Qualcomm AI/Copilot PCs don’t live up to the hype

Watch what they do, not what they say

Qualcomm Snapdragon logoQualcomm’s new AI/Copilot PCs has overwhelming hype but their actions point to a far murkier picture. SemiAccurate really wanted to like this ‘new’ category but we now are convinced that the reality is nowhere near what the shills will tell you.

Words vs Deeds:

Lets start out with a simple fact of product launches, when you have a good product, get it out there to everyone as early as possible. When you have a bad one, make sure no one can independently test it and therefor contradict your rosy messaging with hard facts. Qualcomm and Microsoft have gone far above and beyond the call of duty to make sure independent reviews of their new SoC in the AI/Copilot PCs can’t happen. This isn’t by chance.

We could regale with stories of launches past where companies intentionally broke parts of press samples they knew would not compare well vs the competition but we won’t. To be fair, Qualcomm didn’t do anything like this, physically, they just made sure no one could test what they wanted on their hardware at press events and no one could get samples in time for real testing.

Some people may have been offered samples to cover for the behavior but they were said to arrive a few days before the on sale date. This is akin to breaking ports in the sense that you have plausible deniability when comprehensive testing can’t be done. In any case Qualcomm and Microsoft have a lot to hide and to be fair, they hid it quite well. We know the official excuses but, well, they simply don’t add up.

Behind the Curtain:

What is Qualcomm hiding and why? Lets start out with the why side, it is easier. The x86 emulation sucks and they know it. Since x86 compatibility is a key marketing message from both players, they want you to believe your software will just run, they have to hide this failing at all costs. Let me split a hair here, the silicon is actually quite good on the x86 emulation front, the team didn’t forget everything they did at Apple but coupled with Windows it, well, ends up sucking. Actually that is a bit unfair, lets just say the end user device sucks and not apportion blame to any one part.

When questioned, Qualcomm spokespeople will fall back to to a talking point about apps that they have tweaked/validated/worked on and those run fine, fast-ish, and without hiccups. This is true but that list is mighty short and chances you have a few apps that will blow up. Games, well, again they go back to casual games running fine which they probably do. Real games don’t and most non-tweaked games don’t either, casual or not. Qualcomm knows this and they are desperate to hide it.

Why? Again their message is all about performance and games not running or chugging badly isn’t in line with this talking point. You can be sure all the commonly benchmarked games have been made to work but the rest? Got any older games you like to play? If you have a fast laptop that doesn’t run many of your critical apps, how much standby battery life is that worth again? How much claimed video playback time is your critical enterprise app worth? Is it worth $100 to play Russian Roulette with your software?

And don’t get me started on drivers. Got any hardware that isn’t brand new? A printer bought during Covid perhaps? Scanners? Low level tools? Antivirus/antimalware? Got a game with an anti-cheat thread? VPN? This stuff WILL break on the Qualcomm ARM PCs unless they have tweaked it and chances are they haven’t. And won’t. These companies are desperate for this knowledge to not get out, and they are making sure it doesn’t by making sure only tame reviewers and fluffy youtubers get devices. Even they they won’t have time to dig in to the defects they do find.

This is all to back the official message of fast and compatible, one of which is kinda true. Compatibility is not good enough, games break on the consumer side and corporate apps are hit and miss. Worse yet there wasn’t a single mention of remote management since the November 2023 reveal. If you actually test the product you will find this out and possibly write it up before the official message has a chance to be repeated enough to ‘become truth’. Don’t believe the hype.

AI PCs Aren’t Intelligent:

Lets move on to the elephant in the room, AI PCs, we will stop using Copilot because it is getting annoying. This is Microsoft’s latest attempt at branding and is distinct from Intel’s AI PC branding and AMD’s Ryzen AI/AI PC branding. For this we will focus on Microsoft’s version which we exclusively told you about here. The TLDR version is that if your CPU has 45TOPS in it’s NPU you get the sticker. And the slush funds, lots and lots of slush funds.

