Key takeaways
- A 15% discount means paying 85 for something worth 100. That's a 17.6% gain on the cash you hand over, not 15%, because a discount is quoted off the price you don't pay.
- Payroll contributions buy nothing until the purchase date, so your capital is committed for about half the period. On a 6-month purchase period that's 3 months, which turns 17.6% into roughly 70.6% a year.
- Microsoft's smaller 10% discount over 3-month periods annualises higher, at 88.9%, because the money is tied up for half as long. Period length matters as much as the discount.
- NVIDIA valued one ESPP right at $20.75 at grant in its 2026 financial year, against a $49.13 average purchase price. A same-day 15% discount is worth 17.6% of the price paid.
- For a UK higher-rate participant in a US plan, HMRC taxes the entire spread as pay: 40% tax plus 2% National Insurance cuts 70.6% to 40.9%.
A 15% discount is a 17.6% gain, and the money is committed for half the period
Your employer offers shares at a discount, the paperwork says 15% off, and you want to know what that's actually worth. It's worth more than 15%, for two reasons that have nothing to do with the share price.
Start with the discount itself. Paying 85 for something worth 100 is a gain of 15 on an outlay of 85. That's 17.6%. A discount is quoted off the price you don't pay. Your return is earned on the money you do.
Now the timing. Contributions come out of each payslip across the whole period, and they buy nothing until the purchase date at the end. The first deduction is tied up for the full period. The last is tied up for days. On average your money is committed for about half the period, which is roughly 3 months of a 6-month purchase period.
A 17.6% gain on capital committed for 3 months works out at about 70.6% a year. That figure is real. It's also the most misleading number in this piece, which is why most of what follows is about where it stops being useful.
The averaging is the whole calculation, and it isn't a compounding claim
Here's the method, because the method is the argument.
Say you contribute a steady slice of every payslip. Over a 6-month purchase period you put in half your annual contribution, and your average balance across those months is half of that again, so a quarter of the yearly total.
At the purchase date you buy at 85% of market value and pick up 17.6% on the cash. You do that twice a year. So you earn 17.6% of half your annual contribution, twice over, on an average balance of a quarter of it. Multiply 17.6% by 4 and you get 70.6% a year on the capital employed.
That's a rate, not a pot of money. Nothing compounds here, because the plan caps what you can put in and the gain arrives as stock rather than as contribution room.
Two refinements sharpen it. Six equal monthly deductions land at the end of each month, so the true average commitment is 2.5 months rather than 3, and the same plan then annualises at 84.7%. Using half the period is the conservative version, and it's the one used throughout this piece.
The second refinement is that nothing happens to the money while it waits. Apple's plan, as amended in November 2024, is explicit: "No interest shall be paid on sums withheld from a Participant's pay for the purchase of Shares under this Plan." NVIDIA's 2012 plan says contributions "will be deposited with the general funds of the Company". You're making an interest-free loan to your employer, repaid in discounted stock.
Microsoft's 10% discount earns more per year than a 15% one, because the wait is shorter
Period length is the variable nobody quotes, and it does as much work as the discount.
Microsoft's annual report for the year to 30 June 2026 describes its plan in one line. Shares "may be purchased by employees at three-month intervals at 90% of the fair market value on the last trading day of each three-month period", capped at 15% of gross pay. No lookback, a smaller discount, a much shorter wait.
Run the same arithmetic. Paying 90 for 100 is 11.1% on the cash. The capital is committed for about half of 3 months. That annualises at 88.9%, ahead of the 70.6% from a 15% discount over 6 months. The chart plots both, alongside a 12-month version of the same 15% discount, which falls to 35.3%.
Microsoft employees bought about 5 million shares that year at an average price of $382.92. If that average sits at 90% of the average market value, the implied market price was $425.47 and the discount was worth $42.55 a share. Across a $1.9 billion outlay, that's roughly $213 million of value transferred, in a plan with no lookback at all.
The lookback is the feature doing most of the work, and most explainers skip it
A plain discount pays you 17.6% whatever the share price does. A lookback changes the shape of the thing entirely.
NVIDIA's plan document sets the price at "not less than the lesser of: (i) an amount equal to 85% of such Participant's Offering Date Price; or (ii) an amount equal to 85% of the Fair Market Value of a share of Common Stock on the applicable Purchase Date". Its annual report supplies the shape: "Each offering period is about 24 months, divided into four purchase periods of six months."
So the price is struck off whichever of two dates was cheaper, and those dates can be 24 months apart. If the shares rise, you pay 85% of the old, lower price. If they fall, you pay 85% of the new, lower one. The lookback never costs you anything. It's a free call option on your employer's shares, and almost nobody explaining the plan prices it.
