The road not taken: the deals companies turned down, and the ones they took

The billion dollars Intel kept

In 2017 Intel was offered 15% of OpenAI for $1 billion in cash, and another 15% if it agreed to build chips for the lab at cost. It said no.

There were two reasons, as later reported. The person deciding did not think generative AI models would reach market soon. And Intel's own data-center unit did not want to make products at cost. Both reasons made sense on the day. On the day, saying no cost nothing.

By 2024 the bill had arrived. Intel's whole data-center business, AI chips included, was projected at $13.89 billion for the year; analysts expected Nvidia's data-center revenue alone to reach $105.9 billion. That year Intel's stock fell more than half, it was removed from the Dow Jones Industrial Average and replaced by Nvidia, and its chief executive was forced out in December.

Saying yes has its own way of compounding. In 1993 Warren Buffett paid for Dexter Shoe with 25,203 Berkshire shares, then worth $433 million. In 2007 he called it "the worst deal that I've made". By 2024 those same shares were valued at approximately $14.4 billion.

That is the decision in one pair of stories. A company gets a chance to buy, sell, build or enter, and has to decide while the price of the other road is still invisible. Whichever road it takes, the other one keeps moving. 5 of the decisions in this guide were later taken back. The rest of this guide is about how to see the road you are not taking.

What the road not taken is

Most strategic choices are judged by their results. This one is judged twice: once by what the company did, and once by what the other road turned out to be worth. Intel's no is measured against Nvidia. Buffett's yes is measured against what the shares he gave away became. That second measurement is the one nobody keeps.

Why it is so hard

On paper this is a straightforward comparison of two options. In a real company, three things make it harder.

The ways companies did it

The decisions on record take four shapes. Each is a chance a company could take or let pass.

What 49 companies did

Here is every decision on record, placed by the year it was made and marked by how it turned out. Hover a mark, or open the table, for the company and the outcome.

Of the 47 decisions that can be judged, 36 were never undone, 5 were taken back and 6 were kept only in part.

Never undone is not the same as right. Intel's no is on the never-undone list. So is Bayer's decision to close its roughly $63 billion purchase of Monsanto in 2018, after which its litigation provisions rose to 11.8 billion euros by February 2026.

The walk-backs

Put the decisions that were taken back, in whole or in part, side by side and two things stand out. First, all but one of them was a yes: a purchase, an entry, a launch, a sale. The exception is Apple's refusal to license the Mac, and Apple half-reversed that itself when it licensed clones a decade later. Second, the largest group bought a business the company did not know how to run.

The largest group bought a business it had no way to run: eBay with a phone service, Verizon with a media company, Goldman Sachs with consumer loans, Walgreens with doctors' offices. Each had a story about how the new business would feed the old one. eBay's was that buyers and sellers would want to talk before they traded. None of the stories survived contact with the new business.

The ones kept in part mostly kept what they had wanted and gave up the rest. Google kept Motorola's patents and sold the phones. Microsoft got Activision after handing its cloud streaming rights to Ubisoft. AMD stayed free of its factories, and later paid to be free of the supply contract too.

The refusals barely appear on this list, because there is little to undo. Their cost shows up elsewhere: in Nvidia's results instead of Intel's, or in what Snap was worth on the day it went public and on the day it fell back below that.

A yes can be taken back. A no just keeps its price.

What the share price did

For many of the decisions there is a clean market record: the company's total return over the three years after the decision, against the S&P 500.

Some of the best share-price records here belong to decisions that were taken back. Nvidia's agreement to buy Arm was abandoned, and over the next three years its shares beat the market by 234.5 points. Amazon's Fire Phone failed, and Amazon beat the market by 157.6. Lyft's sale of its self-driving unit stood, and Lyft trailed by 108.1. A company can be far bigger than its mistake, and the chart measures the company.

How to run this decision

If you are weighing a chance like this, here is what the record suggests doing, in order.

1. Write down the option you turn down, with the date, the price and the reason

   One page: what was offered, what it would have cost, why it was declined and who decided. Intel's reasons became public only years later, in press reports. A refusal nobody wrote down cannot be reviewed, only re-argued.

   Seen in: Intel, 2017
2. Price the road not taken again a year later

   Ask what the rejected option would cost now and what it would have returned. Buffett did this in public, restating Dexter's cost in later letters. Snap's refusal looked brilliant at its IPO and less so when its value fell back below that level. A second look is cheap.

   Seen in: Berkshire Hathaway, 1993 · Snap, 2013
3. Read the terms, not just the number

   IBM's DOS license had no price that mattered. What mattered was that it was non-exclusive. AMD sold its factories to survive and later paid $703 million to get out of the supply contract that came with the sale. Exclusivity, duration and exit terms are where the long-term value sits.

   Seen in: IBM, 1980 · AMD, 2008
4. Notice what currency you are paying in

   Paying cash for a bad asset caps the damage at the cash. Paying in your own rising stock does not: Buffett's $433 million became about $14.4 billion. If you expect your shares to be worth more, they are the most expensive way to pay.

   Seen in: Berkshire Hathaway, 1993
5. Ask whether you can run what you are buying

   eBay, Verizon, Goldman Sachs and Walgreens all bought into businesses that worked differently from their own, and all later sold, shut or cut back what they bought. Name who will run it, and what they have run before, before you name the price.

   Seen in: eBay, 2005 · Verizon, 2015 · Walgreens, 2020
6. List who else has to say yes

   Regulators ended Nvidia's purchase of Arm, a court ended Kroger's of Albertsons, and a state attorney general stopped the sale of Hershey. Chevron, by contrast, structured its Hess deal around the rival who could block it, and an arbitration panel upheld that structure.

   Seen in: Nvidia, 2020 · Kroger, 2022 · Chevron, 2023
7. Check whether the no is protecting one unit's margin

   Intel's data-center unit would not build chips at cost. Apple's Mac licensing died to protect margins above 55%. Sony locked down its players to protect its record label. When the person saying no owns the margin at risk, get a second opinion.

   Seen in: Intel, 2017 · Apple, 1985 · Sony, 2004

Warning signs

What it has looked like when this decision was going wrong.

• The deal is turned down because one division doesn't want to change how it prices.Intel, 2017
• The decision has no number in it, so nobody senior reads the terms.IBM, 1980
• You are paying in your own stock, and you expect it to rise.Berkshire Hathaway, 1993
• The product is designed to feed your business rather than to serve the customer.Amazon, 2014
• The deal depends on a buyer for divested stores that runs a fraction as many.Kroger, 2024

All 49 cases

Each case is one company at one decision: what was on the table, what it chose, why, and what followed, with every fact sourced.

Questions

What is a counterfactual decision in strategy?

A real chance to buy, sell, build or enter something that a company has to take or turn down. It is called a counterfactual because, afterwards, the option it rejected can be priced against what happened instead.

What are famous examples of companies that passed on a deal?

Intel declining a stake in OpenAI in 2017, Snap refusing Facebook's offer in 2013, AstraZeneca rejecting Pfizer in 2014, and Apple declining to license the Mac in 1985. Tesla came close to selling to Google in 2013 before a profitable quarter ended the talks.

Is it riskier to say yes or no to a big deal?

On record, every decision that was later reversed was a yes: a purchase, an entry or a launch. A no is almost never undone, but its cost can be just as large; it simply shows up in someone else's results.

How should a company evaluate an opportunity it is turning down?

Write down what was offered, the price, the reason and who decided; read the terms as closely as the number; note what currency you would pay in; and price the rejected option again a year later.