How much? If you play ball like Lenovo purportedly is it means a 50% discount on Windows to start with. You will have to stand up on stage and dance for Microsoft but the cash is about as massive a kickback as SemiAccurate can recall. The money hose is on full blast this time but why? What everyone is calling AI this time around is simply not.

It isn’t powerful enough to make a difference over the TOPS available in the CPU and GPU but it will save a _LITTLE_ energy every so often. The cooked up scenarios shown off over the last few days are perfectly doable with current PCs, just with a little more battery drain. The bunny ears you put on your boss during that endless conference call may be a little more realistic with an AI PC but, well, refresh the corporate fleet NOW! AI PCs have little to no actual advantage over current silicon and that will be the case for a long time to come.

So if this stuff is pretty worthless for the short and mid-term due to lack of performance and lack of apps, why is Microsoft so desperate to push it? Simple enough, money. Microsoft is pimping AI features hard, and in some cases it can have uses, Office 365 is said to be one. Nothing a good template library can’t do better but on occasion it has a useful and occasionally correct outcome. Search on the other hand is plagued by silent errors so that is a long way off.

Why money then? All of these AI features are done where they should be done, at the datacenter. It has the actual performance to do the job right, has the data on hand, memory to run the correct large models instead of far less accurate desktop version, and more. It is the right thing to do in the datacenter for the right reason. The big problem is that Microsoft has to pay for it. It is a cost with little to no direct payback. If they push it to the user, the user pays for those cycles. Genius!

So the user ends up with a far worse version of the result but Microsoft doesn’t foot the bill, in fact they get paid when a new PC is sold. Win/win for them, lose/lose for the user. If you were really cynical you could foresee a program that says something like, “Want better AI results? For only $9.99 a month you can run your AI queries on our servers…”. <sarcasm>Never could happen.</sarcasm> So basically this whole push is nothing more than a way for Microsoft to reduce burgeoning datacenter costs. They know it isn’t a user benefit.

(Lack of) Security:

When we started this story off, we mentioned that the Qualcomm silicon was unsafe to deploy, something we have talked about before. We aren’t kidding nor do we mean mean Windows software directly, we are talking about intentional remote hardware backdoors. This isn’t just a Qualcomm problem, AMD is affected too, but Intel is definitely not. We will explain why Intel is not affected when we write up their Lunar Lake architecture, stay tuned.

Back to the fatal issue with Qualcomm security, it is of course Microsoft’s Pluton. SemiAccurate is aware of the official word from Microsoft on the topic but it intentionally leaves critical information out of the statement to make the technology seems safe. It isn’t nor can it be made safe. We don’t base this on speculation, SemiAccurate spent months talking to engineers at multiple silicon vendors many of who have directly implemented Pluton on currently shipping hardware. We are aware of what is true, what is lies, and what is omitted from Microsoft, AMD, and Qualcomm public statements. The technology is not safe, period.

Why is it not safe? Pluton is a ‘security’ hardware block that is brought up immediately after the TPM which is the first thing to come up when the system powers on. Once the TPM authenticates the Pluton block, Pluton takes over system security from the TPM. What could go wrong? Lots in fact.

Fist up Pluton is not actually a hardware security block, it is a microcontroller that emulates a TPM in software/firmware. (Note: We will use software from here on out for simplicity) Rule one of security, it should not be software at this level, period. Why? Software can be changed or corrupted. Pluton fails here. We will pause for a bit while the security folk we know read this pick themselves off the floor.

Next up is how that software is updated. The official Microsoft word is that it can ONLY be updated through Windows. This is a lie. According to multiple people who have literally implemented the Microsoft developed block in their silicon, Pluton can snoop all system busses. Let me repeat that, Pluton can snoop all system busses. Transparently. With no logs. Almost every modern system has a NIC on the SoC or connected to the PCIe bus. That can be snooped. And can send commands to Pluton, updates included. Officially though, it is Windows update. Trust who you will but the last several laptops SemiAccurate purchased were Intel based for cause.

