AMD Discloses $2.9 Billion Price for Taalas Acquisition

AMD has disclosed, for the first time, how much it is paying to acquire Taalas: a cash-and-stock deal valued at approximately $2.9 billion, including up to roughly $1.4 billion in contingent consideration, according to the quarterly report AMD filed with the U.S. Securities and Exchange Commission on November 4, 2026. The figure fills in the one detail AMD left out when it first announced the deal on August 6, 2026 — the purchase price.

What AMD Disclosed

AMD's Taalas acquisition went three months without a public price tag. That changed with AMD's third-quarter 2026 Form 10-Q filing, which lists the Taalas deal under "Pending Acquisitions" and states that AMD entered into a definitive agreement in August 2026 to acquire Taalas Inc., "a developer of specialized inference silicon," in a transaction described in the filing as a cash and stock deal valued at approximately $2.9 billion, including contingent consideration of up to approximately $1.4 billion. That is the only place AMD has put a number on the deal so far; the original August announcement on AMD's newsroom and in AMD's coverage of its broader AI roadmap did not mention price at all.

The filing does not break the $2.9 billion down into a cash portion and a stock portion, and it does not spell out what has to happen for Taalas's former owners to collect the contingent piece. Subtracting the contingent ceiling from the total implies roughly $1.5 billion in guaranteed, non-contingent consideration, though AMD itself has not published that subtraction — it is simple arithmetic on the two figures AMD did disclose, not a number from the filing itself.

Timeline: From a Price-Less Announcement to a November Filing

The sequence of disclosures matters here, because it explains why search interest in "AMD Taalas acquisition price" spiked only in November, three months after the deal itself became public.

AMD's third-quarter numbers, covered separately in AMD's Q3 2026 earnings report, were released around the same filing window, which is part of why the Taalas price and AMD's quarterly results have been easy to conflate in headlines. They are two different disclosures: one is AMD's revenue and data-center results for the quarter, the other is a footnote inside that same filing describing a pending acquisition that has not yet closed.

Breaking Down the $2.9 Billion Price Tag

Two numbers anchor everything AMD has said about the cost of buying Taalas: approximately $2.9 billion as the total deal value, and up to approximately $1.4 billion of that total structured as contingent consideration. Contingent consideration in an acquisition typically means payments that depend on the acquired business, or its technology, hitting specific milestones after the deal closes — things like product integration targets, revenue thresholds, or retention of key staff through certain dates. AMD's filing does not specify Taalas's triggers, so it would be inaccurate to describe the contingent amount as guaranteed or as an "earnout" with known terms; what's confirmed is only the ceiling AMD has set aside for it.

It's also worth being precise about what "$2.9 billion" is not. It is not a per-chip price, a licensing fee, or a cloud-rental rate; it is AMD's own accounting valuation of the entire acquisition, the kind of figure that shows up in purchase-price disclosures rather than in a product price list. And because the deal has not closed, the figure could still change before it does — SEC filings disclose the terms of a pending deal as currently structured, not a guarantee that those exact terms survive to closing.

What the Filing Doesn't Say

AMD's 10-Q answers the headline question — "how much" — but leaves several related questions open:

• The filing does not disclose the cash-versus-stock split within the $2.9 billion.
• It does not describe what milestones or conditions unlock the up-to-$1.4 billion contingent consideration.
• It does not list acquisition-related transaction costs separately from the headline figure.
• It does not give an exact expected closing date beyond "by the end of 2026."

Any secondary reporting that fills in these gaps with more specific numbers is going beyond what AMD itself has disclosed in its SEC filing, and should be read with that caveat.

Who Is Taalas?

Taalas is a Toronto-based startup that designs what it calls specialized AI inference silicon: chips built around a single trained model rather than general-purpose accelerators meant to run many different models. According to AMD's own announcement, Taalas was founded in 2023, and separate reporting on the company's funding history places its founding specifically in August 2023. The company's pitch is that by finalizing a small number of metal layers in a chip's design once a model's weights are fixed, it can cut the compute and memory overhead that comes from running inference on flexible, general-purpose hardware.

AMD's own description of the acquisition, published on its newsroom, says Taalas "optimizes inference dataflows to reduce compute and memory bottlenecks that exist in general-purpose architectures," and that AMD plans to fold the technology into its AI accelerator roadmap. AMD's Vamsi Boppana, senior vice president of AMD's Artificial Intelligence Group, said in the August announcement that the acquisition strengthens AMD's portfolio with "differentiated inference performance and efficiency," framing it as giving customers more options for matching compute to specific AI workloads.