Some plans even re-strike it downwards. NVIDIA's document lets the board structure an offering so that if the price at the start of a new purchase period "is less than or equal to a Participant's Offering Date Price, then ... that Offering will terminate immediately as of that first Trading Day and such Participant will be automatically enrolled in a new Offering beginning on that first Trading Day."
Why 24 months rather than 5 years? The US tax code caps it. An option priced off the grant date has to be exercised within "27 months from the date such option is granted". Apple's plan repeats the limit in its own definitions, then runs two offerings a year of about six months each. Designers sit just under a statutory ceiling, which is why 24 months keeps showing up.
How much is the lookback worth? NVIDIA's accountants have to answer that every year. They value the rights with the Black-Scholes model, using a "Weighted average expected life (in years) 0.1 - 2.0" and volatility of "26 %- 96 %". For the year to January 2026 the weighted-average grant-date fair value came to $20.75 a share, against a weighted-average purchase price of $49.13 for shares actually bought that year. The fair value is 42.2% of the price paid, about 2.4 times the 17.6% that a same-day discount is worth.
That comparison mixes two cohorts, rights granted during the year and shares purchased during the year, so it isn't a clean like-for-like. The gap is still far too wide to be the discount on its own. An option model that needs an expected life of up to 2.0 years and a volatility input is not valuing a coupon. It's valuing an option.
The price NVIDIA employees paid on average went from $17.74 to $49.13 across those two years, up 177%. In a market moving like that, the difference between paying 85% of today and 85% of two years ago is the entire proposition.
The US tax form carries the fingerprint. Form 3922 has a box for the "Exercise Price Per Share Determined as if the Option Was Exercised on the Date Shown in Box 1", which is the grant date. The IRS asks for the lookback price because the tax bill depends on it.
US tax: the discount is pay, and only the grant-date slice reaches the capital gains rate
What follows is US federal tax on a section 423 plan, as it stood for the 2025 tax year.
Nothing is taxed when you enrol or when you buy. The bill arrives when you sell, and which bill depends on how long you waited.
Section 423 asks for two waits at once: "no disposition of such share is made by him within 2 years after the date of the granting of the option nor within 1 year after the transfer of such share to him". Clear both and you have a qualifying disposition. Miss either and you don't.
On a qualifying disposition the ordinary income is capped at the discount measured on the grant date. IRS Publication 525 for 2025 works an example through. An option at $20 when the stock was worth $22, exercised at $23, sold at $30: "you must report as wages the difference between the option price ($20) and the value at the time the option was granted ($22). The rest of your gain ($8 per share) is capital gain."
Sell early and the whole spread at purchase becomes wages instead. On the same facts, sold 6 months after exercise, "you must report $300 as wages and $700 as capital gain" across 100 shares. Waiting moved $1 a share out of ordinary income and into long-term capital gain, which for tax years beginning in 2025 is taxed at "no higher than 15% for most individuals", or 20% at the top. On those numbers the prize for two years of waiting is the rate difference on $1 a share.
There's also a ceiling on the whole exercise. No employee may accrue rights "at a rate which exceeds $25,000 of fair market value of such stock (determined at the time such option is granted) for each calendar year". At 85% of that, the most you can spend is $21,250 and the most discount you can capture is $3,750 a year. The limit is measured at grant-date value, so a lookback that pins you to an old, low price lets the same $25,000 of allowance buy more shares.
One oddity: the compensation element of a qualifying disposition carries no withholding. The statute says "No amount shall be required to be deducted and withheld under chapter 24". The tax is still due, and nobody takes it off you at source.
UK tax: HMRC charges the entire spread as employment income on the purchase date
This is where UK employees of US companies get caught. The rates below are for the 2026 to 2027 UK tax year.
A US section 423 plan is not one of the UK's tax-advantaged schemes, and our guide to EMI, SAYE and SIP taxed at every stage sets out what a statutory scheme buys that an overseas purchase plan doesn't. What you have instead is an employment-related securities option, taxed under Chapter 5 of Part 7 of the Income Tax (Earnings and Pensions) Act 2003. HMRC's manual sets out the charge: the gain on acquiring shares under the option is "MV - C where MV is the market value of the securities at the time of acquisition, and C is the amount of any consideration given for the securities".
Read that against the lookback. The UK charge is the full difference between market value and what you paid, measured on the purchase date. The discount and every penny of lookback gain are employment income together. There's no qualifying disposition to wait for and no cap at the grant-date discount, so the feature that makes the plan valuable is also the feature that raises the UK tax bill.
