At 6:40 p.m. on a Wednesday in January I drank sixteen ounces of cold brew — 205 mg of caffeine, per the chain's own published figure — because a livestream started at eight and I wanted to be awake for the wearables segment. I was. I also did not fall asleep until 12:51 a.m., against a fourteen-minute personal baseline. I have run the same measurement against every hardware keynote since, which brings us to the Google Pixel 11 launch and the only question this piece exists to answer: what should you drink, and when, so that a Made by Google stream costs you an hour of your evening instead of two hours of your night?
The verdict, before anything else: one eight-ounce cup of ordinary drip coffee — roughly 95 mg — taken forty-five minutes before the stream starts beat both a 205 mg cold brew and a decaf on every measure that mattered, including how much of the keynote I could recall the next morning. The large dose bought alertness I did not need and charged me forty-nine extra minutes of sleep onset for it.
That is the whole recommendation. The rest of this is the timing, the arithmetic, and the part where I admit what I could not measure.
When the stream starts, and why the hour is the whole problem
Google has anchored its hardware keynote to a 10 a.m. Pacific slot for years running, and the livestream goes up on the Made by Google YouTube channel a day or two ahead with a countdown on it. Confirm the exact slot on that listing before you plan anything around it — a sixty-minute slip is not unusual, and it is the difference between a manageable evening and a bad one.
Assuming the usual 10 a.m. Pacific start, that lands at:
- 10:00 a.m. — Pacific
- 1:00 p.m. — Eastern
- 6:00 p.m. — London
- 7:00 p.m. — Berlin, Paris, Madrid
- 10:30 p.m. — India
- 1:00 a.m., next day — Singapore
- 2:00 a.m., next day — Tokyo
- 3:00 a.m., next day — Sydney
If you are in North America, close this tab. A 1 p.m. Eastern keynote is a lunch break, the caffeine question does not apply to you, and nothing below will improve your day. This piece is written for the second half of that list.
One more piece of scheduling arithmetic that most event guides leave out: the keynote itself is short. The last several Made by Google streams have run between fifty-five and seventy-five minutes. The thing that actually eats your night is what comes after — the hands-on posts landing the moment the embargo lifts, the liveblog you scroll to the bottom of, the spec sheet you open in a fourth tab to check whether the telephoto is the same sensor as last year. In my logs, the stream reliably took about an hour and the aftermath took ninety minutes more. Budget for the aftermath or it will budget for you.
What I tested, and what I couldn't
Three livestreamed hardware events between January and July, each starting at 8:00 p.m. local for me, each about an hour long. Same room, same 19°C, same 6:30 a.m. alarm the following morning, same fixed 200 mg of coffee at 7:15 a.m. on every one of those days so that my daytime intake wasn't a hidden variable. One protocol per event.
Sleep onset came from a wrist wearable, cross-checked against the timestamp on my phone charger and a bedside notebook. Consumer wearables land within roughly ten to fifteen minutes of lab polysomnography for sleep onset in validation work, which makes them adequate for detecting a fifty-minute effect and worthless for detecting a five-minute one. I am reporting the fifty-minute effects and ignoring anything smaller.
Things I could not control, in descending order of how much they bother me. The sample is one person, three nights — my palate and my liver, not yours. Nothing was blinded; you cannot blind a person to sixteen ounces of cold brew, and on the decaf night I knew perfectly well it was decaf. The events were different, so a genuinely dull keynote and a genuinely interesting one are baked into my alertness ratings. And I am not caffeine-naive — 200 mg every morning for years — so a person who drinks one cup a week will see larger effects than I did, in both directions.
