Raspberry Pi has not announced a Pi 6. Full stop. What it has done, loudly enough that LinkedIn and the Cambridge tech grapevine noticed, is keep posting ASIC roles, IC design, DFT, tapeout, for a silicon team that looks like it is building something more ambitious than another RP2040 cousin. Meanwhile the public timeline, from Eben Upton in a Reddit AMA on May 21, 2026, is blunt: major platform releases run on a roughly 4 to 4.5 year cadence, so do not expect Pi 6 before early 2028. That kills the late-2026 and 2027 rumor mill that had been circulating since Pi 5 started feeling "due" for a successor. Hiring and shipping dates are not the same calendar. Both can be true at once.
The AMA killed the 2026 rumor, not the silicon work
Upton sat the AMA with James Adams and Gordon Hollingworth. Circuit Digest, OMG! Ubuntu, It's FOSS, and Jeff Geerling's roundup all carried the same core answer: historical cadence puts Pi 6 no earlier than early 2028, and the Pi 5 still has legs as a flagship. I went back and read the secondary coverage next to each other because AMA threads get paraphrased into certainty they never earned. The consistent line across those write-ups is cadence plus economics, not a secret canceled project. Geerling's summary in particular is useful because he sits closer to the maker and industrial audience than a generic launch rewrite does, less vibes, more "here is what they actually typed."
On the product shape, Upton's framing was almost aggressively boring, in a good way. Expect a familiar form factor and feature set with "more": faster CPUs, more I/O headroom, more DRAM bandwidth. Quantitative, not qualitative. No onboard NPU in the picture he painted. Asked about edge AI beyond camera inference, he stuck to the company's long habit of betting on CPUs getting faster and algorithms getting cheaper rather than soldering a fixed-function accelerator into every board. HATs and external accelerators remain the escape hatch for people who need one. If you were waiting for a Pi that ships looking like a Jetson competitor, that is not the story they told.
That is the public story. The private-looking story is the job board. Expanding an ASIC organization in Cambridge, design, test, bring-up, is not something you do for a brochure refresh. Silicon programs eat years. A 2028 ship date for a board that needs its own SoC is not "we forgot". It is "we started and the calendar is honest." I treat the hiring as evidence that work continues, not as a leak that Upton was sandbagging the AMA. Companies hire ahead of product pages all the time. They also hire and cancel. So the signal is directional, not a SKU. A requisition for DFT engineers is not a pinout. It is still a stronger signal than a reseller restock screenshot with Pi 6 painted on in MS Paint.
I keep a soft file of Cambridge listings when they appear, the way some people keep FCC filings. Not because a job description is a datasheet, it isn't, but because the verbs in those posts (tapeout, bring-up, silicon validation) describe a stage of work that only makes sense if die are in flight or about to be. You can staff a research fantasy for a while. You cannot staff tapeout theater forever without something to tape out. That is the thin edge of optimism I allow myself between now and 2028.
DRAM prices are the boring villain
If you want a single reason the roadmap feels stuck, look at memory. Coverage around the AMA and Raspberry Pi's own earlier pricing notes pointed at LPDDR4 costs that had roughly seven-folded over the prior year. That is the kind of input-cost shock that makes a $35–$80 board strategy look like a math problem you cannot solve politely. Raspberry Pi already pushed price increases. Anyone still quoting 2023 MSRPs as if they were physics is arguing with a spreadsheet that left the building. The company has been unusually blunt about memory economics in public. Believing them is easier than inventing a conspiracy about a warehouse full of unreleased boards.
The Zero line makes the constraint vivid. Older Zero boards cling to LPDDR2, which insulated them from the LPDDR4 spike. A Zero 3 that needs newer DRAM would, in Upton's phrasing as relayed by Circuit Digest and others, land at an "un-Zero-like" price. So the company waits. Flagship cadence and entry-level cadence both bend around the same commodity market. That is less dramatic than a hidden Pi 6 prototype, and it matches how component-constrained hardware companies actually behave when the BOM goes hostile.
Industrial buyers already treat Pi availability as a supply-chain topic, not a hobby shop mood. They will care less about whether Reddit wanted a 2027 launch and more about whether 2028 silicon is pin-compatible enough to keep carrier boards alive. Raspberry Pi's habit of long platform life is the quiet product feature here. A delayed Pi 6 is annoying for YouTube unboxers. For a factory floor that qualified Pi 5 last year, "still shipping" is the feature. I have sat through enough redesign meetings triggered by a discontinued module to prefer boring longevity over a flashy SKU that orphans last year's carrier.
Price raises already happened. They are not a hypothetical. That matters for schools and for anyone quoting a multi-year kit budget based on a blog post from the Pi 5 launch window. The educational mission and the public-company BOM are in tension when DRAM traders set the weather. Upton's AMA read to me like someone choosing not to pretend otherwise. Ship the board you can price. Staff the silicon you will need when pricing thaws. Decline to invent a 2026 keynote because the internet is restless.
Read the hiring without inventing a keynote
There is still no official Pi 6 announcement. No SKU. No FCC filing I am hanging a launch date on. No keynote slide. If someone is selling you a pinout diagram sourced from a Telegram channel, close the tab. What we have is a CEO on the record about early 2028 at the earliest, a product philosophy that sounds like Pi 5-plus rather than a reinvention, an explicit coolness toward onboard NPUs, and a company that continues to staff silicon roles in Cambridge while DRAM markets misbehave. Secondary coverage from Circuit Digest through Geerling-style roundups is how most of us met those AMA lines. Go read the thread if you want the raw phrasing.
I have watched this community invent launch seasons out of reseller restocks before. The healthier habit is to separate three clocks: rumor clocks, hiring clocks, and ship clocks. Right now the rumor clock said 2026/2027 and got corrected. The hiring clock says silicon work is real enough to recruit for. The ship clock, publicly, says early 2028 or later. Those three can disagree without anyone lying. Most online arguments collapse because people insist the three clocks must tick in unison.
If you are buying hardware this year, buy for the board that exists. If you are designing a carrier for a next-gen Compute Module you have not seen, leave slack in the schedule and in the BOM for memory pricing that has not apologized yet. And if you are just waiting for a faster Pi because your current one swaps, the AMA's unromantic message is that CPU-and-bandwidth gains are coming on the company's cadence, not on the internet's impatience. Swap space is a software problem you can mitigate today. A nonexistent SoC is not.
Believe the later date until a product page says otherwise. Keep an eye on Cambridge listings if you enjoy reading tea leaves. Just do not confuse a requisition with a release party. Raspberry Pi told the second story out loud in May: work continues, boards arrive later, memory is the tax. August job posts do not overturn May sentences. They decorate them.
- Open Source
