Mythic, an AI inference processor startup developing analog compute-in-memory tech, raises $30M Series B1, bringing the total raised to $86M
There are two trends converging in AI inference and so far, only a small number of companies are enmeshed. — The first trend takes us …
Context & Ripple Effects
Mythic's $30M Series B1 is an extension of the $40M Series B SoftBank led in 2018 to get its high-speed, low-power AI chip to market — a bridge round that lifts total funding to $86M while the company is still pre-scale on its analog compute-in-memory approach. The framing in the coverage itself is that two trends are converging in AI inference, and few companies are positioned where they meet.
The longer arc matters more than the round size: Mythic went on to raise a $70M Series C in 2021, taking the total to $165.2M, and by late 2025 had closed a $125M round led by DCVC under a new CEO recruited from Nvidia. Read together, the 2019 B1 is the point where an unconventional analog architecture committed to a decade-long, capital-hungry path rather than an early exit.
First-order effects
- Mythic buys extended runway to finish bringing its analog processing unit to market without changing its architecture thesis, with SoftBank's Series B syndicate effectively doubling down via the B1.
- Investors backing the round are accepting that compute-in-memory inference requires more capital than a standard fabless chip startup, setting up the larger Series C that followed.
Second-order effects
- Rivals building low-power edge inference silicon now face a competitor whose differentiation is energy efficiency rather than raw throughput, forcing the comparison onto watts-per-inference instead of conventional accelerator benchmarks.
- Each successive Mythic round raises the bar for alternative-architecture chip startups: later entrants must argue they can reach commercial scale on less capital than the $165M-plus Mythic had consumed by its Series C.
Third-order effects
- The pattern across Mythic's rounds — B, B1, C, then a $125M round six years later under an Nvidia-veteran CEO — suggests non-digital inference architectures survive only as long-duration capital commitments, with the eventual test being whether edge customers adopt them at volume or the category stays a research bet.
- If analog compute-in-memory does find its market, the structural winner is whoever controls the compiler-and-toolchain layer around the hardware, since analog designs cannot ride the software ecosystems that digital GPU incumbents already own.
The trend: Edge AI inference is becoming a capital-endurance contest in which alternative architectures like Mythic's analog compute-in-memory must sustain successive mega-rounds for years before proving commercial viability against digital incumbents.