The three protocols, compared
| Protocol | Dose and timing | Alertness at minute 55 (self-rated /10) | Sleep onset | Total sleep | Spec recall next morning |
|---|---|---|---|---|---|
| A — Full send | 205 mg cold brew, 80 min before | 9 | 71 min | 5 h 12 m | 14/22 |
| B — One cup | 95 mg drip, 45 min before | 7 | 22 min | 6 h 58 m | 16/22 |
| C — Decaf | ~6 mg, 45 min before | 5 | 11 min | 7 h 31 m | 9/22 |
| Baseline (no event) | none after noon | — | 14 min | 7 h 20 m | — |
The recall column is the one I find most interesting, so here is how it worked. Twelve hours after each stream, before opening any coverage, I wrote down every hardware claim I could remember unprompted — chip name, display brightness figure, battery capacity, price, colourway, anything specific. I then scored that list against the official transcript, twenty-two checkable claims per event, one point each, no partial credit for "something about the camera."
Protocol A, the biggest dose, scored 14. Protocol B, less than half the caffeine, scored 16. I want to be careful here: two points on a twenty-two-point scale across a sample of one night each is noise, not a finding. What it is not is evidence for the thing I assumed going in — that more caffeine would mean better retention of a dense, fast-moving spec dump. It didn't. On the cold brew night I was unmistakably more awake, in the sense of alert, tense, and typing quickly. I was not better at holding numbers.
One eight-ounce cup, forty-five minutes before the stream, is the entire strategy. Everything above that buys you alertness you don't need and charges an hour of sleep for it.
The science under the result
Caffeine does not supply anything. It blocks. Adenosine accumulates in the brain across every waking hour and binds to A1 and A2A receptors, and the subjective feeling of that binding is sleepiness. Caffeine is shaped closely enough to adenosine to occupy those receptors without activating them. The adenosine is still there, still accumulating, still unbound and waiting. When the caffeine clears, it binds all at once. That is the crash, and it is why a large dose does not feel like extra energy so much as borrowed silence.
The timings that matter for planning a keynote are two. Peak plasma concentration arrives roughly thirty to sixty minutes after you drink — that is where the forty-five-minute pre-roll comes from, and why drinking as the stream starts leaves you peaking during the closing sizzle reel. And the half-life in a healthy adult runs about five hours, with a real-world range from roughly two to eight. Oral contraceptives can roughly double it. Smoking cuts it by about half. Variation in the CYP1A2 gene, which handles the majority of caffeine clearance, sorts people into fast and slow metabolizers with genuinely different evenings.
The most useful published work here is Drake and colleagues, in the Journal of Clinical Sleep Medicine in 2013. They gave 400 mg of caffeine at bedtime, three hours before bedtime, and six hours before bedtime. All three conditions disrupted sleep relative to placebo. The six-hour dose — the one everybody assumes is safely clear — reduced objectively measured sleep time by more than an hour. The participants' own self-reports did not reliably detect it. That last detail is the one that reframes the whole exercise: the sleep was gone whether or not the sleeper noticed.
And sleep onset is only the visible casualty. Work going back to Landolt's caffeine and sleep-EEG studies in the mid-nineties found that caffeine taken in the morning still measurably altered the spectral profile of that night's sleep — less slow-wave activity, the deep stuff. You can fall asleep on schedule and still have had a worse night. My wearable cannot see that, which is precisely why I am citing someone whose equipment could.
Why the "nothing after 6 p.m." rule fails a European viewer
Run the numbers longhand for a London reader on a 6:00 p.m. stream. You drink 205 mg at 5:15 p.m., safely before the folk-wisdom cutoff. One half-life later, 10:15 p.m., about 100 mg is still circulating — roughly a full cup of coffee, poured directly into the hour you were planning to wind down in. At 3:15 a.m. you are still carrying about 50 mg, which is a strong cup of black tea, administered in the middle of the night.
The cutoff rule fails because it is a rule about clock time and caffeine does not respond to clock time. It responds to dose and to your particular metabolism. A 60 mg dose at 8 p.m. leaves you less loaded at midnight than a 200 mg dose at 4 p.m. If you only change one variable for launch night, change the size of the cup, not the hour on it.
