Sentiment doesn't compound.
Runway does.
Relationships are worth more, but your company still has to survive the winter.
This is my first time working with infrastructure. When I started at Chainstack I assumed the competition happened on price and uptime, and found that most of it happens in relationships. That is a comfortable finding until you count how many funded companies it is helping to kill this year. Scroll through the numbers and see where your own decisions land.
scroll down
First, let's measure the cold
Nobody should hand-wave about "the bear market," so here is the thermometer. Two years of all-time highs are gone: Bitcoin trades near $63,000, about half its October 2025 peak of $126,198, and the CoinDesk Q2 review logged a third consecutive losing quarter, the longest streak since 2022.
A price only records what someone was willing to pay that day. Where the money actually went in the second quarter is the harder number to look at, especially for the spot Bitcoin ETFs that were sold as the permanent institutional bid:
Billions of dollars per month · inflow · outflow
Q2 total: −$4.67B, the largest quarterly outflow since the products launched in January 2024. Source: CoinDesk Indices, Q2 2026 Digital Asset Review
Citigroup zeroed its ETF inflow expectations for the next twelve months, cut its Bitcoin target from $112,000 to $82,000, and found something mechanical along the way: every $100 million that exits correlates with a same-day price decline of about 53 basis points, which is enough to make the selling feed itself.
The money didn't get scared. It moved.
Over the same quarter the S&P 500 rallied 14.9% and the Nasdaq 100 gained 27.2%. Risk appetite went somewhere else:
Venture tells the same story from the inside. Eight new crypto funds raised $1.1 billion in Q1 2026, the lowest quarterly count since Q3 2020; annual crypto fundraising is on pace for roughly $4 billion, half of 2025's $8.75 billion; and Paradigm, one of the defining crypto funds of the last cycle, closed a $1.2 billion fund in July aimed at AI and robotics. None of this means crypto is dying; Hashdex's CIO calls it temporary rotation, and he may be right. But you cannot make payroll out of a thesis about when capital returns, and your burn rate keeps running while you wait for it.
The graveyard census
Here is what that environment does to projects. CoinGecko examined every token ever listed on GeckoTerminal: 13.4 million of 25.2 million, 53.2% of everything ever traded, are dead. And the dying was not spread evenly:
Share of all recorded token failures, July 2021 to December 2025
11.6M tokens died in 2025 alone; 7.7M in Q4 after the October 10 liquidation cascade. Source: CoinGecko Research, updated April 17, 2026
Tokens are cheap to abandon, so look at actual companies. RootData counted 99 projects shut down in 2026 by late July, after about 176 across 2025; 62 of this year's dead had raised over $200 million between them. CryptoSlate's tracker recorded 109 by August 5, with the wave peaking in April:
Per month · 109 total through August 5
Jan-Mar shown combined (monthly split not published). Source: CryptoSlate tracker, Aug 6, 2026
Of the 109 recorded shutdowns
Source: CryptoSlate, Aug 6, 2026
Two endings from Q1's closure list are worth holding in mind. Entropy raised $25 million, wound down early, and returned $27 million to investors. Archblock ended with $100 million in liabilities against $10 million in assets. Between those two endings sits every operating decision the founders made, including the boring ones about what they paid for infrastructure.
Nobody ever got fired for buying IBM
Smart teams keep overpaying premium logos out of habit, and the reason has a name and a literature. "Nobody ever got fired for buying IBM" was never an official tagline; it was a fear that circulated through the industry in the 1970s and 80s and did IBM's selling for it. The pitch was that choosing the big name was personally safe for the person choosing. If the project failed on the challenger, you got blamed. On the incumbent, the vendor did.
Behavioral economics gave the reflex a backbone: Samuelson and Zeckhauser's 1988 experiments showed people disproportionately stick with the default option even when alternatives are objectively better, and the pull gets stronger as options multiply. Now try it on yourself:
Two vendors offer comparable service. One's team you love working with; it costs meaningfully more. What do most B2B buyers actually do?
If you want the endgame of choosing by name, the cleanest documented case is concrete and steel, not crypto. Carillion, one of the UK's largest construction firms with roughly 450 government contracts, was the kind of name that made due diligence feel optional. The red flags were public for six months: debt up from £242 million to £1.3 billion, suppliers paid in 43 days, an £845 million write-down, shares down 70%. On January 15, 2018 it collapsed owing around £2 billion to 30,000 suppliers, many of them small raw-material producers who had extended credit to a logo. The suppliers who tracked the numbers and diversified in time got out with something. The name on the letterhead never warned anybody.
