue held by customers who can go straight to your supplier.

Three companies, three mechanisms, one shared property: the deciding number was measurable before the announcement — what the customer would use instead, what the money had actually closed, and what the buyer could build alone.