How Taalas's "Hardcore Models" Approach Works

Taalas calls its design philosophy "Hardcore Models": chips where a trained model's weights are finalized into a small number of the chip's metal layers rather than loaded at runtime from external memory. According to reporting on the company's first chip, unveiled in February 2026 and built around Meta's Llama 3.1 8B model, the approach starts from a nearly complete, reusable chip design of roughly 100 layers and customizes only the top two metal layers for a given model, with Taalas saying a fabrication partner can turn a new model into silicon in about two months. Merging memory and compute onto a single die this way is pitched as removing the need for the stacked high-bandwidth memory and advanced packaging that general-purpose AI accelerators rely on.

Taalas has published its own throughput comparisons — figures on the order of thousands of tokens per second per user, framed as roughly an order-of-magnitude faster than general-purpose alternatives, alongside claims of lower build cost and power draw. Those are vendor-supplied figures rather than independently measured benchmarks, and coverage of the first-generation chip has also noted a tradeoff: an early, non-standard low-bit data format that the company itself has acknowledged reduces output quality compared with mainstream GPU benchmarks, with a planned second generation moving to more standard numeric formats. None of those specifics come from AMD's own disclosures; they describe the technology AMD is buying, not a claim AMD has made about it since the deal was announced.

Why AMD Wants Hardwired Inference Silicon

AMD's stated plan is to integrate Taalas's technology into system-level solutions alongside AMD Instinct GPUs, and to have it complement AMD's Helios rackscale systems, EPYC CPUs, and ROCm software stack. That positions Taalas not as a replacement for AMD's existing data-center silicon — the same broad lineup on display when AMD's EPYC processors turned up inside Google Cloud's confidential computing infrastructure — but as an additional, specialized option for inference workloads where a model is fixed and extremely high throughput per user matters more than flexibility.

That tradeoff is the whole bet behind Taalas's approach: a chip hardwired to one model's weights can't be repurposed if that model gets replaced by a newer version, but it can, in principle, run that one model faster and more efficiently than a general-purpose GPU running the same weights. It's a narrower, more specialized strategy than the flexible-accelerator approach AMD has otherwise pursued with its Instinct line, and whether the tradeoff pays off at data-center scale is untested in AMD's own hands so far, since the acquisition has not closed and AMD has not shipped a Taalas-derived product.

The Founder Behind Taalas

Taalas was founded by Ljubisa Bajic, who serves as co-founder and CEO. Bajic has a long history in chip design: public profiles and Tenstorrent's own announcements describe him working as an architect at AMD and later at Nvidia before co-founding AI chip company Tenstorrent in Toronto in 2016, alongside Milos Trajkovic and Ivan Hamer. Tenstorrent's own newsroom confirmed in 2023 that Bajic was scaling back his full-time role at that company, a transition that lines up with the timing of Taalas's founding shortly after. In the AMD announcement, Bajic said: "We founded Taalas to rethink AI inference from the ground up by building the hardware around the model."

Regulatory Review and Expected Closing

AMD's 10-Q states plainly that the Taalas acquisition "is expected to close by the end of 2026, subject to regulatory approvals and customary closing conditions." That language, carried over from AMD's original August announcement, means the deal is not final. Taalas's Canadian base adds a cross-border dimension that AMD specifically addressed when it first announced the deal, saying the acquisition builds on AMD's "long-standing presence in Canada" and reflects a commitment to retaining Canadian talent — language that reads as an acknowledgment of the kind of scrutiny cross-border tech acquisitions can draw, even though AMD did not name a specific regulatory body reviewing the deal in either disclosure reviewed for this article.

What Happens Next

Three things are worth watching as this deal moves toward its expected close. First, AMD's subsequent SEC filings — including its next 10-K or 10-Q — are the most likely place for a cash-versus-stock breakdown or contingent-consideration triggers to surface, if AMD chooses to disclose them at all. Second, the actual closing date matters: AMD has only committed to "by the end of 2026," and any delay into 2027 would itself be a signal about regulatory friction. Third, once the deal closes, AMD's own roadmap updates — likely around future Advancing AI events or Instinct GPU announcements — should show whether Taalas's hardwired-silicon approach actually ships as a product alongside Instinct, or stays a research integration for longer than the announcement implied.

For now, the confirmed facts are limited to what AMD has put in writing: a definitive agreement signed in August 2026, a disclosed value of approximately $2.9 billion including up to approximately $1.4 billion in contingent consideration, and an expected close by the end of 2026. Everything else — the cash/stock split, the contingent terms, the exact regulatory path — remains undisclosed as of this writing.

Written byMara Lindqvist

Mara Lindqvist edits the hardware desk. She covers graphics cards, processors, memory and storage, the foundries and chip designers behind them, and what the numbers on a spec sheet mean for people choosing a PC. Specifications in her stories come from manufacturer spec pages and datasheets.

More from Mara LindqvistAll Hardware coverage