So Pluton code can be updated without user knowledge via methods Microsoft doesn’t admit to. Who is shocked again? Luckily if malicious code is pushed to Pluton, you will know, right? Actually here is fail number… we forget the count, sorry. In any case you won’t know. You won’t know it has been updated and there is no way to check what is running on the block. The block that controls system security no less. If you poll it, it can answer but it answers via software. Anyone think that chain can be tampered with? If you can write low level malicious code, faking a result string is probably within your technical capabilities.

On the up side, once the system is up and running, the controller is far too slow to read the major busses at speed. This is a plus. That said it can poke at known memory locations to pull keys, corrupt keys, and so on. When asked, one engineer said there is no way for Pluton to send hardware keys out of the block, this is good. When the author asked, “With the current APIs?”. there was a long pause followed by, “Good question”. We don’t have a good answer to that one yet but again, if you can hack Pluton, passing bits out is probably not out of the question.

We could go on at length about key invalidation, keyspaces, hardware bricking of systems, and more but the abject insecurity of the block, and therefor Qualcomm silicon, but it is sort of tangential to this article. When we get the final piece of the puzzle we plan on writing up the full story, sorry it has taken this long.

The take home message is if you deploy silicon with Pluton, be it Qualcomm or AMD, it can be hacked or have code intentionally pushed to it that blows through any system security, hardware or software. These devices are intentionally insecurable, there is literally a remote backdoor that is easily accessed by malicious actors, not to mention governmental abuse. If you deploy this silicon you are mad, if you are not in the US and deploy it, you have even less safegaurds. And don’t ask about who has key/code escrow, spokespeople get very nervous when you do, assuming they understand the gravity of the question.

Legal Issues:

Next up is another problem, legality. The Nuvia IP that Qualcomm purchased is in the middle of a long standing legal fight between the two companies. SemiAccurate has expressed our views that Qualcomm is going to get blown out of the water in court but that is irrelevant to the next point. Lets assume there is a chance it could go either way, forget odds, just that both Qualcomm or ARM winning are possibilities.

If you read the ARM complaint here Specifically paragraph 82 on page 22, ARM asks for the destruction of the IP and other remedies. Qualcomm denies all of ARM’s claims and is slogging it out in court. So far so par for the course from large IP companies with oversized legal teams to feed. Again, at this point it could go either way.

Our question to Qualcomm was, in the however unlikely event of an ARM win, do you indemnify the customers of the silicon against whatever potential remedies are granted? Qualcomm answered with a quote from their 10-K, “Arm Ltd. v. QUALCOMM Incorporated: On August 31, 2022, Arm Ltd. (ARM) filed a complaint against us in the United States District Court for the District of Delaware. Our subsidiaries Qualcomm Technologies, Inc. and NuVia, Inc. (Nuvia) are also named in the complaint. The complaint alleges that following our acquisition of Nuvia, we and Nuvia breached Nuvia’s Architecture License Agreement with ARM (the Nuvia ALA) by failing to comply with the termination obligations under the Nuvia ALA. The complaint seeks specific performance, including that we cease all use of and destroy any technology that was developed under the Nuvia ALA, including processor core technology. ARM also contends that we violated the Lanham Act through trademark infringement and false designation of origin through unauthorized use of ARM’s QUALCOMM Incorporated NOTES TO CONSOLIDATED FINANCIAL STATEMENTS F-23 trademarks and seeks associated injunctive and declaratory relief. ARM further seeks exemplary or punitive damages, costs, expenses and reasonable attorney’s fees, and equitable relief addressing any infringement occurring after entry of judgment. On September 30, 2022, we filed our Answer and Counterclaim in response to ARM’s complaint denying ARM’s claims. Our counterclaim seeks a declaratory judgment that we did not breach the Nuvia ALA or the Technology License Agreement between Nuvia and ARM and that, following the acquisition of Nuvia, our architected cores (including all further developments, iterations or instantiations of the technology we acquired from Nuvia), server System-on-Chip (SoC) and compute SoC are fully licensed under our existing Architecture License Agreement and Technology License Agreement with ARM (the ARM-Qualcomm Agreements). We further seek an order enjoining ARM from making any claim that our products are not licensed under the ARM-Qualcomm Agreements, are not ARM-compliant or that we are prohibited from using ARM’s marks in the marketing of any such products. On October 26, 2022, we filed an Amended Counterclaim seeking additional declaratory relief that certain statements ARM is making in the marketplace concerning our rights under the ARMQualcomm Agreements are false, and that ARM has no right to prevent us from shipping our products, which are validly licensed. Trial is scheduled to begin on September 23, 2024. We intend to continue to vigorously defend ourselves in this matter.