Place yourself on the map
Steph Ango's essay "Many ways to win" catalogs the moats that let anything survive competition: accumulation, price, time, uniqueness, defense, accreditation, collaboration, speed. Run infrastructure through that catalog and it wins on an unglamorous few: defense (reliability, predictability), price at scale, and collaboration with the teams building on top. Prestige is a moat too, but notice whom it protects: the vendor. When you pay a premium for a prestigious logo, you fund their moat with your runway.
So I draw infra decisions on two axes. Horizontal: what wins your deals, brand or measured performance. Vertical: what keeps them, transactions or relationships. Before anything measures you, put your self-image on record. Four gut answers, no overthinking:
Answer the four questions to place yourself; drag the dot afterward if it feels wrong. Nothing leaves this page.
When your team picks a critical vendor, what do you believe decides it?
Honestly, how data-driven are your team's buying decisions?
Your vendor relationships today are…
If your main vendor's whole team quit tomorrow, would it hurt?
Answer the four questions above to place your guess, or drag the dot directly.
Hold that guess. At the end, a ten-question quiz will place you with actual answers, and the distance between your guess and your result is a picture of your own bias.
The scenario machine
Here is why "we'll review vendors next quarter" is expensive. RPC pricing models are deliberately hard to compare: some bill flat, one call one unit, while others use method multipliers where a standard EVM call costs around 20 credits, and pay-as-you-go schemes drift upward with your success. Published side-by-side examples anchor the gap: the same 73.5M-call monthly workload costs $199 flat, $745 on one major provider's usage pricing, or $420 on another's.
Pick a scenario, or build your own. The machine extends those published anchor points linearly, so treat it as a model for direction and magnitude, then verify with your own bill.
Modeled from the published comparisons linked above
1 call = 1 unit
pay-as-you-go
~20 credits per call
Source: Chainstack. The flat-rate row is Chainstack's published pricing (3M requests/month free, $49/20M, $199/80M). The other two models are built from Chainstack's published side-by-side comparisons with major providers, kept unnamed here: ~4M delivered requests on a $49 credit plan; $745 and $420 for the same 73.5M workload on usage pricing (averaged to ~$7.90/M). See comparison A, comparison B, comparison C. I work at Chainstack, so read these numbers knowing where I sit: they are the vendor's published figures, not an independent benchmark, and this is my own reading of them rather than the company's. Real bills vary with method mix, and the whole point of this essay is that you should run yours.
Teams that ran this arithmetic for real found real money: Nexo cut debug and trace costs by more than 5x, CertiK cut Ethereum archive costs by over 70%. In a market where 62 funded projects died in six months, months of runway is the whole game.
The quiz: bias or benchmark
Ten questions about how you actually chose, and actually run, your infrastructure. Scored against the research above. Nothing you pick leaves this page.
10 questions · about two minutes · nothing is sent anywhere.
0-100, higher is stronger. Bias resistance inverted: higher means fewer biases steering you.
Ordered by impact on your runway.
Save it, share it, or send it to whoever approves the invoices.
Price, uptime, performance
So here is where I landed, months after that first research doc at Chainstack. Relationships really are worth more than features, and the CEB numbers prove buyers feel it. In a calmer year I would stop there, but capital is rotating out, more than a hundred of your peers have posted farewell threads in 2026, and funded teams are deciding right now which invoices survive the next board meeting. When the market gets this cold, the ranking turns brutally simple: price, uptime, performance. Those three decide whether you are alive to enjoy any relationship at all.
The DevRel who helped you at that hackathon in 2022 has earned a place at the top of your benchmark list, and nothing more than that. If the benchmark says another provider gives you the same uptime for 60% less, take the savings. Paras Chopra has a line I think about often: you have one life; mindfully choose the important problems you can solve. Every dollar and every hour you leave inside an unexamined invoice is attention taken from the only problems that will get you through the winter.
Pull your last three infrastructure invoices and compute your true cost per million requests. Run your real method mix through two other providers' pricing. Write your decision matrix before the renewal call, and until spring, let price, uptime and performance carry the heaviest weights.
Or start with the side-by-side maths: comparison A · comparison B · comparison C
Sentiment doesn't compound. Runway does.
An interactive essay by danimim on choosing blockchain infrastructure with data instead of biases, in the middle of the 2026 crypto winter.
I work at Chainstack, but nothing here is the company talking. This is my own view, from the first time I have ever worked with infrastructure, and what the numbers did to what I expected going in.
Every figure on this page is built from the sources listed below. The quiz and the scenario machine run entirely in your browser: nothing you type or click is sent anywhere.
Icons are original Windows 95 and Windows 98 system icons, collected by old windows icons.