There is also a ceiling worth knowing about. In the vigilance literature, most of caffeine's benefit to sustained attention shows up in the first 100 mg or so, and the dose-response curve flattens well before the 300–400 mg mark. Past the flat part, you are mostly buying jitter, a faster heart rate, and the sleep debt. My own logs agree: a 9/10 on the cold brew versus a 7/10 on the single cup, for an extra 110 mg and forty-nine minutes of lying awake. That is a bad exchange rate for two points on a scale I invented.
Decaf, reviewed on its own terms
Protocol C is where a magazine about coffee should be honest twice over.
As a drink, decaf did fine. The one I used for the July stream was a Swiss Water-processed Ethiopian natural, and it held its fruit — a stewed-strawberry, slightly jammy thing that survived the whole cup and was still legible at room temperature twenty minutes in. A sugarcane-EA Colombian I used on an earlier night did not fare as well: bright and cherryish for the first third, then collapsing into generic cocoa by the halfway point, the way a lot of decafs do once the temperature drops. If you want the ritual of a cup in your hand while a man on stage says "on-device," a good decaf delivers the ritual completely.
What it does not deliver is the alertness, and I should not pretend otherwise. I rated myself 5/10 at minute fifty-five and I missed a segment — genuinely missed it, learned about it the next morning from someone else's liveblog. My recall score was 9/22, the worst of the three by a wide margin, and unlike the two-point gap between A and B, a seven-point gap is a real difference.
One technical note people get wrong: decaf is not zero. Eight ounces typically runs 2 to 15 mg depending on process and roaster, and a large shop decaf can approach a quarter of an espresso. That's irrelevant at 6 p.m. and marginally relevant at 1 a.m. if you're unusually sensitive.
Who this is for, and who it isn't
This is for you if you're watching between roughly 6 p.m. and midnight local — the UK, continental Europe, India — and you intend to write, post, or argue about the announcements the same night. It's for you if you know or suspect you're a slow metabolizer, because you're the person for whom the difference between 95 mg and 205 mg is an entire night. And it's for you if you have watched a keynote before and noticed that you remember the theatre better than the specs.
This isn't for you if you're in North America, where the Pixel 11 keynote lands mid-afternoon and the correct caffeine strategy is whatever you were doing anyway. It isn't for you if the stream lands at 2 or 3 a.m. local — Tokyo, Sydney, Singapore. I tested an 8 p.m. start, not a 3 a.m. one, and nothing in my data extends there. For that slot, the honest answer is that no cup fixes a 3 a.m. wake-up, and the announcements will be identical in the replay at breakfast.
And it isn't for you if you have a hard early obligation the next morning. Watch the replay. The hardware does not change between the live stream and the recording, and the only thing you lose is the ability to post first.
The night of the launch
Last year I did this badly. I had a double espresso around the time the pre-show loop started, then a second one during the Q&A because the first had stopped working, which is exactly the wrong reason to have a second one — that flat feeling is the adenosine, and the second cup only postpones the same conversation. I filed at 11:40 p.m., went to bed at half past midnight feeling perfectly fine, and lay there until something like two. The next morning I reread what I'd filed and found two transposed figures in a spec table.
So, concretely, here is my plan for this one, since a review that doesn't commit isn't a review. Eight ounces of drip, about 95 mg, at 5:15 p.m. for a 6:00 p.m. start. Nothing after that — not during, not for the write-up. When the stream ends, the liveblog and the hands-on posts get forty minutes on a timer, and then the laptop closes whether or not I've got to the bottom of the camera comparisons. I'll log the onset the same way and I'll report back if the number embarrasses me.
The myth is that caffeine keeps you awake for the event and then wears off in time for bed.
The more accurate version: caffeine doesn't wear off, it decays — half of that launch-night cup is still in you when the lights go out, and the sleep you lose is charged to the size of the cup, not to the length of the keynote.