The above, while relevant to the case, says nothing about the question asked. SemiAccurate asked for clarification but as of this writing has not gotten anything. Without that, would you deploy this silicon? There may be some NDA deals in place to make things right so do think before you buy, and let SemiAccurate know if you get anything concrete. Again, while we have our views, the case could go either way at this point but risk is risk.

Lack of Disclosure:

Qualcomm may tout performance which as we keep saying, isn’t bad. At Computex, Intel released some numbers showing that their upcoming, don’t forget the X Elite isn’t out yet either, CPUs beat Qualcomm in some critical benchmarks. (Note: This was written a few weeks before today’s on sale date) AMD has their Zen5 based offering being released in the very near future too. This isn’t to say that Intel or AMD will best Qualcomm, just don’t leap to conclusions until you see the numbers from both sides, from independent reviewers, not paid shills. That said, performance of Qualcomm PCs is an issue. Why? No, not this but it doesn’t help.

The real problem is that Qualcomm would not allow independent testing on their silicon. They promised that it would happen late last year. They promised us systems in time to test before release. They promised a lot. None of it happened. No systems. No independent testing allowed, only a very narrow slice of canned benchmarks with inadequate disclosure surrounding it as we went on about above.

Author’s Note: We were offered the chance to come down to Qualcomm after this and have benchmarks run and we turned it down for reasons related to protecting our sources. Qualcomm could have seeded devices, Intel, AMD, and on occasion ARM all do, but that would have contradicted their messaging in the worst way.

Back to disclosure, late last year, Qualcomm did right and made large steps on the path to correct disclosure. Then over the last few months they backpedaled badly. What do we mean? Take a look at the disclosures from Intel’s rather trivial disclosures at Computex and the last Qualcomm set we have handy.

Intel Disclosure Example

How disclosure should be done

Qualcomm Disclosure Example

How disclosure should not be done

The Intel disclosure page was one of three in a 16 page deck, 5 pages of which were headers or non-content. Qualcomm’s is the best from the deck, several claims were backed up by nothing. One side allows you to recreate the tests, the other leads you to wonder what they are hiding. Before you think we are picking on Qualcomm a lot, we pick on companies for cause, not to single anyone out. Remember this? AMD cleaned up their act and Intel has done wrong in the past too but not recently. As a general statement, companies often stretch credibility and the best way to hide it is to not disclose details. This is why we are so hot on the topic, and will continue to hold their feet to the fire.

So what is the end result on performance? We think the silicon is just fine and has been for quite a while. It is being hobbled by an unready OS and no software that offers a real benefit to the end user. Both Qualcomm and Microsoft know how unready and simply bad their offerings are. Their words shout out greatness but their actions show fear. Luckily they both have legions of paid, directly and indirectly, analysts, YouTube ignorati, and fluffy influencers to ‘make truth’ for them. Honest reviewers don’t get sampled or rarely get devices with nowhere near the time needed to do the job right.

Confidence, Thy Name Ain’t Qualcomm:

We mentioned the money angle of AI PCs earlier but only from the OEM side. Most people don’t understand how much these slush funds can sway the market. The more confident a company is in their products, the less they have to bribe OEMs and customers to buy them. If a car is a hot seller, you won’t see large rebates or low financing offered. On the PC side this funding is hidden in a way that is rarely obvious. It takes a desperate company to be blatant on this front, and Qualcomm is confident in the performance and compatibility of their AI PCs, right?