Because listed shares are readily convertible assets, PAYE applies. HMRC's test is whether "the securities are quoted or subject to current or future trading arrangement. If YES, then PAYE is due." Class 1 National Insurance follows the same amount. That's the same machinery that runs on an RSU vest, where what comes off at source and what you finally owe are rarely the same figure, as our guide to what an RSU vest actually costs works through.
For 2026 to 2027 the bands run 20% basic, 40% higher above £50,270 and 45% additional above £125,140. Employee National Insurance is 8% on weekly earnings between £242.01 and £967, and 2% above that. A higher-rate participant is paying 42% at the margin on the discount, which turns 70.6% into 40.9%. The $3,750 an American can capture at the section 423 ceiling would be worth $2,175 after the same 42%, and UK employees are often in a separate sub-plan where the section 423 limits don't apply at all.
One thing does go right. The amount charged to income tax is added to the capital gains base cost. HMRC's capital gains manual counts "any amount counting as income on exercise of the option under section 476 Income Tax (Earnings and Pensions) Act ... 2003" as part of the cost of the shares. Selling on the purchase date leaves almost no chargeable gain. Keep the shares and buy more later, and the share matching rules and the section 104 pool start to matter. The rates and allowances that apply from there are in our 2026 to 2027 tax table.
The strongest objection: a spectacular rate on a balance you can't grow
The serious case against the 70.6% is that a rate of return means very little when the capital behind it is capped, illiquid and undiversified.
Start with the cap. Section 423 stops the accrual at $25,000 of grant-date value a year, which is $3,750 of discount. Plan rules bite before that for most people: Apple allows deductions "between one percent (1%) and ten percent (10%)" of eligible compensation, Microsoft caps purchases at 15% of gross pay, NVIDIA allows "up to 25% of their earnings". On Apple's cap the average balance sitting with the employer is 2.5% of a year's salary. On NVIDIA's it's 6.25%. A 70.6% rate on 2.5% of salary is a smaller thing than it sounds.
Then the risk. The discount is a cushion of 15% of the market price, and no more. If the shares fall further than that between the purchase date and the day you sell, the discount has gone and you're holding a loss. The 70.6% is a return on a position, not a payment.
The US tax break makes that worse before it makes it better. A qualifying disposition needs 2 years from grant and 1 year from transfer, which is two years of concentrated exposure to a single stock in exchange for a rate difference on part of one year's discount. Whether that trade is worth taking turns on how large the holding is next to everything else you own, which is the subject of our guide to being concentrated in employer stock.
And the rate can't be redeployed. You can't take 70.6% and run it again on a bigger balance, because the balance is set by your salary and the plan rules. What the figure genuinely does is rank the plan against the alternatives for the same money: a savings account, a pension contribution, paying down debt. On that comparison it usually wins by a wide margin, which is a more modest claim than the headline number implies.
What this calculation can't tell you
Three plans is a sample of three. Apple, Microsoft and NVIDIA are large US technology employers with liquid shares, and their terms differ from each other even so. A plan with a 5% discount, a 12-month period and no lookback is a different instrument wearing the same name, and the same arithmetic on a 12-month period gives 35.3% rather than 70.6%.
The fair-value comparison has a real limitation. NVIDIA's $20.75 covers rights granted during its 2026 financial year and the $49.13 covers shares purchased during it, so the ratio is indicative rather than exact. The implied Microsoft market price of $425.47 assumes the weighted-average purchase price sits exactly at 90% of the weighted-average market value, which holds only approximately.
The tax figures are jurisdiction-specific and dated. The US material is federal tax for the 2025 tax year and ignores state tax and the net investment income tax. The UK material is for 2026 to 2027 and uses the rates that apply in England, Wales and Northern Ireland, not the Scottish bands.
And there's no data here on what participants actually do. Whether people sell at purchase or hold for the tax treatment, and how that has worked out for them, is not something these filings record.
What would change the conclusion
If the 27-month statutory limit moved, offering periods would lengthen and the lookback would be worth more, because a longer free option on a volatile share is worth more than a short one.
If plans paid interest on accumulated contributions, the committed-capital penalty would shrink and the annualised figure would fall towards the headline discount. Both plan documents quoted here go out of their way to say they don't.
If you can't sell at the purchase date, the whole framing changes. Insider dealing windows, minimum holding rules and an employer's own share-ownership guidelines all convert a short-duration discount into a long-duration bet on one company's shares.
The three lines that decide all of this sit in your own plan document: the discount, the length of the purchase period, and whether the price is struck off the lower of two dates. LedgerTouch tracks the position once the shares land in an account. What the position cost you, in money and in time, is written in the plan.