Best Buy Front Page with Qualcomm AIPC free TV offer

Technically not bribery

Now we get to a humorous aside, confidence. As we beat into the ground above, when you are confident in your product, you do things one way. When you are not, well, there are lots of hoops a marketing or PR organization can jump through. As you can see above on the front page of Best Buy, if you buy a Qualcomm based AI PC, you get a free 50″ TV. Those PCs start at $999 and run up to $1499 MSRP, $1349 for the cheapest Samsung model. More on that in a sec.

That 50″ TV is either a Insignia, the Best Buy house brand, F30 or if you buy a Samsung laptop, a Samsung DU7200. These devices retail for $299 and $379 respectively but seem to have a price drop between the time this was written and publication to $219 and $349 respectively. This is what we mean by slush funds and kickbacks, and they are pretty massive. Pre-price drop, the non-Samsung kickback was 30% of _RETAIL_ pricing. That is WAY more than the CPU that Qualcomm is selling plus the OEM OS cost for volume Microsoft customers.

So why is this happening? SemiAccurate thinks it is desperation on Qualcomm and Microsoft’s part, you don’t give such astronomical kickbacks if you have any faith in your product, period. Everyone involved wants massive day one sales and no facts that contradict their messaging until much later. They seem to have succeeded on both counts. Then again, so did Apple with their Vision Pro and you know how that turned out, right?

As a humorous aside, a finance professional SemiAccurate talked to a few months ago asked what we were looking for at the launch of the AI/Copilot PCs. I said, “returns”, which prompted him to ask what I meant, the amount of money the various parties get from the launch or something similar? I then replied that I meant it literally, IE the stacks of boxes in the back of Best Buy once people find out what they actually bought. 29 days until we know if I am right. These kickbacks put the Intel Ultrabook money hose to shame and you remember how badly that distorted the market, right?

Last and Probably Least:

The last bit is probably the most important even if most people won’t care. During Computex SemiAccurate asked a Qualcomm engineer who focuses on the BIOS/firmware for the laptop about the bootloader, specifically if it is locked, and if so was it at Microsoft’s request. He confirmed it was locked and it was Microsoft driving the issue. Fair enough, we expected no less but the honesty was refreshing.

Why is this important? Because between that and the Pluton backdoor, you are not buying a laptop, you are buying a closed Microsoft device that they can do things to without your knowledge, consent, or ability to turn down. It is a phone with a big screen and poor compatibility with the existing software base, and worse yet you don’t really control it, just own it. Sort of. Go read the EULA and be afraid.

The End or The Beginning?:

So what do we have? We have a device where the rhetoric is diametrically opposed to the actual product. The CPU silicon is good but it is dragged down to unacceptable by everything above it. Those that control such things went well above and beyond the call of duty to make sure there was no independent analysis before the messaging was fixed in the mind of the public. This wasn’t an accident.

Toss in mediocre independent reviews, eye-wateringly large subsidies, lack of disclosure, insecurable backdoored hardware, and a locked bootloader and you have a perfect storm. There is no up side here unless you are tallying Microsoft’s server costs, and even then it is tenuous. This launch is far too early to hype up like they did, the software is half baked and will poison the waters for a long time to come.S|A

The following two tabs change content below.

Charlie Demerjian

Roving engine of chaos and snide remarks at SemiAccurate
Charlie Demerjian is the founder of Stone Arch Networking Services and SemiAccurate.com. SemiAccurate.com is a technology news site; addressing hardware design, software selection, customization, securing and maintenance, with over one million views per month. He is a technologist and analyst specializing in semiconductors, system and network architecture. As head writer of SemiAccurate.com, he regularly advises writers, analysts, and industry executives on technical matters and long lead industry trends. Charlie is also available through Guidepoint and Mosaic. FullyAccurate